310 52 33MB
English Pages 318 [299] Year 2005
GEORGE GILDER
HHmmI
USA
ISBN 0-393-05763-1
$22.95
CAN. $32.00
Thanks tion,
to the digital
technology revolu-
cameras are everywhere— PDAs,
phones, anywhere you can imaging chip and a lens. Battling two-billion-dollar market
is
to
put
an
usurp
this
Silicon Valley
a
company, Foveon, whose technology not only produces
become
a
superior image
the
eye
artificially
in
may
also
but
intelligent
machines. Behind Foveon are two legendary figures
who made
possible: Carver
the
Mead
personal computer
of Caltech,
one
of the
founding fathers of information technology;
and Federico Faggin, inventor
of the
CPU— the
chip that runs every computer.
George Gilder has covered the wizards
of
high tech for twenty-five years and has an insider's
knowledge
of Silicon Valley
and the
unpredictable mix of genius, drive, and luck that can turn a start-up into a Fortune 500
pany. The Silicon Eye of
some
of the
is
a rollicking narrative
smartest— and most colorful-
people on earth and their race entire industry.
com-
to
transform an
PUBLISHED TITLES
THE
IN
ha
NTE RPR
I
S E SERIES
George Gilder
The
Silicon
Tim
Eye
Parks
Medici Money: Banking, Metaphysics, and Art in Fifteenth-Century Florence
Ken Auletta Media Man: Ted
Turner's Improbable
Empire
Rich Cohen
Machers and Rockers: Chess Records and the Business of Rock
&
Roll
FORTHCOMING TITLES Richard Rayner California Capitalists
Stanley Bing
Rome: The Rise and
Fall of the First
Multinational Corporation
James Buchan
Adam
Smith
Jagdish Bhagwati Terrified by Trade
GEORGE GILDER
BY
The Party That Lost
Its
Head
Men and
(with Bruce
Chapman)
Marriage
Visible
Man
Wealth and Poverty Recapturing the Spirit of Enterprise
Microcosm Life After Televsion
Telecosm
The
Silicon
Eye
The Silicon
Eye
George Gilder
& S
Atlas Books
W. W. Norton
8c
Company
New York London •
© 2005
Copyright
by George Gilder
All rights reserved
Printed in the United States of America First Edition
For information about permission to reproduce selections from this book, write to Permissions. \V. \V. Norton
&
Company.
Inc..
500
Fifth
Avenue.
Manufacturing by The Courier Companies.
New York. NY
101 10
Inc.
Book design by Chris Welch Production manager:
Amanda Morrison
Library of Congress Cataloging-in-Publication Data
Gilder.
The
George
p.
cm.
—
date.
F.,
George
silicon eye
Gilder,
(Enterpr>
Includes index.
ISBN 0-393-05763-1 1. i
Foveon (Firm)
Santa Clara County
i
3. Digital
electronics.
Electronic digital computers.
5.
8.
(hardcover)
High technology industries
2.
Visual perception.
9.
4.
I.
California
Photography
6. Artificial intelligence.
Cellular telephones.
Federico. 1941-
—
Title.
II.
Valley
Digital
7.
Computer
Mead. Carver.
10.
Enterprise
— Santa Clara — techniques. vision.
11. Faggin,
(New York, NY.
HC107.C2H5334 2005 _ 338. "62 367'09-94 3—dc22 1
2004030994 \\. \Y.
Norton
& Company.
Inc..
500
Fifth Avenue.
New
York.
N.Y 10110
www.wwiMHton.com \\. \\.
Norton
& Company
Ltd.. Castle
House. 75/76 Wells
Street.
London
WIT 3QT
"-90
.
For Louisa
Contents
Prologue
Who
Is
Carver Mead?
13
THREE MEN AND A CAMERA Delbrtick Bursts In
23
The Fourth Camera Merrill's
View
31
47
The Charged Competition Annealing Neurons
57
63
THE FORGOTTEN FOUNDER FF
69
Beyond the Tobermory
Silicon
Gate
87
The Hopfield Net
93
jj
Part
THE CAT AND THE CAMERA
Three 1
U
Carverland
101
in
Michelle 1
2
Misha
1
D
Synaptics
139
4
The Analog
Perplex
Seizing the
Sword
16
125
The Cat
147 159
175
Teleology at Oxford
White Rabbit
Part '
ZU
Four
Merrill's
185
193
FOVEON
Magic
201
Lyons Den of Many Colors
209
Photokina 2000 and Beyond
ZZ
Foveon's Disruptive Challenge
Holy Kammoli! Foveon
RAW in Vegas
Epilogue
Camera's End
Glossary
287
Acknowledgments Index
303
251
299
277
261
223
233
The Silicon
Eye
Carver
Mead and
Federico Faggin in 1988 in front of a giant model
of a microprocessor resembling Faggin's historic designs. At the time,
Faggin was
CEO
and Mead was chairman of Synaptics, a company
devoted to creating a
new
kind of microprocessor inspired by the mas-
sively parallel operations of the
human
brain. Courtesy of Foveon.
Prologue
Who
//
—"^
"|
D
Carver Mead?
Is
eath Valley Nuts and Sweets/' beckons the neon sign.
Amid
mountains,
the lunar desert in a little
rimmed with remote
shop claiming
to
be "the most
beautiful gas station in the world,'' a sixty-nine year-old
named Carver Mead assumes century pilgrim.
among
the gas
The glow from It is
ently
full
the sign
falls
of the electrical debris. is
this
hunched on
mind
woman
—an
array of appar-
might intrigue a teenaged
of an airport security guard.
sits
October 2003, haplessly trying
this before.
relics?
And
midst of Death Valley in early
to catch
13
A
patiently in the car in front
She has been through
at a dive in the
phone.
his rental car.
Carver Mead, collector of electrical
he doing
prayerfully
his cell
on the backseat of
electrical devices that
small middle-aged blond
is
is
trying to get reception
tinker or strike terror in the
what
he
of arcane century-old technology
random
Who
the iconic pose of the twenty-first-
A trim bearded figure,
pumps
man
up with the world?
i
George Gilder
4
following his own visions and disciplines, whims and raptorial will, Mead has always managed to remain ahead. Although known
little
outside the silicon valleys of America, he has been
the leading intellectual figure in four decades of American electronics,
through
three
generations
fastest-advancing
of the
gauntlet of world technology. Crucial to his lead has been his intuitive sense of phase, his special feel for the timing of the
crests
and
and troughs of opportunity
history.
His intuition
is
in life
nicely
and
electricity,
summed up
exhortation, ''Listen to the technology; find out
in his
what
science
famous telling
it is
you" (meaning to go with the flow of the basic physics rather than twist
it
to
own
your
desires).
Born the son of an engineer tric
in a
Southern Pacific hydroelec-
plant in Kern Valley in California's
come
to learn of the
tentous relationships
power was
in phase,
immense power
among
its
High
Sierras,
Mead had
of electricity and the por-
phases in the plant.
When
the
with the crests and troughs aligned between
the generator and the line, electricity the eighty miles to Los Angeles.
would rush
But
between the generator and the transport
if
reliably
down
the waves of force
line
were out of phase,
the voltage shift could blow up the plant.
The Gordon and
Betty
Moore
professor of science and engi-
neering at the California Institute of Technology (Caltech), in
Pasadena,
Mead
early in his career
the millions of watts of
power
had moved
his
in his father's
concerns from
domain
to the
nanowatts (billionths of a watt) of power in exotic microelectronic devices. In
1963 he
built the world's fastest transistor.
He
The
Eye
Silicon
15
researched the proposition about the rate of advance of semi-
conductor technology, doubling every eighteen months, that he
named Moore's Law
Gordon Moore. He con-
after its author,
ceived the chip design technologies that soon dominated the industry.
He
proposed and taught a historic course on computa-
tional physics with the late
Feynman, with
whom
Nobel Laureate physicist Richard
he shared a belief that electromagnetic
waves can move both ways
He
in time.
wrote an important book
on quantum theory and superconductivity called Collective Electrodynamics. Crucial in ships.
Whether
all
these pursuits were phase relation-
they resonated and grew, or dissipated
mined whether useful
The waves
—whether and dispersed —
the waves of force were in alignment
effects
deter-
were achieved.
of influence from his classes at Caltech often
impelled
Mead
advising
Moore
issued, he spent
north to Silicon Valley. at Intel,
much
From
his early years
where Mead wore the seventh badge
of his career
commuting
to the
Bay Area
and guide
their endeavors. His quiet
voice, steady gray eyes, trim goatee,
and swirly colored open-
to consult with his students
necked
shirts
When
became
familiar in the high councils of the indus-
As
to await
an
alignment of the phases of his thought with yours, he pauses.
He
try.
he says something, you
listen.
if
thinks before he speaks, and as he thinks he radiates, and the radiation
reaches the listener and opens his mind for the
impending
idea.
During the course of
his career,
tened attentively enough to act on his concepts.
many have
He
lis-
has served
as a founder of twenty companies.
Prominent among the twenty
is
a photographic
imager com-
George Gilder
16
pany
off
Mead lens
Zanker Road
in
Santa Clara
named Foveon, where
serves as chairman of the board. Imagers take light from a
and transform
it
into electrical signals containing an image.
new
Until early in the
century, nearly
all
cameras were analog,
using film that directly transcribed the image into a visible pic-
But
torial pattern.
this is a digital age. In digital
direct chemical analogies of a scene;
numerical values of the
light colors
you don't make
you translate
and
intensities,
it
into a set of
which can be
processed by computer into the desired picture that you choose to print out.
In
September 2003, one month before Meads stop
Death
imagers that converted the picture into a pattern
Valley, digital
of bits
in
and bytes
to
be stored on a computer memory exceeded
the unit sales of the usual analog film imagers for the
first
time.
Sold in 2003 were close to 35 million digital cameras. Incorporated in cell phones were a rapidly growing additional 10 million digital imagers.
Seven years before,
in 1996,
new
Mead founded Foveon
He
to
an imager for
this
imager chip
superior to any other imager in the world.
many
is
experts
in
era.
launch
devoutly believes that Foveon's
agreement are analysts
at
several
Among leading
photography magazines in Japan and the United States, John
Markoff of the azine,
New York Times,
futurist
and the photography team
Yet here
it
was seven years
the market for
new
at
John Dvorak of PC mag-
Consumer
after the
Reports.
company's founding, with
imagers exploding, and something at Foveon
was way out of phase. Of the nearly 50 million microchip imagers sold in 2003, the Foveon device
commanded
a share of
The
Silicon
the market less than one-tenth of
pany was an interim
scientist,
1
17
percent.
Running the com-
CEO who planned to return to his previous
role as chief financial officer.
nent chief
Eye
Richard Lyon, the company's emi-
was often on the phone, embroiled
in solv-
ing technical problems for purchasers of the Sigma camera that
used Foveons imager. Morale was declining. The most prominent board member, Federico Faggin, the Silicon Valley legend
who
built the first microprocessors at Intel thirty years before,
had chosen
to take a vacation
inventor of the imager, with
with his wife in China. The key
Mead, was an engineer named
Richard Merrill, recently returned from a photography tour in Africa.
The key designer
of the crucial software systems
and
imaging algorithms for the device was chief scientist Lyon.
was regaling the industry with learned speeches on the
He
history of
photography.
Conducted by Lonergan Richards
of
Redwood
City, a leading
new
Silicon Valley executive search firm, the quest for a
was foundering. With the venture investors
capitalists
and company
on the Foveon board bickering over future
one knew how
to
choose among Lonergan's
which included major
figures
CEO
strategy,
no
set of candidates,
from leading semiconductor com-
panies such as Intel and Micron.
Mead, however, to leave the
is
company
often a contrarian, and he had his reasons to its
own
devices for a spell. In the midst
of this critical transition, he chose to turn off his cell set out
with his
"life
partner," Barbara Smith,
on a car
phone and trip
across
the country. Said to be hidden deep in the Appalachians in
North Carolina were intriguing power-line
artifacts that
could
George Gilder
[8
aid
Mead's new
power
efforts to write a history of the creation of the
had served
grid that his father
For ten days,
Mead and
for a lifetime.
Barbara had found no relevant
But on the eleventh, deep
in the
mountains where no one had
meddled with the equipment, they began and
nal original amplifiers
the power grid of 1909
work
tems worked. Everybody had is
an incredible
uncover "phenome-
was
No
one knew how these
sys-
wrong,"
it
home
up
Mead
New Mexico,
weeks,
Mead
and across
to
a rental car with the arti-
in California
Death
decided to check his
down
phone
His Verizon service only stored three of the messages that a
woman named
for
know whether
was
it
messages.
But one
a voice mail
to
Mead
a thing or a person."
be Phil Bond, the undersecretary of
"Has anyone talked
more than two
from
Porter.
"She alluded to a Fillbond or something," didn't
40
into Texas
calls at a time.
somehow got through was
Mildred
Interstate
down
Valley. After
cell
"To find this
said.
through the Blue Ridge Mountains in Tennessee,
and
made
stuff
North Carolina. "This
in
thrill." Filling
they set off for their
facts,
to
insulators," the key devices that
not covered in any of the histories.
stuff
relics.
Commerce
It
recalls. "I
turned out to
for Technology.
you?" she asked as the
cell
phone
crackled. "I
can't hear you," said
She wanted
to
Mead, hunched by the gas pumps.
know: "Can you do a film
for the presentation?"
"What presentation?" Mead asked. "You don't know!" said Porter. Within a few minutes,
managed
to learn that
Technology
to
White House.
he was
a recipient of a National
be awarded by President George
Mead
Medal
W Bush
of
at the
The
In the East
Silicon
Eve
Room ceremony a month
19
later in early
November,
Bush would yawn through many of the presentations. But he was concerned about the future of U.S. manufacturing and
When
employment.
he heard of Mead's twenty companies, he
exclaimed, "Wow. That's what
we
need.''
Sure. But the president did not
there in the
White House
Foveon was indeed
project.
that
a
failure in his
company
vital to
and manufacturing employment.
petitiveness
retrieve leadership for the
Mead, standing
after a fall of travels
was on the verge of a
researches,
know
United States
in
It
and
historical
most promising
American comwas targeted
cameras and
to
digital
cell-phone imagers, long lost to such Japanese firms as Sony, Fuji,
and Canon. But
tions, in the
energy and
this
company
creativity,
most promising of American innova-
into
was
which Mead had poured the most
still
drastically out of
phase with the
market.
With
its
roots in a research project twenty years before at Cal-
tech in a field called neural networks, developing into a com-
pany called Synaptics through the and emerging
at
Foveon
imager had once seemed
late
1980s and early 1990s,
at the turn of the century,
likely to
Mead's new
be a capstone of his career. Yet
U.S. technology had suffered through a devastating crash in the year 2000. Attempting to bring the Foveon imager to market, the
company had found
the possible U.S. customers, from
Kodak
to
Motorola, in disarray, and the Japanese camera titans such as
Canon and Nikon for their
unwilling to rely on a small American start-up
key imager technology.
cesses, failure at
Amid
Foveon seemed a
century dawned over Death Valley.
all
the honors and suc-
distinct possibility as the
new
Part
One
THREE MEN AND A CAMERA
Nobel Laureate
on the beach
Mead
Max Delbriick playing backgammon with
at
his son Tobi
Baja in 1977. Physicist Delbriick pushed Carver
into biology
and Tobi became Mead's student and a founder of
Foveon. Tobi Delbriick
1 Delbriick Bursts In
Carver
Meads
path to the Foveon camera began one
spring day in 1967
Mead
when Max
the door between physics and biology. Storming
into his Caltech office with a
down
Delbriick burst open for
a challenge
on
his
bang and
a flurry, Delbriick hurled
desk that lasted the young professor
for
a lifetime.
A
burly
briick
man
had gained
with heavy hornrims and robust humor, Delhis doctorate in physics in the early
1930s with
quantum theory pioneer Max Born and then had joined arch Niels Bohr in Copenhagen. There he had
with the responses of living things to phototaxis. Defecting to biology, he icine in inal
1969
—two years
work conducted
All his
view,
and
days as a
life,
at
after his
become
light, a field
would win
meeting with
a
patri-
intrigued
then called
Nobel
Mead
in
—
med-
for orig-
Caltech.
Delbriick was ready to question the established
his questions
quantum
were seldom solemn. In
theorist
his
he authored an irreverent
*3
Denmark satire of
George Gilder
24
quantum theory formed
—"The
Copenhagen Faust"
—
that
for the physicists at Bohr's Institute in
was
per-
1932 with the
twcnty-six-year-old Delbriick himself serving as a top-hatted
master of ceremonies. At the other end of his career, studying the brain in his book
Mind from Matter?
the sixty-nine-year-old
Delbriick jibed at the neuroscience of his biologist colleagues.
Their efforts to explain the mind as mere material brain, so he wrote, resembled nothing so struggle to extract himself
own
much
as
"Baron Munchausen's
from a swamp by pulling on
his
hair."
Now
in
1967 Delbriick wanted Mead's opinion on
a recent
outpouring of scientific papers comparing the electronics of ion transfer through nerve channels in the brain with the electrical
behavior of transistors. Thirty-one years old, the lean, sprucely
bearded at
on
Mead was
already a professor of electrical engineering
Caltech with a growing reputation as a world-beating expert transistor physics.
'These guys are saying that a nerve membrane works transistor. Is this right?"
Delbriick asked
ferring electrical signals through
Mead
membranes
like a
brusquely Trans-
in the
human
nerv-
ous system, the nerve channel in biology corresponds on a superficial level to the conductive
channel
in a transistor, a tiny
switch or amplifier of electricity invented at Bell Labs in 1947
and used by the millions today Currents in both seemed to researchers "I
assumed
rise
in every
advanced microchip.
exponentially with voltage, and
that the processes
were the same.
don't know, I'd have to look at the papers,"
curtly.
At the time,
Mead
did not
know who
Mead
replied
the imperious pro-
The
Silicon
Eye
25
was and envisaged papers from
fessor
biologists with a faint
distaste. "All right/' said Delbriick, "here they are."
He
tossed a formi-
dable pile of printouts in front of Mead. "I'll
look at them,"
Mead
said doubtfully.
During the next week, apprised of the identity of the intruder
and
rising to the gauntlet hurled
papers.
Quickly warming
to
on
the
his desk,
Mead
subject,
he found himself
perused the
increasingly dubious of the confident calculations and comparisons
between electronic and
biological functions.
He
called
Delbriick and commented, "Those papers were interesting, but the model they are using
is
really garbage."
"In that case," Delbriick responded, "we'll have to find out
what's true."
Thus began lowing a at
summer
Konstanz
1971,
a long relationship
Mead
in
between the two men.
studying electrical currents in nerve channels
Germany with
Delbriick ally Peter Lauger in
designed a special electronic device for plotting the
conductance of nerve channels using "black
These recently invented
artificial
the physiology of the nerve
lipid
membrane. Using
membranes conductive, Mead showed
membrane
special catalytic to
make
that the flow of current
did not increase exponentially with voltage,
but that the number of connections across the
The
bilayers."
structures exactly reproduced
molecules acquired by Delbriick that were known
across a
Fol-
brain turned voltage not directly into
membrane
power but
did.
into con-
This finding became the prevailing wisdom in the
field
and deepened Mead's relationship with Delbriick, which
also
nectivity.
George Gilder
26
led to a tutorial link with Delbruck's son Tobi at Caltech. After
Delbruck's death,
a kind of uncle to Tobi
and Tobi
of the founders of Foveon Corporation.
became one
The
Mead became
elder Delbriick focused
Mead on
Delbriick called transducer physiology:
all
the subject of what
that goes
on
ceptual process between the physical inputs and the icant output,
all
in a per-
first signif-
the transformations that happen between sense
and symbol, between the blinding rush of electromagnetism on the eye and an intelligible image in the mind. Delbriick that
what we perceive
(the
image of the world)
is
showed
radically dif-
ferent from
what we receive
Vision
more than photodetection. As MIT's Anya Hurlburt
is
far
and Tomaso Poggio put
it
(the flood of
waves on the
years ago, "Vision
is
retina).
not a sense but an
intelligence."
Biologists
knew
little
about these transducer processes
beyond rough vectors of connectivity between the retina and the cortex that to vision.
seemed
Mead
to assign nearly half of the brain's resources
believed that he had found ways to model
perceptual functions. tions
were some
He assumed
fantastic,
that in general brain func-
mostly analog, concoction of currents,
chemistry, and timing, with baffling labyrinthine links
them, "high fan-in and fan-out"
Although past
silicon
some
—
lots of
among
connections to and
fro.
technology could not yield analog
devices with densities remotely comparable to brain tissue, by the early 1980s
work
Mead was moving
that rendered "toy" retinas
kind of neural net-
to create a
and cochleas out of very
scale analog-integrated circuits, or analog
large-
VLSI. These were
chips that used thousands of transistors not as digital
number
The
Silicon
Eye
27
crunchers but as sensors and collectors of real-world inputs.
Mead and
his
team created electronic cochleas
that
would prove
useful in hearing aids and also contrived a series of retina chips.
Although
still
vastly simpler than the brain, these
offered the glint of
new
Once you
insight into perception.
build something, you understand
in a
it
way
models
you never
that
can through speculation or mere measurement. And once you understand something, grasp tions,
you can build
it
its
underlying physical founda-
ever better. This
is
the virtuous cycle
of industrial creation that began with Delbriick's transducer
challenge.
The process digital is
ultimately led to the creation of
imager for a
digital
camera. The
rise of
what
is
called a
the digital camera
the most sudden and transformative industrial event since the
emergence of the personal computer. After the provision of series of
new
digital storage devices,
(CD) and the
video disc (DVD), and after the prolifera-
camera opens up the world of electronics
eye.
of
such as the compact disc
around the globe of the wired and wireless Internet, the
tion ital
digital
Anywhere
human
dant
that the
vision
digital
and
network can reach
vigilance, art
product of the
new
and
is
to the
now
insight.
than the idea, even the phrase,
no such
thing.
The
a possible site
It is
the ascen-
millennial decade.
digital
digital
the camera addresses what
is
camera. There
camera
is
is,
more
after
all,
an oxymoron because
chiefly an analog problem: the
challenge of creating within a machine a model colors, shapes,
dig-
power of the
Yet nothing reveals the hypertrophy of the digital idea
really
a
—an analog—of
and textures from the world outside
it.
George Gilder
What
a so-called digital
images from
camera does
to convert analog
is
light detectors into three arrays of tiny dots called
pixels, or picture
elements. In most digital cameras, each pixel
bears a numerical value for the intensity of one of the three pri-
mary at
colors
each
—
red, green,
pixel, the
and blue.
each
pixel,
it
but one color
Gauging only one of three colors
computes the actual mix of
of the value of colors of neighboring pixels. it,
all
conventional digital imager throws away two-
thirds of the light at the outset. at
Filtering out
colors
on the basis
As Carver Mead puts
the digital camera "guesstimates" the colors.
Carver Mead's obvious, but profound, insight that eventually led to
Foveon Corporation was
and use the
to
banish the
—the substance of
silicon itself
kind of permanent film, like a retina.
determined scientist,
From
to avoid
From
digital
guesswork
a microchip
the outset,
—
as a
Mead was
throwing away information. As Foveon chief
Dick Lyon, explains, they wanted "no guessing
at all."
the beginning, they resolved to "measure every color at
every pixel." In the Foveon camera, every pixel would register real features of the
image rather than
digital simulations of
At Caltech, long before Foveon was a gleam
and
his team, inspired
by Delbriick, learned
goal of silicon "film" not by trying to build a
in his eye,
how
Embracing the
computer
that
silicon as the physical layer,
undertook a deep twenty-year exploration of ties.
its
Mead
to achieve this
be better than an eye, but by building an eye, actually a in silicon.
it.
would retina,
Mead
analog capabili-
The Foveon camera began not by attempting
to
transcend
analog processes or render them irrelevant, but by exploring
both the biology of the
computers.
human
eye and the limitations of digital
The
Silicon
Eye
29
Computers can perform instantaneous calculus and
create
models of short-term weather or store and search the entire contents of the Library of Congress in a disk-drive database. But
they cannot see. Even today, recognizing a face glimpsed in a
crowd across an
image processing than together.
The
and
all
the supercomputers in the world put
pursuit of a fully successful camera, therefore,
must begin with brain,
two human eyes can do more
airport lobby,
human
a study of the
this study, for all the
retina
and the human
grander claims of brain science,
has only just begun.
Now being used
in
cameras
for the first time, the first
based on a serious study of the is
the Foveon X3.
The
story of
human its
retina
imager
and neural system
creation combines themes of
neural science and engineering with a tempestuous narrative of
business history.
many
From Delbriick and Feynman
to
Mead and
his
students and colleagues, led by Richard Lyon and Richard
Merrill, the story brings together
some
of the
paramount
both in the history of Silicon Valley and in the research nity at Caltech.
It is
a story of
minds and
brains,
figures
commu-
and the
still
mysterious transductive interplay between them, in conflict and creation, in
triumph and self-destruction.
Foveon chief
scientist
Dick Lyon with
his
high-school yearbook displaying the panoramic
product of his third camera, huilt sixteen.
His fourth camera
industry. Dick Lyon
may
when he was
transform the
The Fourth Camera
As
a high-school youth,
Dick Lyon showed scant
an incipient revolutionary or dasher of icons.
signs of
He was
a
nice kid with crisp, smooth features, dancing eyes, and short
brown
hair
brushed neatly
to the side.
You could see the
eyes dance in a photograph he had concocted for the yearbook
where
his face reflected clearly in the bulge of a shiny faucet in
a basin in the physics laboratory.
He was head
photographer of
the yearbook for two years, a good job for a dutiful teenager
who
respected his elders and the equipment. In other pictures in his yearbook, nearly
you could see the eyes of some of the
The cute blond and smiling
pictures, definitely
no
black and white,
dancing around him.
Leslie, for example, holding the
in the
Math Club and
girls
all
huge
Math Club photo and decorous
in
slide rule
many
other
was intrigued by Lyon. President of both the
the Science Club, teachers pet and polisher, with
signs of a fierce Edisonian scowl or
Feynman
mischief, Lyon
looked twinkly, a bright, well-behaved teenager of the 1950s.
V
George Gilder
32
Except
it
was not the 1950s, but
a proxy: the late sixties
Roundup,
earlv seventies in El Paso, Texas. In his yearbook, the
1970 high-school
for his
class, signed
many
by
girls call
"Einstein" and in a neat, leaning-over-backwards pen, pay ble tribute to his brains
better
and bright
future.
But
and were not especially impressed.
was not up
gently imply, he
to the
and
his parents
him
humknew
Intellectually, they
standard set by several of his
brothers.
"You should ask
me
Da\id," says the dad.
about
Why
Tom and Jim and Bob
was
I
.
.
.
yeah and
focusing on Dick anyway?
"Dick couldn't even spell 'diaphragm' correctly for his Westinghouse Science Prize application," recalls the
'Tm not
saying he was slow or anything, and he usually could
But he
spell.
messed up on 'diaphragm.' And
really
the expert on cameras.
Ellen,
is
ration
and made
there wasn't
all
The Lyons basement on
Most
in the house.
much room
that
lived in a
his sister,
Eastman corpo-
to
But no question,
That wasn't
at issue.
to start with.
modest structure with
and no
a garage
L.
Lyon,
Jr.,
and
his writer-editor wife
their nine children kept the
of the kids
would
star
Willie Ingles as
Dad
academically
at
Austin High
did before them, and go on to gain
science degrees at major universities. to
work
at
Verna
scene mostly under con-
School in El Paso, study algebra and precalculus under
went
And
a quiet El Paso street, called Fillmore Avenue,
where William
trol.
Went down
a collection of early Kodaks."
Dick took the most room
Mae. and
mom.
A
civil
Airs.
math
or
engineer, Ellen
Eastman Chemical Corporation and became
intrigued with cameras. David, Jim,
Tom, and
Bill
would go
to
The
The
Princeton.
Eye
Silicon
33
oldest, David, a mathematician,
now
does busi-
New
ness consulting at Aurora Market Modeling in Manchester,
Hampshire.
Bill
Lyon
III
applies advanced
jections at the Prediction ied in
mathematics
to business pro-
in Santa Fe.
Jim also stud-
Princeton and went on to work at Microsoft
Redmond, Washington.
A
Unix
badge" in
at
Company
math
at
star at Bell Labs,
Sun Microsystems
(right
behind
the "eighth
Bill Joy).
He
helped
Unix-based operating system that became
adapting the
1988, he also
Solaris. In
Tom would become
became famous
in Silicon Valley for
leading an engineering group in an overnight redesign of the
executive offices of
Sun CEO, Scott McNealy, and
his top assis-
tant into an indoor one-hole golf course, complete with sand trap to
and water hazard. The can-do
Tom was
doing what he could
CEO around the Sun headquarters, but finally
keep the golfer
gave up and defected, to found the IP Router-switch Ipsilon,
company
which was bought out by Nokia.
Brother Bob, a Cornell man, also became a serious star
at
Sun, leading the team that created Sun's search software and industry-standard
dominance
in
Network
File
System (NFS), keys
networked business computing.
to Sun's
When Sun
failed
to follow his ideas for multiprotocol storage systems, however,
he went on
to
co-found the prominent computer storage
ware company Legato, now owned by storage
were well behaved
All these kids really
or
keep up with
Tom and Bob
David and
at
home.
titan If
soft-
EMC.
Dick could not
his two-years-younger brother, Jim, in
math
in the intricacies of getting rich in software, or
Bill in
abstruse business consulting,
maybe
that
was
George Gilder
34
why he went
was why he worked so hard,
trying to get in
on the ground
the physical layer, building things and tearing
One
day
Dick had
floor.
it
floor,
apart. his office at
and was
trying to put
Mr. Lyon was chief engineer
Power Company, but he found
that
that the four-year-old
entirely dismantled the door lock
back together on the
them
home from
Power Company and discovered
El Paso
it
1956, Mr. Lyon returned
in
Maybe
Caltech rather than the Ivy League.
to
at the
hard to reassemble the thing so
the door would lock again. Mr. Lyon
hoped
Dick had
that
learned a lesson. But a year later Dick would be at work dis-
mantling the radio.
At the age of
ten,
Dick went with
his father to his office at El
Paso Power, where he showed the young boy the $100,000
Leeds
&
Northrup computer that controlled the generators
two power plants and transmitted output data back computers have two key aspects
tral office. All ital.
Analog functions connect the computer
essentially continuous characteristics,
radiations of light
bers,
all
to the cen-
—analog and
to the real
dig-
world of
such as flows of
fluid,
and sound, temperatures and pressures. The
screens, mice, scanners, microphones,
put devices are
at
and other input and out-
basically analog. Digital
computing
is all
based on binary on-off codes that can be shuffled
at a
numpace
of billions of computations per second in digital processors.
Mediating between these two worlds
numbers ers
—
are
complex devices called
(ADCs) and
Almost poses
—
all
—analog flows and
analog-to-digital convert-
digital-to-analog converters
processing
is
now done
chiefly simulations
— there
digital
(DACs).
digitally.
are entire
But
for
some pur-
computers using
The
Silicon
Eye
35
analog continuous voltages and currents to represent continuous
phenomena. Based on
real-world that Mr.
and gun-aiming device
a radar
Lyon had used during World War
II,
the Leeds
&
Northrop machine was an analog computer with a complex feed-
back system.
power
Its
electricity
purpose was
to simulate the flows of high-
through the circuits of the power grid with
flows of low-power electricity through the circuits of the puter.
The computer would compare
com-
the simulated values with
desired values and then "feed back" the differences or deltas to control the real
and the
power
lines.
The machine kept
the generators
balanced and loaded
tie-lines to other utilities
to a level
of the "lowest feasible incremental cost."
Lyon had persuaded
computer
As
in 1956,
a child,
and
his boss to it
had paid
buy the Leeds
for itself in less than a year.
its
intricate
a
dome. Dick
modern computer also
watched
or
camera the
fly
model airplane
parts,
under radio control.
were physical or analog machines, using real-world
materials and functions tal
size of the Astro-
closely for years in the garage as his
be assembled into planes that could
All these
you could proba-
today,
father plied a metal turning lathe to form to
this
arrangements of internal gears
and Selsyn motors. Using these devices
make
Northrup
Dick Lyon had been intrigued by the power of
analog machine and
bly
&
—inputs and outputs—
rather than digi-
numbers. Watching his father work with these machines,
Dick Lyon was learning
to
be an engineer
touch with the
in
physical foundations of his creed.
Now, however, back
in the utility
the entire project of raising this boy
room next
seemed
to
to the kitchen,
be getting out of
— George Gilder
36
hand. In there, with the washer-dryer and the water heater, Dick
had hung curtains and and dangled films
set
five
up
a developing
tank as
tall
he was
as
He had
feet long with laundry pins.
spread out boxes of cheap lenses, Kodak E-3 Ektachrome guidebooks, cans of black paint, and other
now
the whole toxic operation
where
it
Mom") were
full
siblings should
come
as a last straw,
in
On
What
if
in the garage using Dad's
the
A&W
one of
from the El Paso heat and take
he was
and
into the kitchen,
nine one-gallon
of malodorous chemicals.
Root Beer jugs
to
was overflowing
debris,
it?
was getting mixed up with food preparation.
kitchen table ("Don't worry,
Now,
—what was
his
a swig?
metal lathe
shape Plexiglas into pulleys to attach to the motor he had
stolen out of Mrs. Lyon's
handheld mixer. Soon they would
have to move out, eat
Pecos Bob's, and convert the whole
house into
at
combined darkroom and
a
laboratory.
What
all
did Dick
think he was doing?
Mr. Lyon had to admit that his son showed some promise in engineering feat of
stuff.
Dick had read about
photographing a bullet
with the noise of the gun.
ment he found
He
in flight
by triggering the camera
built a strobe light out of equip-
in his dad's garage
and
a
Xenon
ordered from the Allied radio catalog.
coil
Harold Edgerton's
Dr.
flash
and
trigger
set
up the
Then he
apparatus to photograph a Ping-Pong ball as his father hit
it.
But
Mr. Lyon and his son could not coordinate the equipment, including the Ping-Pong paddle, the well
enough
debris.
way
It
was
—more
in the dark. frustrating.
scientific
—
to
The
ball
ball,
and the strobe
kept getting
Dick decided photograph
it
lost
light,
amid the
would be better any-
a flying
moth.
The
Silicon
Eye
37
Delivering the El Paso Times and the Herald Post, he had
money by age
saved up enough
Olympus still
Pen-F.
He would
twelve to buy his
Dad tions
tried to
beam, and
The moth would
fly
out
persuade young Dick that making the connec-
would pose serious problems of impedance. Connecting a
would resemble linking
would need
a high
came home from
Dick
to lower the current to a
could accept. But by the time Mr. Lyon
a contraption using a light-sensitive
photo
one of the few vacuum tubes that Lyon ever used.
Rather than working with a direct wire connection, the actuator on automatic doors.
through a
to his strobe
the office that evening, Dick had bypassed the
impedance issue with cell resistor,
on the wires
a drinking straw to a firehose.
impedance source
level that the transistor
like
then
set off the exposure.
tiny transistor with the high currents light
camera, an
trigger the shot with a transistor,
a rarity in household experiments.
of a box, through a light
first
light
strobe flash. Mr.
beam focused on Lyon
still
it
functioned
The moth would
fly
make
the
the photo cell and
has several sharp pictures of the moth
in flight.
At the high school, though, the tivated by snapshots. trived
Moths
girls
in flight,
photo of a water drip in the
air,
were not
even an ingeniously condid not impress them. By
the time Dick was a sophomore, he had
crowd-pleasing strategy. tures
—pictures
the goal of
that
He
would
sufficiently cap-
embarked on
a
more
decided to make really gigantic pic-
wow
everyone
at
Austin High. With
becoming head photographer of the yearbook, he
set
out to contrive panoramic color shots of school scenes and events. Since he could not afford to
buy
a professional
pano-
George Gilder
38
ramie camera, he would have to build one from scratch. Mr.
Lyon had given him some an
In
effort to
ideas.
synchronize the exposure of the film with the
around
a tuna-fish can. Well,
tripod, the
it
was
at
a beginning.
This was his dad's idea, and
it
it,
as the
got
him
each point the image would be moving,
tures.
film
Attached to a
can could be turned by a hand crank that also moved
the lens around the film, exposing site.
35mm
young Lyon wrapped
rotation of the camera, the
Dick
however, that
recalls,
"it
was
a
camera scanned the started.
But because
made
blurred pic-
it
good idea
turned around: By fixing the tuna can and using
it
if it
were
to drag the
film around a roller to reverse the direction relative to the lens
movement and then can,
I
got the film
movement behind
rotating the
whole camera around the
movement synchronized with the lens
get the picture exactly, but
—
Anyway, Lyon showed himself
seem an Einsteinian all
moving
sun.
He
to
You may not
did.
be capable of what might
feat of conceiving several objects in El Paso
in relation to
figured that
the local image
for a sharp picture."
you may be sure he
fixed
if
one another, a piece of
the focal length of the lens
diameter of the tuna can, a
realistic picture
and the
film,
matched the
could be exposed.
Dick took several black-and-white pictures with the contraption. It
was
his first
panoramic camera.
This system never performed to his satisfaction, though, since the negatives
—an inch wide and
—were
a foot long
enlarge but not big enough to enable contact prints. to replace the
motor was
at
too big to
He
decided
crank with a small electric motor (Mom's mixer hand). With the mixer blades removed,
its
speed
The
could be varied and fingers.
He
Kodak
the
Edmund
used
aerial
would not
He
39
grind, blend, mince, or
chop
his
across
film
single-element lens from
the
box of cheap lenses went
for
$5 through
constructed a balsa-wood-and-plywood box for
He
the camera.
Eye
simultaneously to rotate the camera and pull
Scientific ("a
the mail").
tially
it
it
Silicon
painted
black. Trying out the device,
it
he
ini-
botched the coordination between the rotation of the
camera and the
film, yielding tall
skinny images. Working in
the garage, with the lathe, he contrived Plexiglas pulleys that
worked with
ball bearings to adjust the
movements, so
that
the one motor could accomplish the two rotation speeds for
and the camera. Finally the device began
the film rollers
producing handsome black-and-white pictures. That was Lyon's major camera number two. Black and white, however, would no longer do the 1968,
when Dick began
are
color,
brown and
in color. For a yearbook,
he noted, "even
gray."
in El Paso,
to
was hard enough just
to get
more demanding than
fatefully learn
Lyon had planned a
them
and hues. At the time,
it
an accurate-enough exposure (much
for black-and-white negatives)
do precise time and temperature control
that
would
light of the
also have to matrix the colors, integrating
for all the intermediate brightnesses
panorama
where most things
measure the intensity of
three primaries, blue, green, and red, as he
you
soon you
But color greatly complicates photography.
Not only do you have
later in life,
By
the project, most movies and magazines
and even family photos were
would need
trick.
in processing.
truly ambitious shot.
would capture the
entire
and then
He wanted
a color
quad of the school,
all
George Gilder
4o
long sepia buildings, the gray three-story main structure, the
its
gymnasium, the cars parked
in the
back
lot
by the tennis courts,
the flowers along the lawns, the whole polychromatic shebang in
one image. For
this,
he needed special Kodak Ektachrome
and a quarter inches wide and
five
aerial color
a problem.
Nowhere around
such images be cheaply developed. There was no
would need all
to figure out
El Paso could
alternative.
by himself the entire E-3 development
fix,
bleach, rinse, wash, stabilizer, and
That meant he would have
to take over the utility
room
developer tank and then the kitchen table for the overflow.
mom
think his
ever forgave him," said
the results were worth
ever centerfold.
showed
120 degrees.
Dad
"I
don't
in a jocular way.
Opening up over
of 1970 contained
But
one
photo covering
the other side was a near 360-degree
of the quad. Yellow flowers
its first-
a full four-page spread,
a black-and-white professional class
On
for the
it.
The Austin High School Roundup
side
Lyon
the fifteen separate steps, including developer, rinse,
hardener, color developer, dry.
designed for
five feet long,
photography
Here there arose
process,
film,
panorama
bloomed along the green lawns.
Parked in front of the three-story main building was a bright
VW bug.
crimson
In the utility areas
was an
array of blue
and
green trucks. Lyon had wanted to have an entire 360-degree cylindrical image, with a green
that
would have taken another page. He
pager
an
bush appearing
—
a
NBC
at
both ends. But
settled for the four
unique high-school yearbook displaying peacock.
yearbook picture
Many
in the
years later Lyon
its
colors like
would bring out the
course of courting his future wife, Peggy.
She was impressed: ''Wow! What a nerd!"
The
Silicon
Eye
41
Underneath the yearbook panorama was
a story about Dick.
Describing the prolonged calibration process for the coordinated
motion of the motor, the
film,
across the scene, he said,
"I'll
when
I
finally got
and won
my
first
Of
some
'til
before making any off to Caltech.
noncamera
the
panned
wrote up the process
of the spelling, he later
said: "This is
my
my fourth one"
fourth camera, Dick Lyon at seventeen
There he embroiled himself
also
in a series of
Mead.
worked with the
scientist of signal processing,
after retiring
won
of Sciences, senior divi-
projects under his advisor, Carver
Echo and
world's pre-
John Pierce, spearhead of
Telstar satellite projects,
who came
to
Caltech
from Bell Labs. Pierce got Lyon a summer job
Bell Labs. There,
still
only a teenager, Lyon invented
intensive digital signal processing digital
He
camera, Dick
you see
As an undergraduate, Lyon eminent
it
did fourteen somersaults
Academy
his future plans for the
third camera. Just wait
went
I
as
third prize in the Trans Pecos Science Fair. Refining the
place in the Texas Junior
sion.
bet
good one."
presentation and correcting first
and the camera
camera that took
(DSP) algorithms
digitized signals
example, and prepared them to be
from a
mapped onto
an undergraduate, Lyon became such a
DSP
at
new math-
for a kind of satellite,
for
a screen. Still
prodigy that he
taught the subject to undergraduates and faculty members,
including Mead. After graduation from Caltech in 1974, Lyon
Xerox
PARC
(Palo Alto Research Center),
moved on
to
where most of the key
components of the personal computer were
first
put together
George Gilder
42
into practical systems, including the interface, the laser printer,
mouse, the iconic user
and the Ethernet. Working with Eth-
Bob Metcalfe and David Boggs, Lyons won two
ernet inventors
patents on an Ethernet timing recovery circuit. In 1980, Lyon
invented the optical mouse. Used in various workstations from
Xerox and Tektronics, emitting a detectors.
beam It
it
defined screen locations with a mouse
of light detectable on a grid of solid-state photo-
was
a kind of device that later
as a neural network.
Simulating the intuitive and associative
processing of the brain
—with
inhibited neighboring neurodes
putations
among
became known
signals that either spurred or
—
it
clusters of pixels.
performed collective com-
Thus
it
could approximate
constantly changing and uncertain signals from the user
who
was moving the mouse by hand.
From Xerox, Lyon went
first to
Schlumbergers Artificial
ligence Lab, and then in 1988 to Apple Computer,
Intel-
where he
spearheaded the development of the second version of the Newton
handwriting recognition system
(the
one that actually
worked rather than the one that became haplessly famous
in the
much hype from
Doonesbury cartoon
for not working, despite
Apple). Through
these research roles, Lyon was piling up
all
patents by the score, most of cessing, pattern recognition,
By immersing himself boy, however,
them
related to digital signal pro-
and image processing.
in the
Kodak
color film process as a
Lyon had learned the most successful imaging
technology ever created. Invented in 1935 as Kodachrome, this multilayered film, using silver halide phosphors, represented the
climax of a series of color systems that dominated photography
The
Silicon
Eye
43
hundred years following James Clerk Maxwell's invention
in the
of three-shot color and additive color projection.
tems
captured the three primary colors, one
all
then green then blue
—and then combined them
or print, while the photographer
moved
ing had
second technique pursued
—red
for a projection
that noth-
moving subjects improved.
A
decades of the twen-
photographic transparency
patented by Auguste and Louis Lumiere of France, sensitive
time
at a
hoped against hope
in the first
century was autochrome.
tieth
earlier sys-
meantime. As the combining process
in the
accelerated, the ability to handle
A
The
it
used color-
dyed molecules of potato starch, covered with an emul-
sion spread across a transparent plate like a frame of film, to filter
the three colors at once.
when
a color transparency
The random mosaic
light
was reversed
filter
yielded
into the emulsion
through the same random pattern of grains. While
this
ungainly
process of "back propagation" yielded inaccurate colors and
modest
resolution, the pastel images
Twentieth-century
digital
were pleasing
to the eye.
cameras would repeat these essen-
three-shot and mosaic concepts with three-shot and mosaic
tial
charge-coupled device
CCD
(CCD)
systems.
With the advance
of
technology, this essential process of collecting and
filter-
and then computing the colors
after-
ing light intensities
first
wards has become quick and robust.
But the underlying
insight
—separate
colors
and postexposure mixing and mapping them
in a pic-
James Clerk Maxwell's. And the
CCD pic-
ture
—
is
the
same
registration of the intensities of three primary
as
tures necessarily suffer
from the process of
first
throwing away
two-thirds of the light (the other two primary colors) at every
George Gilder
44
pixel at the
time of exposure. Digital cameras then recalculate
the colors from the hues of neighboring pixels.
As Lyon came
camera technology
to see, this digital
taking over in the
new millennium
of capturing colors, a major step
backward from Kodachrome.
James Clark Maxwell and mosaic
filters
transcended both
by capturing
colors at
once
the film.
By allowing one quick shot on one piece
tering
all
three
as they penetrated to different vertical levels of
colors
all
was
was, at the most basic level
CCD, Kodachrome
Unlike the one-color
that
and
all
the light at
Kodachrome eliminated the motion and vious systems. Banishing
random
all
of film, regis-
locations
at
once,
overlay problems of pre-
filters
and processing guess-
work, Kodachrome would rule color photography for nearly the next
fifty years.
Eventually, though,
new method
tions to a
CCD
Lyon himself would make
digital
vital
of imaging that could eclipse both the
camera and the Kodachrome system. Laboring
through the fifteen steps for developing one color
he
let his
contribu-
mind wander
slide,
perhaps
to the possibilities of the transistors that
intrigued him. Perhaps there began to develop on the subcon-
scious substrate of his
Like the Ektacrome,
it
mind the
idea of a solid-state camera.
would capture three
colors of light at
three different levels at every point on the image plane, with no special color
enough
to
filters.
combine
developing process
With still
all
colors
on one chip,
and motion
at all. All
capabilities
it
would be
fast
and require no
the photographic procedures could
occur on the single silicon chip that also captured the image.
Advancing
at the
speed of the computer industry rather than the
— The
speed of the camera industry,
Eye
Silicon
this
45
breakthrough chip would rep-
resent a total revolution in imagers that would achieve both
speed and resolution exceeding either Kodachrome Technicolor movies.
And
it
slides or
would be hundreds of times cheaper
than the equipment of the day.
That was merely a dream. Amid
all
his distractions,
got around even to trying to build his promised fourth or the fourth great technology in the history of digital until
he came
to
this
new machine
—
to
cameras
and Lyon would need
a lot of help
from a
his
old
in building
be named the Foveon imager,
high-resolution color-sensitive "fovea" in the eye
Vermont.
camera
Foveon Corporation and rejoined
Caltech mentor and collaborator, Carver Mead. But
he never
after the
—both
silicon wizard
Mead from
Dick Merrill with a Sigma 10 camera containing his
handiwork
—
the
designed and built. will
become
as
PCs. According
Foveon imager that he
He wonders when
common to
largely
"Foveon Inside"
a logo as "Intel Inside"
Carver Mead, Merrill
creative engineer I ever met.
"
Dick Merrill
is
"the
is
on
most
—
3
Back—
to the
ice
Merrill's
View
Future. "Doc-tor"
Emmett Brown
played by Christopher Lloyd
at
your serv-
—there you have
it
the antic image of Dick Merrill. Wire-rim glasses, long
white coat, electromagnetic blond
hair, a bright feral glint in his
skyblue eyes, Merrill will be a familiar figure to
time travelers. Just step into this vehicle, with a
monochronometer
wavelengths of
light,
and attach
a
your highly analog flux capacitor called
it).
Then read out
all
you
faithful
new all-aluminum enclosed inside to emit just the correct
few score wires (that's
to the pins of
what Steven Spielberg
the data. "Doc" here will have you wing-
ing back into the future in a nanononce.
The reason he can do
it,
you understand,
is
that he's ignorant.
"Never underestimate the power of ignorance," Merrill doesn't
mean
that he
is
stark ignorant like an actor
says.
who
thinks
he can predict global weather over the next hundred years. is
ignorant and
of knowledge
—
knows he is
is,
and conscious ignorance
power. In other words, he
47
is
is
He
a
He
form
an intuitive exper-
George Gilder
48
[mental scientist. That
is
different from being a highfalutin
expert or an academic Ph.D. Merrill
man who went
to
work
first
is
way
of a
electrician
and
ignorant in the
as a carpenter
and
He
only slowly built up to chip design and silicon manufacture.
learned his trade without "being
how things
bombarded by you
are for so long that they give
endurance and
professors about a doctorate for
servility/'
"You know." he sums up, "I'm not one of those Caltech guys. "I
.
.
might not
school. six
.
Nowhere
reasons
why
too ignorant.
in
fit
I
vour
near.
I
story.
can't just
my
class in
do the equations and
tell
not going to work, like Dick Lyon can.
it's
will
go build the chip anyway
build chips
—and maybe
some other
useful effect, and
In other words,
wasn't at the top of
I
he
is
it
—
I
I
am
know how
won't work exactly, but it
you
I
to
will find
solves the problem."
a genius with silicon.
His problem-solving orientation served him well back in
Woodstock, Vermont, where he grew up, of a prominent local physician, he things. After graduation
became
a carpenter.
in his first job.
w as oriented toward building
from high school
While on the
The son
in the late 1960s,
job, though,
serving chiefly as an electrician. At that time in
he
he found himself
Vermont you did
not need any degrees or certificates to wire up houses, and unlike his boss he wasn't afraid of electricity.
None
of the houses
wired by Merrill burned down, to his knowledge, and during the course of his work he figured out
could turn the lights on or off
back and forth three times. Merrill savs, he
is
none too
It
how
to
equip a room so you
at either side, folding the wires
was
swift
a real puzzler for a while
—but he worked
it
out.
—
as
The
Five years
and physics
later, after
Silicon
Eye
he had made
at the University of
49
the
it all
Vermont
way through math
in Burlington,
tured across the Connecticut River to Hanover, shire,
he ven-
New Hamp-
twenty miles northeast of Woodstock, to study electrical
engineering at Dartmouth's Thayer School. There the experi-
ence with three-way wiring saved him a night of homework. In
one of
stumped most
his first classes, the professor
classmates with exactly that problem
—how
of Merrill's
to wire a
room
so
the lights can be switched at either door.
After graduation with a double neering), Merrill at
went
off to
E
(a
degree in electrical engi-
IBM's famous Watson research labs
Yorktown Heights, where he learned the alarming lesson that
ignorance can be even more effective
when
Exploring early Metal Oxide Semiconductors
—he
worked
microchip technology
endary inventor of
As and
memory
for
laced with alcohol.
(MOS)
Robert Dennard,
stint of
went
off for a long
it
might be good
IBM
inventor
lunch in April 1964 and began a
real work.
if
sudden pang of conscience. They
they accomplished something before
manage
all
of
them were
So they decided
to
do some Really Important Work,
leaving for the day. But
worthy of their glorious
memories
ing producer of
too drunk to
state of inebriation.
Dennard asked, "What agreed,
leg-
competitive boozing. Later, stumbling back from the
restaurant, they suffered a
thought
basic
chips.
Merrill heard the story from Dennard, the
his staff
—the
is
really
are important;
memory
chips.
important to IBM?" Everyone
IBM was Memory
then the world's lead-
chips consist of a tiny
grid of wires attached to an array of "cells" that store a desired
information bit at each point in the grid where the wires cross.
George Gilder
50
The problem
is
how
to stop the electrical
the bit from draining
At the time, each memory
to electricity. sistors that
silicon, cell
which
is
a sieve
consisted of six tran-
kept the bits alive by passing them back and forth in a "flip flop." Together with the wires
what was termed them, the
away through the
charge representing
six transistors
took so
much
between
space on the chip that
these memories were limited to the low thousands of bits rather
than the hundreds of millions or even billions of bits possible today.
"Well,
let's
invent
By the end of
some
better memories,"
a roisterous lunch, the group
idea of a single transistor storing the bit in the capacitor,
two
Dennard
memory
form of an
which consists of
a
cell.
It
said.
had arrived
at the
would be based on
electrical charge
on
a single tiny
conductor sandwiched between
insulators.
"Wouldn't "Yes,
it
just leak off?" asked
guess the capacitor would flux," Dennard said, "but
I
could be measured itance
one of the group.
on the
cell
at the
sum
of the capac-
You would have
to refresh the
end of the
and on the
wire.
it
line as a
cell constantly."
"Well, a
good
it
might work," said one of the team. "That sounds
idea,
you might
call
it
a
dynamic flux
like
capacitor. Call the
lawyers." In
memory berg,
IBM
1968, cell,
patented this one-transistor flux capacitor
and the
rest
is
history.
Bob Noyce, Steven
Ted Hoff, Emmett Brown, and the
rest of
Spiel-
them adopted
it
in various
ways and dynamic random access time-travel became
a surefire
theme from
Feynman
to
Silicon Valley to Hollywood,
Michael Crichton.
from Richard
The
Silicon
Eye
51
Anyway, that was how Merrill remembered Dennard remembering the invention of the now-prevailing technology for the
DRAM,
dynamic random access memory
the
Research offers a more sober thin film magnetic
work"
in
memories and
"many months
unique research environment"
the "one-transistor dynamic the working
memory
based single-transistor in
entailing
conference on of
an arduously "disciplined innovation process" using the
scientific resources of "a
all
tale, starting at a
IBM
chip.
memory
At the heart of nearly
PCs
in the world, the capacitor-
cell is
one of the key inventions
in all the
DRAM
cell."
to create
semiconductor history and according
to
IBM
bravado, "the
most abundant man-made object on the planet Earth."
Out son.
of his experience with Dennard, Merrill took a vital les-
He came
tern detection
to
understand that the secret of invention
and extension. "There's
a lot
you can get
is
pat-
in this
world just by looking for symmetry, looking for patterns," he says.
The tors
pattern of advances in silicon
per
bit.
The
memory was fewer
zero transistor bit? Sure.
IBM
transis-
had already
invented the hard drive in 1956, based on magnetic domains.
"Look
one area and apply
for a technological trend in
another," Merrill said. Using this technique, Merrill ating patents for National eight a
month when
meeting on tor)
CMOS
Semiconductor
in early
at a
it
to
was gener-
pace of some
1997 he met Carver
Mead
at a
(complementary metal oxide semiconduc-
imagers at the National offices on Lawrence Avenue in Sun-
nyvale. Since the mid-1980s,
structure of nearly
and other electronic Merrill
all
CMOS
has defined the essential
the digital chips in personal computers
gear.
met with Carver Mead on
several Fridays at National
George Gilder
52
he did not even know
who Carver
and
at first,
this
guy doing imagers that were not going
to
to
was.
"He was
just
work. They'd seem
work, they gave great demos, but those Caltech guys, they
know
a lot of stuff, they
they don't
know
— they
know much, much more than
know
can't
making something work once repeatedly, in volume, in that stuff because
I
mass production
do, but
difference between
and making
in a laboratory in a
wafer
fab.
I
it
know
have had twenty years of agonizing experi-
ence making things by the millions
in
Merrill's frequent habitat, where raw ingots of
—the
I
silicon
wafer fabs."
a wafer fab,
is
a chip factory,
from beach sand are transformed into
microchips through the most complex and rigorous manufacturing process ever conceived by
minuscule devices
man. Although fabricating the most
in all industry,
commercial wafer fabs are now
mostly too large to compete for real estate in Silicon Valley. Sur-
rounded by moats and metal fences and chemical septic pools, bulbous with huge tanks of deionized water and chemical gasses
and buttressed and suspended
to
absorb the shock of 8.0 Richter
scale earthquakes, attended by workers hermetically sealed in
white
suits,
these structures
modern version society's
smack
of
some
sleek and angular
of a medieval fortress, designed to protect the
most precious treasures from the outside world.
Sequestered in remote parts of the globe, wafer fabs loom on islands such as
Kyushu
in
Japan or bleak desert reaches such as
Chandler, Arizona, or even on the wilder edge of Portland,
Maine, where National Semiconductor makes
its
chips.
— The
Silicon
Arrayed inside these factories
Eye
53
a concatenation of shiny
is
metal machines linked by robotic arms moving quartz boats of silicon wafers
—
thin gray circular disks usually either eight or
twelve inches wide
—on and
may run on automated equipment less
like a
automated
may be borne by a controlled
is
These containers
tracks that traverse spaces
model amusement park
sites
machines
ing the
off of carriers.
between the
roller coaster or at
fab workers. Surround-
environment with mazes of
pipes carrying the millions of gallons of deionized water or
minute quantities of exotic chemicals into Turbines and
strategic positions.
constantly clean and process the atmos-
filters
phere in "clean rooms" that are classified by the number of particles smaller than
foot of
air.
particle
Though
500 nanometers
in
diameter in a cubic
entirely invisible to the
looms across
a
microchip as
naked
a "boulder."
eye,
such
a
Leading-edge
clean rooms at companies like Intel or National Semiconductor are "class ten" or
even "class one," meaning that there are
between zero and ten such
particles,
making
a
microchip fac-
tory tens of thousands of times cleaner than a hospital's cleanest surgical facilities.
Using hundreds of millions of dollars worth of the exquisitely calibrated equipment,
wafer
lasts six
the fabrication cycle for each silicon
weeks and
entails
between 700 and 1,000
specific
process steps. Transcribing the chip's circuit design onto the silicon surface
is
photolithography. visible
a specialized photographic technique called It
bathes the wafer in near-ultraviolet light
wavelengths being between four and eight times larger
than the 90-nanometer features that must be inscribed on the
George Gilder
54
The
chips.
ultraviolet rays shine
through reduction lenses
in a
process that resembles a slide projector with the optics reversed to miniaturize rather
than expand the image.
Interspersed with the photolithography steps are hundreds of repetitions of chemical and physical etching, cleaning, ion
implanting, diffusing, and "doping" the silicon layers with needed
elements to create millions of nanometer-scale devices on every chip.
Each wafer holds hundreds of separate chips
cooked
batches
in
at the appropriate
that are
temperature for each chem-
reaction or physical deposition and then diced apart after
ical
they are probed and tested.
equipment tweaks, and specialized
Full of black arts, this wafer-fab
process lends
lore,
few masters the mystique of
its
a
kind of priesthood. Whatever grandiose plans for microchips
may emerge from
the systems engineers or computer-aided engi-
neering teams at
company design
microchip firms such as Foveon, rill
who
ultimately determine
Merrill, however,
chips and invent
was
new
centers or from "fabless"
the fab masters like Mer-
it is
what can be
a rare fab master
built
who
and how.
also could design
devices himself. Merrill thus could inno-
vate not only in chip design but also in chip fabrication, enabling
new
devices that could not be created by the usual processes.
This capability would prove crucial
where the usual industry-standard be adapted
to create
new forms
Patents had been issued on
CMOS
large to enable
Foveon Corporation, processes needed to
of analog photodetectors.
CMOS
But until the early 1990s, the feature
were too
at
imagers as early as 1964.
sizes of
CMOS transistors
an imager with a reasonable number of
The
pixels.
Silicon
charge-coupled devices (CCDs), a custom compo-
nent partly invented by Carver less
the
55
For imagers, the industry turned to specialized technol-
ogy, chiefly
up
Eye
Mead and
his students that took
space on a chip than transistors did. Like
field,
many
others in
though, Merrill calculated in the mid-1990s that by
then microchip progress had advanced enough to enable an
imager that uses a cleverly tweaked and controlled version of
cheap high-volume
CMOS
transistors instead of
CCDs.
At the time, the entire industry of analog semiconductors was pursuing what was termed "complementary technology." By
—two
using both positive and negative electricity every measurement
immunity
mon
to
to noise
— the
complementary approach offered
and other environmental conditions. Com-
both of the complementary
cancel out
when
sources for
signals, these influences
the difference between the two inputs
is
calculated.
In 1994, three years before meeting create a pixel with
Mead,
Merrill decided to
complementary output. Exempt from
ronmental interference,
it
would allow more
stable
envi-
and accurate
imaging. Building this complementary pixel, he discovered that
the two channels responded differently to different wavelengths of light. Different wavelengths penetrated different depths into
the silicon, making tary pixel. this effect
it
impossible to
But Merrill
—
make
a reliable
complemen-
listening to the technology
—thought
might be used in a silicon imager.
With three
colors penetrating different depths, a single silicon
device might render a full-color image, employing the physical properties of the material.
His
first
tries
yielded impossibly
George Gilder
56
murky
results,
but he thought something might be
might not be sharp but the
manufacturing
at least
details,
National's patent lawyers tion.
One
it
made
of
it.
"It
would be cheap." Elaborating
he registered
and proceeded on
his
concept with
to his next inven-
patent idea down, seven to go for June 1994.
The Charged Competition
proposing a
InMead,
Dick
new
silicon
Merrill,
challenging one
of
image plane
for cameras,
Carver
and Dick Lyon of Foveon would be the
proudest and most profitable
redoubts of Japanese industry. True,
like
almost every other early
invention in electronics and communications, from the transistor to the laser, the at Bell
charge-coupled device
(CCD) was
Labs, then the giant research arm of AT&T,
invented
now
serving
Lucent and Telcordia. But the Americans went on
its
offspring,
to
more adaptable memories. The Japanese adopted the
and converted light
it
from a memory
cell into
CCD
an imager, a kind of
meter that could convert photons into electronic pulses.
In inventing the device in 1969, Bell Labs
was seeking what
might be described as a tape player on a chip: a with a capacity in megabytes that offered tinuous stream. At a time
huge, and
Intel's first
held just 500
bits, a
when
DRAM
its
transistors
serial
memory
contents in a con-
were
still
relatively
(dynamic random access memory)
successful megabyte
57
memory
chip meant
George Gilder
58
memory domain
shrinking each
space as possible.
to as tiny a
This goal prompted Dennard's pursuit of the one-transistor
DRAM
cell at
IBM. The
netic spot, like the
was not a
ideal
domain on
a disk drive.
seemed
answer the need
to
Because the rectangular
that could efficiently pass
cameras shutter
dense memory device.
capacitors proved to be
contents to a
as
an imager
memory when
the
closes.
Leading the quest for the lab chieftain Jack to costly
its
Holding a movable
showed promise
also
on
for a
CCD
grid of
CCD
sensitive to light, the
mag-
Dennard's "flux capacitor," the
flux of charge in a capacitor, like
CCD
transistor but a
CCD was
first
Bell Labs' solid-state
Morton, who was seeking a
silicon alternative
magnetic memory for picture phones or answering
machines. Responding to Morton's quest were two engineers, then in their
thirties,
The analog
named George Smith and
in silicon of a
Willard Boyle.
magnetic domain
is
an electrical
charge on a capacitor. Smith and Boyle conceived of a long array of capacitors that passed their charges one to the other, as in a bucket brigade,
when dragged
along by a voltage stepping along the top.
Patented by Bell Labs, this prototype charge-coupled device
was
a first step toward the technology that ultimately
ply imagers for
would sup-
most of the world's cameras. However, when
Smith and Boyle presented the device
to
an International Solid
State Circuits Conference in 1971, they were surprised to find
young Mead student written with
Mead on
Mohsen and Mead, voltage to
Amr Mohsen
move
improved
a sharply
the
presenting a paper he had
CCD.
new charge-coupled
the charge
down
the line
Conceived by
device used a push
more
swiftly
and
reli-
ably than a pull voltage could. Although a Bell Labs researcher
The
Silicon
Eye
59
stood up during the session and declared that
work,
current
all
CCDs
columns
would never
use the Caltech design. Pushed by a
shifting positive voltage, electron charges registers or
it
to a horizontal buffer.
move down
vertical
Read out by an on-
chip amplifier, the charges are converted into a digital image. In the great imager confrontation
Foveon and the Japanese,
Mead
now under way between
thus can claim at least partial
authorship for both of the contenders. But back in 1971, Intel perfected Dennard's
DRAM for "working memory" and the eras-
programmable read only memory (EPROM), now
able
for software storage.
memory
The
CCD
went the way of
But when Bell gave up on the took over the project.
ist at
moved on
CCD,
Now
key developer, Gil
its
famous
to other inventions.
Ame-
as a turnaround special-
Rockwell, National Semiconductor, and Apple Computer,
Amelio was then
a
young researcher working
Boyle at Bell Labs in Murray turn around the
When
CCD
Hill,
CCD
sufficient
Smith and
He
resolved to
Jersey.
Labs
to Fairchild
for imaging.
Amelio moved from
and made
for
New
Bell
ductor, he set out to adapt the device for this
the
specialized
devices unfavored by the onrushing advances of silicon
chip technology. Bell Labs soon
lio,
all
"flash,"
improvements
to
new
Semicon-
application
be credited as a father of
imager. In the 1970s, Fairchild created a small busi-
ness in the product for
NASA
remote sensors on
satellites
and
space probes.
making many humble
visits
both
to
Murray
Hill
and
to Sil-
icon Valley from the 1950s through the 1970s, however, were
George Gilder
60
Sony founders Masaru Ibuka and Kazuo Iwama, together with
Makoto Kikuchi,
a scientist
and the builder of Japan's
from Tokyo's Electrochemical Lab
first transistor,
who became
Sony's
research chief. As American companies Bell, then Fairchild,
then Texas Instruments, and then Rockwell
Sony wrestled with the technology
field,
dropped from the
all
for
some
thirty years,
gradually eking out a device that could be manufactured by the
thousands, with a yield of one or two per wafer. Sony's cofounder, third president, and transistor
devoted
much
of his career to pushing the
unforgiving gauntlet. of his gravestone in
Iwama never But
Embedded
Kamakura
is
lived to see the
Japanese-made
in the
a
CCD
through
this
upper right-hand corner
CCD
chip.
widespread success of
CCDs.
pioneered most successfully in Sony
like the transistors
handheld radios and in
champion Kazuo Iwama
like
the lasers that were used most widely
CD players, CCDs from Japan finally outper-
formed those of the American inventors. The Japanese are now manufacturing ing a
them
CCDs
by the scores of millions and incorporat-
into ingenious
camcorders and
digital
cameras that are
key source of profits for Japanese industry. In Japan, the
CCD saved the solid-state division of Sony Cor-
poration in the 1980s after sors
and
some
calculators.
It
it
failed to
six million a year, to
become one It
microproces-
early twenty-first century. Far
down
at a rate of
of Sony's most profitable
became
highest-performance-image plane for
become one
in
enabled the camcorder, made
products through the 1990s.
has
keep up
the lowest-noise and
digital
photography
the learning curve, the
in the
CCD
of those established products that has gener-
The
Silicon
Eye
61
ated an entire industry, supply chain, financial keiretsu, and of manufacturing skills
and machines.
terintuitive but remorseless
It
web
benefits from the coun-
law of business philosopher and
pundit Peter Drucker ordaining that no innovation can displace
an established system unless
it
is
ten times better. Almost no
inventor wants to hear this law, which disgruntles patent holders
around the world. But
it
remains mostly
true.
5 Annealing Neurons
A
the heart of the science of late 1960s neural net-
t
—computers inspired by model —was analogy with the
works
.brain
steel. Steel is
annealing cooling
a
a bizarre
tempered and strengthened by
—heating
it,
of the
human
processing of
a technique called
up
again,
in a repeated cycle of diminishing amplitudes,
some-
it
up, cooling
it
down, heating
it
times climaxed by plunging the metal into icy water.
Try
it
some time on your own neural network,
yen for nerves of
Sauna,
steel.
if
you have a
oscillate, sweat, jiggle,
shudder,
shake and bake, and dive into the wintry bay by the Golden Gate as
some people do every year
refresh you. For a to
new
in
San Francisco.
to find a
will surely
—planning going cameras —you
utterly
Silicon Valley start-up
transform both computers and
have
It
superman
or two,
men
are
to
of steel. Perhaps annealing
offers a recipe.
What
is
going on in an annealing process, like the process of
developing a film, or even bringing closure to a meeting or pur-
63
George Gilder
64
pose to a company,
is
a quest for a desired condition of the mol-
ecules of the material. With a film camera you want an arrangeof silver halide molecules that convey a targeted image.
ment
The developing process
tips
and sloshes the film through
when
cession of chemicals toward a "program" created
image
is
result of the computation,
With
a
conforms
to the correct
such as the recognition of
a face (hi,
meeting you want an enduring consensus. With
a piece of steel,
you want an arrangement of
that will not break
A
the
exposed. With an analog computer you want a configu-
ration of electrical energy patterns that
Carver).
a suc-
crystalline
under pressure or sag
form
—think
of ice
softly
molecules
ferric
on
—arranges
a hot day.
the molecular
grains of a metal in an aligned structure, riven by shear lines
prone to fracture. Ice resembles a
digital
system
—
and
perfect at
every step, and perfectly fragile in the face of a mere bug.
Whoops! Emerging from the brine with you too may
feel
crunchy and
frost
crystalline,
on your eyebrows,
ready to shatter into
shards.
On
A
heated metal gravitates to a homogeneous
fully
leable will
the other hand, you don't want to be cooked either.
clump
that does not hold
its
form.
Out
array, a
mal-
of such a glop you
not even be able to extract an image.
What you
desire, as a
of molecules that
is
man
of steel,
is
a
random
distribution
free both of the fault lines of a crystal
and
the ductility of heated metal. To arrive at this point, you need to
heat up the molecules, jiggle
them around, and shake out
the crystalline clumps and shears, without melting
shapelessness. Your
man
of steel cannot be a blob.
them If
into
you are
annealing a sword, you need a bendable blade with a sharp cutting edge.
When
you get what you want, you plunge
it
into icy
The
water to preserve the
Silicon
result.
Save
Eye
as:
65
Deadly-weapon.doc. Or
Supermancomic.gif.
Consider strike
you
metal,
it
a computational result. If
computing does not
as a plausible analogy for the processing of a
don't
computing
That
worry.
as a
is
because schools tend
lump
to
of
teach
branch of mathematics, an abstract discipline
separate from the physical world. But the physical world, as it
changes,
the state of
is
its
constantly
doing computations,
transforming
molecules in accord with the algorithms of phys-
ical law.
Thus computer
scientists see the universe as the ultimate
analog computer, processing cosmic cycles of astrophysical, chemical, and geological change toward a graphic result ren-
dered as the landscape, weather, and climate you see through your windows. In that way, computer scientists differ from
plumbers, for example,
who
rarely claim that the universe
is
merely a Kohler machine, a particularly clever arrangement of
cosmic pipes and valves, though such a notion ble.
Carver
Mead
is
equally plausi-
teaches computer science to biologists and
other beginners as a form of plumbing hydraulics. In annealing, the successive steps of heating
down
a material represent efforts to jiggle
up and cooling
and seduce a system
toward a desired outcome, shaking out suboptimal fixations or energy disequilibria height.
—
errors
—on the
way. Analogous to heat
is
Both hold potential energy. Imagine a pinball machine the tray up and
with two knobs that enable the player to
tilt
down and back and
maneuver
forth using gravity to
several balls) forward toward the goal.
the bottom (or pops into a hole),
it
Once
will not
a ball (or
the ball pops out at
come
back.
It
has
George Gilder
66
arrived
at
its
energy equilibrium.
physics
In
math, such
processes are governed by formidable procedures called "Hamil-
Rowan
tonians," after the great Irish mathematician William
Hamilton. In 1834
method
general of
all
at the
age of eighteen, he introduced a "a
dynamics, by which the study of the motions
in
free systems of attracting or repelling points
is
reduced
to
the search and differentiation of one central relation, or characteristic function,"
which might be
gravity, or voltage, or
thermal
entropy.
Neural networks treat computations as downhill movements toward equilibrium. During the hot phases, the atoms material are high
up
in the
in the topography, full of potential energy,
hopping around unpredictably. The solution comes when the system cools off and gravitates to the valleys, the basins, of a low energy equilibrium, the solution set by the computer program or
camera image. But on the way they may come trap,
caught in a tree or a high
shaken
free.
When
some suboptimal
a
to rest in
from which they must be
valley,
company becomes
goal
—
it
often
ship. Early in the history of the
an upland energy
so ensnared
—pursuing
must be saved by new
Foveon
leader-
project, long before there
was a Foveon company, long before Dick Merrill and Dick Lyon actively entered the scene, the crucial leadership
legendary figure of Silicon Valley Faggin
who
called
on Carver
named
Mead
new
science of the
—the
processor
human
Federico Faggin.
to help
Faggin was shaping called Synaptics.
Its
came from
him with
goal
brain to create a
central processing unit in a
was
new
a
It
a
was
company
to exploit the
kind of micro-
computer or camera.
Part
Two
THE FORGOTTEN
FOUNDER
Federico Faggin at the beginning of his career at Olivetti in 1961 in
Borgolombardo,
Italy,
flanked by technicians helping him
culator system he built boards.
At
Intel he
test
a cal-
using some one hundred printed circuit
would put the same computer power on a
chip. Courtesy of Federico Faggin
single
6 FF
Many is,
Americans have no idea of who Federico Faggin
why
his
his
name
to the cognoscenti,
a
expectation in the Valley.
who knew enough
Fah-jean, Federico Faggin
circuit design,
company would cause
a
at
comment and
susurrus of
However,
arrival
was
to
pronounce
a paragon of integrated
and the man who contrived the basic microchip
technologies that enabled the creation of silicon computers and
cameras.
If
fabulations
you could see through
—
McKenna PR journalists
suits,
the Intel publicity fogs and
sweet smiles and smoke from
Regis
with their jargon and gentle guidance for
and historians
was comparable gods,
the
all
all
—you
might understand that Faggin
in his contributions
even
to the legendary Intel
Bob Noyce and Gordon Moore.
Working inside
Intel,
an
Italian
immigrant, age twenty-three,
with skewed English and few friends in the Valley, Faggin had
been the true creator of the microprocessor, the central processing unit in nearly every personal computer. This
69
was the product
George Gilder
jo
that
had made
Intel the
main company of the California econ-
omy, the core holding of every technology investment portfolio,
and the crucial physical foundation the
The microprocessor
north.
Japanese
— seemed
memory
and the
to
be slipping away
Key
chips.
in the
fast
to
that
mid-1980s
—and
amid
Intel's
a tide of
computer revolu-
to the personal
rise of the Internet, the
empire
was the product
the United States' lead in semiconductors
profits as well
tion
also
from Japan
retrieved the technological edge
when
for the Microsoft
microprocessor could even
be deemed the basis of American commercial power and thus ultimately
its
military might. Lying at the heart of nearly
all
major products of U.S. industry, microprocessors inform the cruise control
and power brakes
television sets,
manage the
in cars, regulate the tuners of
interior
environments of tanks, and
guide missiles to their targets.
One-tenth of the weight of
rival solutions,
the original micro-
processors thrived in applications requiring absolute minimization of
power and
size.
The
first
one, Intel's
developed, fabricated, and tested by Faggin
Apollo spacecraft that took
men
to the
silicon space.
—was
—designed, used in the
moon.
In those days of large transistors, there
on
4004
was
a
huge premium
You could not waste area on the chip. In the
design of the 4004, Faggin adhered faithfully to the rule of
minimization. But on the face of the chip, integrated with the
thing extra. area were sistors.
where
it
could not be removed, he added some-
Consuming
a precious square millimeter or so of
silicon circuitry
what appeared
Through
a
to
be two malformed additional tran-
microscope, the concerned historian could
The
Eye
Silicon
71
see that on the border of the chip Faggin had
miniature Fs
—
FF. Faggin's English
was
embossed two neat
still faulty,
but he was
not shy. Yes,
we
knew,
all
all
of us in the know, that
erico, to the extent that
signed the all
first
man
one
it
was
really
could be credited,
Fed-
who had
microprocessor and was the proximate source of
the bounties of
creation.
its
Even more important, though,
in
enabling the whole industry to prosper, was Faggin's previous
development of a new fabrication method called the
silicon gate
process.
Electronics cations links
is
a
technology of wires and switches: communi-
combined with
controllable transistor cross-points
that can perform logical or sensory functions. In a world of ever
more
trillions
Whether
of transistors, the switches
become
in brains or microprocessors, all
relatively easy.
computing
is
ulti-
mately limited by the wires, by the reach and bandwidth of communications. results,
If
you cannot integrate the various intermediate
you cannot
square of the
finish a computation.
number
of nodes or transistors, wires congest the
silicon surfaces of chips
of the brain
Multiplying by the
and the dendritic wetware of the cortex
and the backplanes of computers and the ganglia of
eyes and the wiring closets of businesses and the ionic passages of the central nervous system
and the under-street
circuitry of
big cities.
As they increase exponentially with the increase
in transistors
and processors, wires eventually constrain computation and force localized solutions
and
distributed architectures.
As Mead
remarks, "In the end everything gets choked off by the wires."
George Gilder
72
That
is
why, after a few billion years of evolution
several decades
of
—and despite
enhancements and upgrades from IBM,
Cray, and Intel, despite even the rise of fiber-optic worldwide
webs
of glass
ticular brains
and
light
—
intelligence
is
still
distributed in par-
and machines, rather than centralized
in
some
transcendent supercomputer in the sky or some pedestal in an air-conditioned computer shrine.
As
historic as
his
Faggin's role in
more fundamental
processors,
was
was
making the
first
in the industry's long-run
micro-
advance
contribution to relieving the problem of the wires.
Before Faggin, every transistor on a chip had to have a metal wire attached directly to
minum necting
formed
it.
This was called a metal gate, an alu-
link to the switching transistor giving it
it
power and con-
to all the other transistor switches that collectively
a processor or a
memory. Since the
of silicon and the wire of
transistor
was made
aluminum, the main problem of semi-
conductor manufacture was joining the two incompatible substances together in a robust, reliable way. If
need
your brain used metal gates, your head would weigh a ton, a direct link to the
Grandma's face
in a
power
grid,
photograph.
and
still
The need
sharply curbed the industry's ability to put
couldn't recognize for a
more than
dred transistors on a chip. Metal wires melt, absorb at temperatures far below
silicon's
metal gate
drift,
thermal
few hun-
a
leak,
and
limits, often
reached during the process of manufacture, which entails multiple
cooking steps. Restricted to temperatures below the melt-
ing point of metal,
impossible.
many microchip
fabrication steps
become
The
While
still
at
Fairchild
Silicon
Eye
73
Semiconductor Corporation with
Noyce, Moore, and Andrew Grove, Faggin developed a way to replace these metal wires with wires
the
amorphous version of the
Although
scientists at Bell
made
of polysilicon,
crystalline silicon of the chip.
Labs and
at
Hughes had
patented some form of silicon gate, they had never
previously
made
the
technology work.
Outside the semiconductor industry, the romance of patents
and inventions captivates observers. But within the industry patented inventions are mainly a nuisance that allow others to obstruct your work but do not enable
That
is
why
and shared.
them
to
consummate
it.
across the entire industry patents are cross-licensed Intel's
patent portfolio, for example,
is
Texas Instruments, National Semiconductor, and
shared with all
the other
microchip firms. In general, companies achieve competitive advantage not by patenting ideas, exposing them to the world, but by hiding crucial process steps that are latent in densely
wrought and many-layered
silicon devices.
In electronics, ideas tend to be relatively easy.
They mostly
consist of taking functions, such as computers and cameras,
already performed on a large scale, and translating small-scale
microcosm of microchips. What
is
them
into the
hard are not the
patents (patent derives from the Latin word meaning to
open") but the latents (from the Latin word meaning to den"):
linking
"lie
"lie
hid-
thousands and then millions of transistors
together in an inscrutably complex chemical and physical weave of layers that reliably executes (and conceals) the idea.
matters
is
process:
making the devices
actually function.
What
George Gilder
74
Thus the world may
for "inventing" the integrated circuit. For his
won
a
Instruments
exalt Jack Kilby of Texas
achievement, he
Nobel Prize and appears on a postage stamp. But the
industry saw that Kilby's design, entirely metal gate, with
its
gold wires arching through air across the top of a chip of germa-
nium, was not manufacturable. The industry knows that before co-founding
Intel,
Bob Noyce invented
the real silicon inte-
grated circuit while at Fairchild Semiconductor and with Gor-
don Moore, Jean Hoerni, Faggin, and others made In
developing the silicon gate,
exquisitely accurate
method
work.
Faggin also contrived an
down
for laying
wires on the chip substrate and connecting tor so they aligned
it
these polysilicon
them
to the transis-
themselves in exactly the right place.
It
was
a self-aligned silicon gate.
Allowing the creation of transistors
forms of silicon
made
entirely of various
—sand—the Faggin breakthrough
of silicon gates
gave the technology at least a temporary reprieve from the tyranny of wires.
Most of the chip could be completed before you had
worry about the metal and
its
melting point. After Faggin, wires
and switches could be made of the same Although no one recognized the
first critical
it
essential material.
at the time, the silicon gate
step toward a totally
new
which
was
paradigm: a global net-
work of wires and switches dominantly wrought of silica,
to
are various permutations of sand.
The
silicon
and
substrate of
chips would be opaque silicon, and the network would be trans-
parent silicon (fiber-optic glass). Ultimately, nearly tion technology,
from networks and
cell
phones
all
to
informa-
computers
and cameras, would consist of various forms and aggregations of silicon gates.
The
Enabling
Silicon
Eye
75
this all-silicon infrastructure will
be
all-optical long-
distance systems, where the wires evanesce into mere wavelengths on fiber threads and the switches evolve into passive optical
waveguides
silicon gate
tion in
—
like prisms.
his first step for
This consummation of Faggin's
mankind
2005 and skeptics believe
company
it
—
will
is still
under construc-
never be
built,
though a
called Corvis has already run such a network for
its
subsidiary Broadwing. Jerome Wiesner and the other science advisers of President
John Kennedy, remember,
all
believed that
the Apollo project was impossible as well.
By sending the wings of
light,
signal
from origin
to destination entirely
on
the all-optical network obviates the electronic
conversions and regenerations, synchronizations and retransmissions that raise cost
Vastly simplifying
and complication
and accelerating
all
in existing networks.
communications, optics
can enhance the audio and video resolution of every
film, tele-
conference, medical image, game, or cultural event that plays across the Internet. Digital
mostly
computers and cameras and
silicon.
An
cell
phones are already
would be
electronic camera, in particular,
almost impossible with metal gates, since metal cannot detect colors of light. Eventually, silica fiber optics will replace
all
the
copper cages and metal harnesses of existing telephone systems,
and the network readily as
it
now
will
be silicon too and
will transmit pictures as
handles voices. Faggin's concept
will
be every-
where. With the wires and switches, computers and cameras,
made work
of the at the
same
all
essential substances, they might ultimately
same speeds, allowing the
distribution of
computing
and vision across an entire worldwide web of sand and
light.
Beyond the
Back
in 1969,
Gate
Silicon
while Neil Armstrong capered on the lunar
surface, silicon
cameras and
networks
all-optical
lay in
the long and still-embattled future. At that time, as
Robert Noyce and Gordon Moore to
form
Intel,
Hobbled by behind
left
Fairchild
Semiconductor
Federico Faggin faced more immediate problems. restrictions
at Fairchild,
management back
on
his visa, the
young
Italian stayed
while Sherman Fairchild and the company
in
Long
Island tried to figure out what to do
with their depleted Silicon Valley subsidiary. Faggin was not
at
the forefront of their plans. Meanwhile, the U.S. government did not
want the young
Italian
with his physics doctorate from
the University of Padua to take any jobs from Americans without
due process of bureaucracy. His experience with a stint with Olivetti at age nineteen, the former typewriter
company with
seen as a "small" computer (well, circuit boards,
deemed
heroically
11
it
in Italy
when he had
a design for
had only
few
had begun
a
provided
what was then
hundred printed
at the time).
Think how
George Gilder
78
many
jobs for low-income Americans that kind of mischief
could cost.
So Faggin had Valley
rumor
to stay
mill,
put at Fairchild. Through the Silicon
he heard that
Intel
was experimenting with
silicon gates. "If they are," said his boss at Fairchild, "we'll sue
them." At the time, Faggin did not
and was
startled
by
know about American
He
this idea.
technology, and Fairchild was letting
seemed, another key to depart
had
man would
wanted
just
it
drift.
to
lawyers
work on
Each
his
day, so
it
leave the company. Nearly last
was the young Hungarian chemist Andy Grove, who
at first resisted
this frothy talk of
someone from
moving
to Intel out of suspicions
compensation
in shares. Faggin
Intel didn't call him.
toward
all
wondered why
They could help him with
the visa, and he could help with silicon-gate processes.
He
still
And he would be more
than happy to
take shares. But he had to stay back. Bearing their
first child, his
knew many
wife, Alvia,
useful secrets.
became
restive.
They had few
Perhaps Federico should return to It
took some
six
months
friends in the Valley.
Italy.
work out the
to
visa details. Finally
Faggin called up Les Vadasz, his former Fairchild superior, then at Intel,
and asked
Vadasz thought for a a mysterious
project for a
was
a
to
come
to Intel to
moment and
new project.
It
work on
He had an idea for this new silicon-gate
said "Yes."
turned out that
complex device that
silicon gates.
Intel
Japanese calculator company,
had contracted
to create
named Busicom, and
that
could not be readily built without Faggin's intimate knowledge of polysilicon processes. Crucial
were the so-called buried
or "doped" silicon contacts that
made
layers
the connections from
Faggin's silicon wires to the transistors below.
The
The
Eye
Silicon
79
some 2,400
silicon-gate process enabled Intel to put
transistors
on a chip, enough
ing unit on a single sliver of silicon.
was the microprocessor,
an entire central process-
to build
A "computer-on-a-chip," this
a feat long targeted
by various visionar-
the industry but only vaguely outlined at the time and not
ies in
economically manufacturable without Faggin s technology.
The day
Vadasz hired Faggin to build the
after
Masatoshi Shima showed up
check out
Shima had lessly
at Intel
from Busicom
their microprocessor project.
persuaded
finally
item
which
at Intel,
Shima had signed slipped five
still
off
I
itself a
on the project
angrily
it
already see those.
circuits?
fancied
in
remained a low
priority
memory company. December,
Since
had already
it
months behind schedule.
As Shima put
There
came
is
all
when
Where
Faggin showed him the specs,
the logic design?
is
nothing here. This
way over here
is
the project since he had
and help him make
hand crippled
in a
left. it
Where
not a device.
to check.
check." Shima was furious that Intel had
ities
December 1969,
unorthodox one-chip concept. But by April 1970,
Intel's
arrived, the microprocessor
stay
Japan to
company, which had been clue-
his
when Shima
idea.
In
in
device,
seeking a calculator chip set with ten separate designs, to
accept
"I
new
There
is
made no
are the
It is
only
nothing to progress on
Faggin had to persuade Shima to
into a real microprocessor.
With
a
boyhood rocket experiment, Shima s capabil-
were doubted
in corporate circles in Japan,
intrigued by the opportunity to
come
to the
and he was
United States.
He
stayed for twenty years, until rehired by Intel to start a research
center at Tsukuba in Japan.
From
the silicon-gate concept later evolved a "floating gate."
George Gilder
80
memory
Crucial in the
power goes
when
chips that do not lose their contents
out, a floating gate
an insulated island within
is
the silicon device treated to accept electronic charge and hold
it.
Faggin inadvertently discovered floating gates for himself in the rendition of the microprocessor
first
when
the end of December 1970" and he could
Chips are made
number
in
came
"the big day test the
wafer
in
at last.
batches on wafers, each containing a large
of actual chips.
It
was
on that December
after 6 p.m.
evening after Christmas, and most of the other engineers had the day. Alone in the lab in Mountain View, Faggin was
left for
measuring the
electrical behavior of the device
loscope, the ubiquitous box with a
window
that provides a
oscil-
blue or green screen
for the engineer into the invisible elec-
He
on the chip.
tronic world
little
through an
attached the delicate probes of
the tester to a die (an unseparated chip on a wafer containing
many
dies).
As he wrote ing.
mean
I
scope.
I
I
pushed the
it's
bad
a
button and
.
.
.
noth-
die, let
me
another one:
7
try
Same
changed the wafer and repeated the process. Same
By now
I
was profusely sweating, thinking:
have screwed up so bad? wafers.
start test
nothing, not even a wiggle anywhere on the oscillo-
told myself,
story ... result.
later, "I
No
life.
I
continued
to test
After 20 minutes of agony
take a look under the microscope.
more I
How
dies
finally
could
I
and more
decided to
Something must be wrong!
It
didn't take long to find out that [Faggin's unorthodox] buried-
layer
masking step was
fab team]
.
.
.
Most
left
out of the process [by
Intel's
wafer
of the gates were floating!"
Insulated from any electrical connection, a floating gate
is
The
Silicon
Eye
81
useless as the control element in a transistor, but
useful as a
memory
cell that
would prove
could hold an electrical charge for
mere milliseconds of
a capacitor
but in years or even decades. Also in Mountain View
at the time,
a period
an
measured not
named Dov Frohman was independently
Israeli
floating
in the
gates
as
memory element
a
mable read only memories (EPROMs). 1971, Intel launched "a
announced four
new
perfecting
program-
erasable
for
When
on November
15,
era in integrated electronics,"
it
chips: for calculations, Faggin's microprocessor;
for software storage,
Frohman's
EPROM;
dynamic random-access memory; and
for
working storage, a
to interface
between the
processor and memory, a small register chip. In sum, Intel intro-
duced
all
the key components for a personal computer and
them used
as
it
EPROM, ories
of floating gates especially intrigued Faggin,
off.
who
an extension of his own silicon-gate concept. After the the floating gate enabled a series of nonvolatile
—memory chips
went
of
silicon gates.
The concept saw
all
that kept their contents
Based on the floating-gate concept
memory, holding programs and data on your
when is
cell
the
mempower
todays "flash"
phone, start-up
data in your computer, and images in your digital camera.
Faggin was a in 1986, still
some
vital catalyst for all
these developments.
fifteen years later,
he was
still
And yet,
in Silicon Valley,
building chips, and almost entirely unacknowledged by jour-
nalists
and historians
in the industry.
Only the experts knew
personal computers had Faggin's work inside.
that
The key reason
for
Faggin s eclipse was his departure from Intel in 1978 with his collaborator the "nobushi" (cantankerous) genius
Shima
to start
George Gilder
82
a
new company. At
when
a time
prowling the globe to protect erty it
and
to deter similarly
would not do
to let the
the silicon gate. So the
the company's lawyers were
Intel's
precious intellectual prop-
pregnant defections from Intel
itself,
world know the story of Faggin and
man who had
transferred the key intel-
and
lectual property of Fairchild to Intel in his spruce
Paduan pate was written out of
volatile
some kind
Intel histories as
of
embarrassment. Faggin had
many reasons
to leave. Intel
was too
large, too pre-
occupied with the memory market, too concentrated
owner-
in
ship in the founding group, too oriented toward a hagiography of
Noyce and Moore
to celebrate
of Faggin. In addition, that has
and reward the accomplishments
Andy Grove had
become infamous
at Intel. "If
instituted a sign-in sheet
you arrived
morning, you had to sign
after eight in the
in." Clearly,
time to go. Faggin was willing to sign in his (FF), provided
was a bridge too
As
was
it
far.
at the office
initials
on
it
was
silicon
a device of epochal importance, but this
He
felt
"subjugated."
a practical matter, though, even in Silicon Valley,
not just spin out and start a
new company without
banks and venture
capitalists.
opportunity. Intel
had yet
For Faggin, the
to focus fully
you can-
support from
vital factor
was the
on the microprocessor.
Faggin and Shima formed Zilog Corporation to do the microprocessor right! At Zilog, Faggin led the design of the Z-80, a
new
microprocessor, which was decisively superior to the Intel
processor and would enjoy a longer
life
span embedded in
low-rent electronics products. However, the central role of the
microprocessor in a computer architecture
is
to
define
its
The
Silicon
Eye
83
"instruction set," the portfolio of things that
it
can directly do,
such as add, subtract, or fetch from memory. Directly addressed by every
system (OS)
—such
as
computers software operating
Windows
or Linux
— these
instructions
vary subtly between different microprocessors, even very similar
ones
like the Intel
among
8088 and the Z-80. The subtle
instruction sets pass on as a kind of
DNA
variations
into the soft-
ware operating systems that run the basic operations of a computer,
and then from the
Word and PhotoShop
OS
into
that use that
all
the applications such as
OS.
Faggin set the Z-80 to execute Design Research tions, thus obviating the
need
to
pay Microsoft
DOS instruc-
fees.
the Z-80 was Tandy Radio Shack for the 'Trash 80" puter.
Adopting the 8088 from
Intel
was
IBM
Adopting
home com-
for its personal
computer. As a microcontroller embedded in everything from
automobiles to copiers, the Z-80 eventually sold in the billions for single-digit dollars. In the long run, however, going
the back of Bill Gates and his In the exalted realm of
DOS
behind
did not turn out to be wise.
computer central-processing
units
where
margins lurked near 50 percent and chips sold for hundreds of times more, Intel would crush the Z-80 entirely. Crush Faggin. In all
those benighted days before the
trial
lawyers banned
pithy or revealing corporate communications, before casual
e-mails
came burdened with seventeen
lines of disclaimers at
the bottom, and before the Justice Department focused
guns on various forms of managerial faux pas, term "crush" a tors.
Its
lot to
convey
its
Intel
goals in relation to
its
its
big
used that competi-
much-acclaimed campaign against Motorola's 68000
George Gilder
84
microprocessor was called Operation Crush. 1
crush
le liked to
rivals, too. In this
it
He would
was faggin who History
is
launched the
but to rewrite
privileges of greatness it.
processor would
ture of the chip.
based as puter, the
it
in Intel's version of
is
make
was on
man from
had conceived the general architec-
was not an exceptional architecture
at all,
Equipment Corporation 's minicom-
Digital
PDP-8, and not
history
inventor of the micro-
be Ted Hoff, the nice owlish
It
Ham-
a great company,
is
not only to
The company-certified
New York, who
Rochester,
really a
new
idea. Expressive of
what
PDP-8 devoted most
the industry called "spaghetti design," the its
not even get
but he and Shima
play,
written by the winners. Intel
and among the
of
ele-
alone famous.
became Rosencrantz and Guildenstern let.
Gates noticed.
slam dance of industrial
phants, Faggin was lucky to get out alive. rich, let
Bill
space to metal wires between the components on the
printed circuit board.
Put those wires on a modern micro-
and the thing would
processor, without Faggin's silicon gates,
sprawl over several blocks of Santa Clara.
Hoff was a smart and creative
man who
integrated solution for a calculator chip. didn't
proposed the obvious
He
stayed around and
make waves, and mused modestly about
recognition. But Intel
microprocessor
was not
company,
interested.
period,
and
It
He would win
well and he
went
had become the
more
supererogatory. Hoff had had his certified idea
man.
chips for speech
ideas
were
and he was
Intel's
the kudos, until Intel got bored with Hoff as
off to
become
a fellow at a research institute.
— The
However, the
feisty
In a Silicon Valley really a it
man
until
Silicon
Federico
you
85
— FF —was not the
macho moment, he
moved
had ended up with
"pushy" wife, so they said
"fellow" type.
observed: "You aren't
start a company.'' If
wisely in Padua and had
Clara, he
Eye
to jive
he had come it
money, and
who thought
should get credit for what he had done. Imagine Faggin
Faggin outside
—was the
wive
wealthily in Santa
relatively little
at Intel,
to
a
her husband that. Forget
Intel consensus.
Largely crush-proof, however, Faggin was difficult to forget.
And Alvia,
a technical writer,
was
grasp her husbands role in the
still
first
pushing
to
make
the world
microprocessor. She would
ultimately prevail, but the world of technology
was slow
in per-
sonal matters, and Intel-oriented. In 1986, Faggin thought he
might make history yet again, find yet another chance
for
fame
and fortune and enduring achievement, by founding yet another
company a
resolved to replace his
company poised
to take
first
invention with a
advantage of the
new
one,
latest innovations in
another computer model, the neural network.
8 Tobermory
many smart engineers the Federico FagLike deemed the prevailing microprocessor designs — even own — cumbersome, and too dependent in
industry,
gin his
on software
as
inflexible,
for interfaces to the outside world.
Based on the
architecture credited to the great Hungarian genius John von
Neumann,
all
these microprocessors followed a serial regime,
step-by-step, getting instructions
and data from memory, pro-
cessing the operands one at a time, and sending the results back
memory. "Step
to
n'
Fetchit"
was the
drill.
Von Neumann
machines were so dominant that any other design was termed
a
non-von.
Although von
Neumann computers were
efficient at per-
forming routine mathematical functions and simple step-wise algorithms for spreadsheets and word processors, they bogged
down on fail
to
a variety of
more complex problems. Not only did they
meet the famous supercomputer grand challenges of
chaos and nonlinearity, fluid flow and phase changes, weather
87
George Gilder
88
modeling and molecular chemistry, they also recognition, vision,
summed up
and haptics
as simulating
—
human
all
failed at pattern
functions that might be
sense and sensory input.
Daniel Hillis was the designer of several formidable "non-von"
"Thinking Machines'' employing a massively parallel design called cellular automata.
Neumann's
Although based on yet another of von
computers optimized
ideas,
automata
for cellular
were deemed "non-von" by an industry that could not quite terms with the polymathic genius of von Neumann.
come
to
Hillis
described the problem of von Neumann's original step-by-
common sense,'' or why computer science. Among humans,
step architecture as "the paradox of
sense
is
common
common
not
in
sense springs from
lots of
of the environment. In general, the ter they function.
and
see.
the
dumber
But
Neumann
in a it
knowledge about the
more humans
The more they know,
computer,
gets.
it is
the
learn, the bet-
more they can sense
the opposite.
The more knowledge
The more
that
is
it
functions. That
is
it
learns,
put into a von
computer, the bigger and more crowded
and the slower
details
its
memory
why your Google
search
engine runs not on one huge computer and one centralized data base but on one hundred and twelve thousand networked small
computers, using Pentiums, the great-grand-kin of Faggin's microprocessor design. Hillis
explained the problem in his book, The Connection
Machine: "As we build bigger machines with more equivalently, as area, the
we squeeze more
machines have
power and
silicon, or
transistors into
each unit of
memory
to processing
a larger ratio of
are consequently even less efficient. This inefficiency
The
remains no matter
how
fast
Silicon
we make
move
This time
89
the processor, because the
becomes dominated by the time
length of the computation
required to
Eye
data between processor and memory."
sump
was the
is
von
called the
overcome
it
Emerging
in the universities
Neumann
much computer
grail of
was
David McClelland, William
bottleneck. To
science of the day.
new movement
a
Chuck
Dally,
Seitz,
led by Hillis,
Steve Colley,
and many others who advocated removing the von Neumann bottleneck by multiplying the processors and coupling to
each one, distributed through the machine.
names, the concept was massively part on models of
human
memory
Known by many Based
parallel processing.
brain functions, the
would eschew the step-by-step regime and adopt
new
in
designs
parallel
meth-
ods that could simultaneously process scores or even thousands of bit-streams.
One
persistent line of research focused
on what were called
neural networks, which were computers based on a primitive idea of
human
how
billions of
neurons might work together
in the
brain. In the early 1980s, for all their popularity as
an
academic pursuit, neural networks remained an obscure backwater in the computer industry. In the enthusiast,
MIT
computer
scientist
league Seymour Papert, had
shown
late 1960s, a
one-time
Marvin Minsky, with
col-
that neural networks could
not perform logical operations critical for serious computing. At the time, the most resourceful advocate of neural nets was Frank
Rosenblatt of Cornell, inventor of the "perceptron," an early
form of neural network optimized Today, Rosenblatt's device,
for imaging.
named Tobermory
after a fictional
George Gilder
90
cat that could understand
Smithsonian Institution first
blatt
computer model,
in
as
human
Washington, next
if
losing a debate to
believed
it
was
to
He
von Neumann's
died in a somewhat myste-
on Chesapeake Bay
Minsky
in the
sits
comparable inventions. But Rosen-
never saw this vindication.
rious boating accident
conversations,
"I
apparently
most fervent
Rosenblatt's
a heinous crime.
after
disciples
accuse: Dr. Minsky on the hay
with an algorithm"
Minsky showed
that while any digital
computer can mimic
the series of additions and multiplications of a neural network, Rosenblatt's simple one-layer perceptrons could not perform key digital
computer functions, called Boolean
logic,
necessary for
the execution of digital computer programs. Since the digital
computer could do any calculation the Tobermory could do, but Tobermory could not perform key
digital functions,
others dismissed the utility of Rosenblatt's machine.
Minsky and As
a result,
the perceptron idea lay in limbo for two decades.
By the 1980s, however, the flaws led to a
new
Neumann
British genius
Machines (named
who conceived them
German Enigma
after
Alan Turing, the
during World
code).
A
digital
War
II
world.
But
digital
They were
num-
computers could not interact with the deaf,
dumb,
blind,
knows
that
real
and insensate. "Dumber
than an airport urinal," as Nicholas Negroponte puts urinal at least
while
computer might
simulate anything in the cosmos that can be reduced to bers.
design
look for alternatives. Digital computers might be
Universal Turing
breaking the
of the von
it,
since the
you are there. Computers may aim
to
be "general problem solvers" but they could never solve the prob-
The
lems of driving a
car,
Silicon
Eye
91
emulating a retina, or taking a photograph.
After decades laboring in the Artificial Intelligence Lab at MIT,
Minsky, himself, concluded that the most challenging form of intelligence
was not the manipulation of numbers, whether
in
high-school arithmetic or advanced calculus, but the perception of the real world
been
is
solved.
all
the seeing, hearing, touching, and motor
learned in the
skills that are
the input
—
first
year of
life.
Once
the image or
converted to numbers, the problem has basically It
becomes
SMOP
(simply a matter of program-
ming). Neural networks were designed to address these prob-
lems of pattern recognition that the
do
at
all.
That neural nets
machines seemed But
to
like
digital
computer could not
Tobermory might be slow Turing
irrelevant.
push neural nets
to the point that they could create
an
image or register a touch or recognize a face or take a picture
would require
a series of innovative jolts imparted by rivalrous
professors at Caltech.
The Hopfield Net
1980s revival
its
In
ral
at California Institute of
Technology, neu-
network research reached a moment of revelation
at
an
evening meeting of the American Science Association.
There
a
Caltech physics professor
duced the Hopfield
net, a
model
named John Hopfield
for a neural
network based on
an abstracted cartoon of the functions of neurons
in the brain.
These neurons had none of the features of wetware, no complex biochemistry. But the mathematical their behavior in electronic
and
terms
DNA or
rules that governed
their massive connectivity
— seemed
intro-
—
fan-in, fan-out
crudely congruent with the way
the brain functions in seeing, hearing, feeling, and recognizing patterns.
Comparing
his
computational model to the
digital
model, he
offered the analogy of two different approaches for a committee to
make
members
decisions. "In a digital computer-style
vote yes or no in sequence; each
committee the
member knows
about
only a few preceding votes and cannot change a vote once
93
it
is
George Gilder
94
cast. In contrast, in a collective-decision
committee the mem-
bers vote together and can express a range of opinions; the
know about
bers
opinions.
mem-
the other votes and can change their
all
The committee generates
a collective decision,
or
what might be called the sense of the meeting."
A
meeting that gravitates toward a consensus
meeting where even one buys
duce
a
more
stable
in to the decision
and robust outcome than
Japanese companies do not spin out a
that
lot of
a binary
up-down
to create devices
in the face of errors, bugs, or fuzzy inputs
Such
the real world.
likely to pro-
competitive compa-
Advocates of neural computing hoped
were robust
is
analog
Supposedly run by consensus,
vote from a digital meeting.
nies.
—
— an
a
from
computer could conceivably become
a
camera, or an eye.
What made tists,
Hopfield's vision irresistible to
and entrepreneurs
physicists,
was
alike
foundation in the natural physics of spin arrays
of magnetized glass
heated into an excited
could
on
it
tip a polarization
mathematical
its
These were
glass.
(silica)
brain scien-
that could
states
from neighboring
by the magnetic
grains.
Small effects
one way or another, triggering
of change to neighboring grains.
a
cascade
Governed by the laws of
action and lowest energy, the synthetic neural network gravitate as
much
ing a consensus
—
as possible to all
its
it
would
—which represented
solution to the targeted problem. Because a spin glass ural object, researchers thought
least
lowest energy state, signify-
argument subdued
teries of brains, eves, ears,
be
with each molecular grain "frus-
between two binary
trated," or poised
forces impinging
flux,
molecules
many
might offer clues
was
a nat-
to the
and other natural objects.
a
mys-
The
One way
Silicon
Eye
95
simulate such a network
to
to
is
interconnect
densely an array of simple processor elements, usually termed
neurons but better named "neurodes" to save the biological
name
for real neurons. Like cellular
the inputs of
all
automata, neurodes
contiguous processors. This
is
sum up
a "high fan-in and
fan-out" arrangement with lots of connections both in and out.
Responding
to the collection of inputs
that in turn
becomes an input
system over time arrives
at
—pattern
at
for neighboring processors, the
an optimal point: the solution. By
making the topology
grammable
with an output action
of energy peaks and basins
—
pro-
each processor through a pattern of weights on the
neurodes, the neural network designer can create a general-
purpose computer without a memory bottleneck.
Each configuration
of the entire
machine could represent
a
particular pattern of sounds in a speech recognizer, or of finan-
data in a securities analyzer, or of cranial shapes in a face
cial
recognizer.
In theory, then,
and
in practice
soon
after,
such
devices could identify fingerprints, buying opportunities, bacteria,
credit-card fraud, power-plant perils, or terrorist maneuvers.
The network may be inputs,
which
it
a desired pattern of
result in a particular weight for
pattern of weights
mon
trained by showing
is
each neurode. The
the program for the device.
Thus the "com-
sense" of the machine was not concentrated in
tralized
memory but
some cen-
diffused through the entire network, like
the information in a hologram.
Faggin had been tracking
oped first
his
these developments as he devel-
new company, which he would
insight
become
all
was
that a
a preferred
call
Synaptics. His
programmable neural network could
form of processor
for all
pattern-matching
George Gilder
96
computer
functions, such as hearing, imaging, touch, and other inputs. Since every
spend
much
of
its
computer or
cell
phone
or
camera would
time recognizing patterns of various kinds, a
general-purpose pattern recognizer could find a market that
might be comparable microprocessors
in size to the
like the
market
von
for
one Faggin had created
Neuman
at Intel
some
fifteen years before.
Faggin's further insight
function as the
was that
memory element
his "floating gates"
for the
could
neurode weights.
He
envisaged a programmable neural network based on floating gates that could be trained to perform any pattern-recognizing role.
"Guess, measure the
back"
error, adjust
the answer, and feed
the basic routine of neural networks:
is
an
it
iterative
process that step-by-step approaches the answer to problems,
such
as: "Is that
object in the
Bin Laden's face or a bagel?" or "What
Tumi bag?"
that cannot be
is
that
computed by ordinary
machines.
Once
again, as in the case of the original microprocessor,
intellectuals thought they
knew what
a neural
network was, but
no one could build an effective and marketable device. Like
most researchers who explored the
field,
Faggin found himself
gravitating toward the
work of Mead, then building
scale neural networks
on microchips. Even though
his largehis neural
networks were only roughly related to the models of Hopfield
and McClelland and the other pioneers who had attracted
Faggin,
Mead was
originally
pursuing a quest for practical
devices rather than for exalted but elusive theory.
He was
begin-
ning to transform his Caltech lab from a center of leading-edge
The
digital devices
and design
Silicon
tools into
Eye
97
an arena of what he called
Analog VLSI.
Mead and VLSI chip sors,
his
students began converting ordinary digital
substrates such as are used in Faggin's microproces-
made by
the billions in semiconductor factories around the
globe, into analog circuits that could crudely reproduce brain
functions. Because he
was using the most common semicon-
ductor processes, his devices could be cheap, containing thousands of transistors. Because he used these transistors in analog
form
—not switching on and
ing continuously like an
analog sensor with as
off like a digital device
AM
many
radio tuner
—he
but operat-
could
make an
transistors as a digital microproces-
sor has, eventually millions.
company, Faggin would have
To tap these resources to
come
to
for his
new
terms with Carverland.
Part
J
hree
THE CAT AND THE
CAMERA
—
10 Carverland
Beginning
in 1983, the center of
his students called
above the
janitorial
it
Carver
Carverland
Meads
—was
world
his laboratory
warehouse and machine shop
at Cal-
tech. Originally classed as "second-class space," this secondstory facility filled
became
up with equipment and avid students and
a hive of creative activity
glimpse of some of the players,
network
at
LAX and
weight
its
let
on the campus. To get
us set up a pretend neural
synapses for
school males headed for Pasadena for the the synapses for acne, ponytails,
ham
brilliant
fall
nerdy high-
semester. Jack up
radios, plastic pocket pro-
tectors, model-airplane kits, copies of Godel, Escher, Bach,
advanced graphic calculators. (Let that big glasses
and denim
overalls
a
girl
and
with granny
and the comic books under her arm
pass right through, shunted by the neural network into the
art-
school queue.)
One way or another, the rest
—
the synaptic machine might catch
many of
including, for the purposes of this tale, David Gillespie,
IOI
George Gilder
I0 2
[obi
I
Tim
)elbriick,
Allen,
Adam
Greenblatt,
Gribble, David Feinstein, John Piatt, ipants fit
Tom
and even
a
Tucker, Glenn
few older
partic-
— instructors — such as Dick Lyon, who wore a ponytail and
the mold. In 1984, Lyon had joined the Schlumberger Artificial
Intelligence
Lab
in
Mountain View
From these perches regularly eration,
down
to
after his stint at
in Silicon Valley,
Xerox PARC.
he continued
to
commute
Carverland in Pasadena to guide each
work with Carver on
new
gen-
problems, and
difficult technical
tutor exceptional students.
Now that our neural
net has caught them,
imens. Near the entrance to 1
980s was
a fuzzy-faced
he was introduced ("This shift
screen.
on
He
is
his feet, look
really
had
meet our spec-
lab throughout
most of the
blond boy also with a ponytail, leaning
toward the large screen of an
and
Mead s
let's
to get
HP Chipmunk workstation. When Dave
Gillespie"),
down and back
he would shuffle
away, and back at the
to his project.
A
Caltech soph-
omore, he was ginning up a design tool for timing parameters on digital circuits that
Carver considered "pretty neat." And "Carver
had the coolest computers," so the young undergrad began
hang around the "I
managed
to
He ended up
Soon he was there day and
lab.
horn in on
all
night.
the coolest professors," he says.
as a teaching assistant in
course, "The Physics of Computation."
He
tech to victory in national Association of
(ACM) programming
to
contests and then
Richard Feynman's also twice led Cal-
Computing Machines
became coach. Taking
the Hopfield course in neural nets, he would not only do the
homework but homework
also
code computer programs
automatically.
to
perform the
Even before graduating, he
started
The
actually
Eye
freshman course
the
giving
Silicon
103
computer science.
in
"Except for Carver, most of the top professors steered clear of freshmen.
I
found them
great.
I
liked turning
them
on.
I
liked
explaining the stuff."
Within an important group of other scholars such as Glenn Gribble, Telle Whitney, and John Lazzaro,
who
supplied the lab
with design tools and software, Gillespie became a fixture in Carverland. In an extraordinary feat, he contrived to spend a
decade
at Caltech,
including six years in graduate school.
It
full
was
not until 1991 that he turned off onto Interstate Route 5 and drove up the coast to Zanker Road in Santa Clara, where he
much
joined another lab that looked pretty
same
faces
the same, with the
and much of the same equipment. The company was
Synaptics, almost five years old at the time, the vessel of the imager project that
would end
Often working with David Gillespie 1980s was Tobi Delbriick. Max's son, a
unassuming and engineer in the
boards
(test
practical, Tobi lab,
as the
corporate
Foveon X3.
at Carver's lab in the
large,
shambling man,
was perhaps the most adept
ingenious at contriving workable bread-
boards with an array of working electronic devices
on them), incorporating the various by
first
Mead and
the other students.
and devices created
circuits
He
frequently collaborated on
imagers with colleagues. Later contriving a stream of retina
—
chips himself ers,
for
camera imagers, door openers, check read-
and autofocus detectors
at Synaptics.
—he
built
many
of the imagers used
Nonetheless, industrial engineering was never his
true calling. Like his laureate father, he
research on the frontiers of science.
was called
to basic
George Gilder
104
Sharing an office next to the Caltech lab were John Piatt and
David Feinstein. Even
Caltech, where "beautiful minds"
at
abound, these paragons of mathematics were a rare and coveted resource. Although Piatt his services
was Mead s graduate advisee, he shared
with John Hopfield; Feinstein worked both with
Hopfield and with Richard sis
Feynman while
writing a Ph.D. the-
with Mead.
With
curly black hair
twenty years teen.
He
later,
a boyish look that
Piatt
had not
left
studies.
When
was
in
log project
incident
and came
in three years,
he arrived
at
cramped quarters was
light.
to
in
in
ful collaboration
with
low power
to the
new
machine shop
Mead on
—
for graduate
to
be run on
Carverland
Piatt
began
— the
a fruit-
the use of timing dynamics in
neural models, delays and buffers that give
counterpart of short-term
Long
Booth Hall, where the only ana-
1983
larger space over the janitor s
Pasadena
at
age eighteen in 1982, Carver's lab
a circuit of sufficiently
Moving
him
exhausted high school by age four-
passed through the computer science program
Beach State
still
John
and
memory
them the analog
in digital systems.
Mead was
sure that crucial to the brains computational process
was the
use of time. Rather than marching to the beat of an abstract clock, the brain uses the time of arrival of an input as a source
of information
mode
and
its
relationship to other timing events as a
of computation. For example, the brain apparently distin-
guishes a word from a chaotic flow of sounds by the appearance of several
phonemes enjambed back-to-back.
Presenting a prototype system to Carver, Piatt in his
first
year
declared that his scheme was asynchronous (without a regular
The
Silicon
clock cycle) and thus would be
Mead
speed of components. matter
if
you take out
Eye
105
immune
to variations in the
challenged him. "Then
he
this capacitor?"
shouldn't
it
said, pointing to a small
circular device that stored or delayed electrical charge
on the
prototype board. Piatt
hedged. "Well, maybe." Piatt had a mathematical theory
but had not tested
it.
He
took a deep breath and yanked the
capacitor off the board while the circuit was running.
ued
to
It
contin-
work, meaning that the circuit was truly asynchronous
and thus did not depend on
a particular
cadence of timing and
delays.
Carver burst into a smile and slapped him on the back. "That's terrific!"
he
said, delighted as usual
achievement by his students. a
mode
of
computing eventually led
and Dick Lyon
human
Piatt's
in the creation of
work on the use to collaboration
to the
of time as
with
Mead
an ingenious model of the
cochlea, which imitated the
sound frequencies according
with every show of
human
ear by separating
time they take to pass
the auditory channel. After getting his Ph.D. under
down
Mead
in
1985, Piatt joined Lyon as an intern at Schlumberger. In pure intellectual fireworks, though, Piatt his
match
stein
in his office
more than met
mate David Feinstein. Unlike
was not awesomely precocious. Until age
eight,
beat his electrical engineer father in chess for the
was deemed dyslexic and otherwise
slow.
Piatt, Fein-
first
when he time, he
But he got over
it.
Described by the department head as Caltech's best student in applied mathematics in thirty years, Feinstein talks so fast that
Mead
coined the "Feinstein" as a unit of speech pro-
George Gilder
106
duction and described
other students as speaking in "milli-
all
down
Feinsteins." Unfortunately, "Feinsteins" slow
when
crawl
translated into writing,
thesis for several years.
minated the
When
and he
failed to finish his
came, however,
it
to a viscous
it
brilliantly illu-
between Claude Shannon's information theory
links
and thermodynamics. Although above
all
deeply philosophical.
else a mathematician, Feinstein
He
communications
assistant
He became
theory.
and go-between
wave dynamics
more recherche enigmas
the nature of the soul, and the or
for the course
putation" jointly taught by Hopfield,
to
make
useful
it
to find their
who had come ing
"The Physics of Com-
who
failed read-
hung around
trying to be
it
One
in.
of these
was Tim Allen,
Caltech chiefly under the influence of watch-
to
Feynman on
way
of quan-
Mead, and Feynman.
into the inner circle, but
enough
to
the lead teaching
Also influential in Carverland were students ily
also
can commute readily among such sub-
jects as Jungian psychiatry, the relationship of
tum
is
a tape of a
Nova program.
scores of times," he says.
"He
didn't
"I
must have played
have particularly more
screen time than any of the others, but there was something
about the others,
man
I
—he was more
just liked
and hearing him
talk
exciting than any of the
about things made
me
excited
about them, too." Allen's
most profound early encounter with technology was
a
Hewlett-Packard HP-35 scientific calculator, which his physician dad brought
was old enough captivated
me
home from
the office
when Tim was
to appreciate that arithmetic
to
hold something in
was
nine.
difficult
my hand which was
and
"I it
'smarter
The
than
I
Adept
was."'
at
the calculator "the
Eye
Silicon
107
mechanical technology as a boy, he found
first
thing
I
had ever encountered whose
inner workings were totally impenetrable, invisible, inscrutable. Just a at
lump
of plastic,
it
didn't
Caltech nearly a decade
relevance to this machine.
35 worked, In his
"all
the
way
whir or
later,
He
click."
When
he arrived
he chose his courses for their
set out to figure out
how
the
HP-
to its physical underpinnings."
sophomore year he encountered Carver Mead teaching
CS10. Mead opened
the introductory computer science class, his first lecture
with an analogy between the
a car: "Imagine trying to describe
how a
computer and
car works to a technically
savvy alien who'd never seen one before. believe that an engine that
digital
It
would be
worked by explosions
difficult to
rattling
cams up
and down could ever produce the outward appearance of smooth quiet motion. Digital computers were the itive internal action,
fluid function." Allen
same
its
frenetic, prim-
well-hidden, causing an illusion of smooth
concluded, "Carver was articulate, charis-
matic, and informal, and struck exactly the that videotaped
—
image of Feynman.
.
.
.
same chord
in
me
as
Caltech had delivered on
promise."
Like his friend Dave Gillespie, Allen was attracted to Carver's
Chipmunk
lab of
dained the
little
HP9836
pre-PCs. Real computer "geeks" dis-
Chipmunks, preferring
centralized near-"mainframe" Digital
computer that the
one
all
center.
HP was
time-share a large
Equipment "VAX"
in the
But Allen and Gillespie quickly discovered
half as powerful as a
to himself,
to
VAX and
each student had
while the VAXes had to be shared with
other students. Setting up the
Chipmunk
lab,
fifty
Mead showed
his
George Gilder
108
grasp of the promise of the personal computer back in a period
when
it
fast that
seemed
a radical idea.
This technology has moved so
today Allen with his team at Altera Corporation, a major
Silicon Valley innovator, has built a microprocessor unit, given
away in a
as a promotion, that has the
power of 150 VAXes and
fits
corner of one of Altera s chips.
Like Gillespie after him, .Allen
CS-10. But he
still felt
more
looking
elite lab,
that he
became
a teaching assistant in
was on the outside of Carver's
In his senior year, he signed
in.
Meads "Analog VLSI and Neural Systems" largely taught recalls,
"It
from
his
forthcoming book by that
for
which was
title.
As Allen
was new, important, leading-edge glamorous fun,
—
interesting
class,
up
everything.
I
experienced the feeling of being in
exactly the right place at the right time to catch the perfect wave."' cle.
But he
To do
still
that,
had not made
he had
for a silicon retina that
to build a
his
into Carver's inner cir-
custom piece of
was of urgent
step on the road to Foveon.
way
lab
equipment
interest to Carver
and
a key
Misha Mahowald gazing
into Tobi Delbriick's camera, taking a break
from a research conference they attended
in 1993. Tobi Delbriick
11 Michelle
Into
Mead's band of
digital brothers
1983 strode a woman. Hey, no big
woman
to play
one day
her,
December
She was not the
deal.
an important role in the
such as Telle Whitney before
in
lab.
first
But the others,
and Mary Anne Maher, had
been adaptive, supportive, working within the frame, conceiving
new
valuable
woman, Michelle Mahowald,
contrast, this transistor
design tools for advanced transistors. By
digital
was
"a
arrived thinking a
kind of radio." She really did.
And
radios
were
not her thing.
Michelle was a different kind of
some
said a "bionic
She was shy what
to
do
.
woman," interested
at first, .
.
—an analog woman,
woman
get this, gardening.
in,
but she rather soon began telling Carver
and, for nearly ten years
finish her thesis nor leave the laboratory.
.
.
.
she would neither
She would decorate
its
walls with artsy-dipsy posters, concoct chip designs with an antic hexagonal flare, post lab reports with bright overlapped
patches of color, mauves and yellows galore, and release random
George Gilder
in
analog beasts of prey from their safe digital cages. After her passage, the lab
would never be the same.
Her landing
in the lab
was unplanned and unexpected. After
the usual prodigies in academics and college boards obligatory at
Caltech admissions, she had arrived on the campus three years before from an Full of cautions
all-girl
Catholic High School in Minneapolis.
from her intellectual mother on the
"Californian lifestyle," Michelle
was
perils of the
set to study astrophysics
and become an astronaut. She wanted
to
walk
in space, tran-
scend her past. She even wanted to change her name, transform her mind.
"I
wanted
to burst out of
my
shell'
"
—her Michelle
Catholic constraints, the concept of original "sin," her structured family, their corseted beliefs, their rules
the parents
who had adopted
She sought
rosaries.
She loved
and named her
her, raised her,
become "Misha,"
Michelle, but she wanted to idyllic, free as a cat in a
and
bionic, lyrical,
commune. wonder
to recapture the
of a trip to Disneyland at
the age of seven, on a ride recalling the Alice of Lewis Carroll or the "incredible shrinking Slick's
Disney
man"
of film, or perhaps even Grace
"White Rabbit" pill-borne descent of the 1960s. In the ride,
she remembered dwindling
enter a water molecule, as
dance. With
little
good, flushed
down
into
a
size of a basketball.
Brownian
froze into a dazzling crystal
Disneyland netherworld,
maybe
face-to-face with a nucleus
it
in a
and she thought she was gone
a stream,
microcellular catacombs,
small enough to
bounced around
it
Michelle inside,
and then melted into
down
—
forever.
lost
for in
Then she came up
a bright red blinking sphere the
Glancing above her she became aware of
a
The
gigantic eye, a
huge pulsing
a microscope,
seeming
to
was half giving up hope,
Eye
Silicon
retina, looking
down
half hoping for the ride to last forever,
into the nucleus?" "Certainly
I
she began to grow back again
and aggregating the atoms
into
"Do
And
dare not." .
.
could
.
all
dare to go further,
I
it
then astonishingly,
be
.
.
.
recomposing
the higher levels of structure,
and blood and bodies, and restoring her
cells
her through
at
peer into her mind. Just as Michelle
the spectral voice of her guide asked,
in a
113
to her original size
conventional family group. There was Papa Al smiling, and
embracing
her,
and Mother Joan,
cerned as usual, and
intelligently, judiciously
sister Sheila, all as before.
the trip into the molecular
her world. As she recalled,
Her
family.
microcosm had permanently
"I
thought
it all
was
real.
I
Because you
that
I
had been led
were normally
found that
it
to believe.
invisible to
was supremely
me
much more
a
There were
that
exciting."
imagined that an astronautical
more
Mahowalds embrace it
space would bring a
link with distant lands over
ham
a
mere
abstraction.
and rebuild
model planes and radios.
five,
televi-
rockets, or
She painted
vivid pic-
tures, with rich colors, consciously suggestive of Picasso.
age of
still
Unlike her classmates or her
associates in the lab, she did not disassemble sion sets as a child, or construct
I
and beauty.
was never
as a reunion with nature.
these things
As the years passed, she
trip into
of science
all
interesting
were actually there and
ecstatic high of natural revelation
She saw
down
never thought to doubt."
She concluded: "The world was place than
But
altered
could see the other people on the ride moving in miniature a tiny plastic pipe,
con-
At the
she saw patterns in nature. Pointing happily, she told
George Gilder
n4
The wind blows through
her father: "Look.
them over
the grass and knocks
dominoes." She helped plant a maple tree
like
in the
backyard and dreamed of having a garden of her own. She hid in her room and gorged on books of fantasy Rings, then
Hermann
Her teachers and avid artist
loped
it
Tolkien's Lord of the
Hesse's Steppenwolf.
friends from
Minnesota remember her
and an able gymnast, part of
third in the state.
moment
—
One
in kickball.
and
it
a girls'
team
as an
that
was
of her eighth-grade teachers recalls a high
"She just stepped up
was gone.
When
I
to the plate
and wal-
think of Michelle," she said,
"I
see her standing there, hands together, great big eyes and a shy
smile and the kickball zooming out of the park."
remembers her dancers
as the
at a local
Her mother
most intense and perfectionist of the tap
dance studio. Afraid
at first to
go on stage,
nervous and evasive, she emerged once there as a dervish of a dancer. In the audience, they
Joan,
beamed
Everyone
all
noticed her, and her mother,
in pride.
recalls a
happy and successful
more than expected. But Misha remembers
ment
in
always giving
a sense of confine-
her pleasant suburban household and her structured
Catholic schools. As a child, door, but
girl,
I
"I
could not leave through the front
could read books and take that way out." She remem-
bers that at school the teachers tended to devote most of their attention to the slower students, so she again retreated to books
and then
was neglected, and
to college extension courses
on
a
campus near Minneapolis. Misha never became a type
can be defined
a typical Caltech science
at all
on probably the
wonk,
nation's
if
such
most
truly
The
diverse
study physics, she already
By then the
Eye
On
and meritocratic campus.
from Minneapolis, headed
ence.
Silicon
115
the plane west to
September 1980
for college in
felt
LAX
qualms about the usual
life
to
of sci-
brunette with the theatrical smile and
little
the already fierce intelligence had grown into a big, willowy,
owlish
girl
—what the French
conveying a para-
call "jolie laide"
dox of coarse features and luminous femininity.
With granny too,
and
glasses,
broad shoulders, a secret forbidden
a taste for sixties rock
and
roll,
plane on her way to Caltech with a pile of
comics and some colored pencils, given
she comes onto the
Conan
to her
She finds herself
sitting next to
bound
campus, flaunting
a
and a
copy of
the Barbarian
by a rakish
friend Diana. for the
tat-
artist
another youth
Scientific
American
pile of scientific texts. Sixteen years old, too "nerdy" to
believe, "he
made me
feel old
and
I
wasn't even eighteen." Hor-
monal, aggressive, splaying peacock plumes of mathematics and
away
physics, he chattered proto-professor.
Why
can't
in the adjoining seat like a
he keep
runaway
his voltage fields, his calcs,
megahertz, his knees, his vectorials, his Hamiltonians to
his
himself?
Why all
way
at
this
mental molestation?
Is this
going to be the
Caltech? As on several other occasions on her
pil-
grimage into Southern California's most outre circles of high
sci-
it
is
ence, she considered turning back. In those
moments, she would cherish her memory of Min-
nesota, the quietness of
it,
the peace. "The seasons were pro-
nounced and you have an acute sense of time early-summer
day, the
passing. In an
sun would shine through the elm leaves
bowering the paths along the water, and the
light
would shim-
George Gilder
n(.
mer and
crystallize the scene,
and grass growing
trees
short time to just three
"Then out a
lot
live. It
and
it
was
as
if
you could see the
in real time, as if life there
has to get
it all
way
out of the
has a very
in a rush of
months. in the
and
of diamonds.
winter
ice
it is
cold but beautiful because the sun
grows on the branches and the world
The sense
is
made
is
and yet
of seasons, of things changing
staying the same, isn't there so strongly in California.''
came new
at caltech
of Disneyland or the
shocks and changes beyond the ken
all-girl
Catholic High School. Perhaps only
her mother really had the picture. Assigned to the Ricketts coed dormitory, she encountered "the sexual dance."
Palmy Pasadena
resembled something out of Dr. Seuss "with tufted palms trees everywhere and strange gawky animal adise that looked as
if
they might eat you
In a college that grants
up
in "his or
sion, in
her
own
freedom galore
way,"
meaning
an institution that did not
and Flowers of
life if
you strayed too
for
each person
virtually
officially
Par-
near."
to
grow
no adult supervi-
recognize the differ-
ences between the sexes, the heat built up.
Misha found herself under
who resembled and
visible
a two-year siege by
male creatures
her airplane companion, with prognathous math
cutaneous
oils.
Retreating to literature as was her
wont, she was taken with the theories of The
Territorial
Impera-
Robert Ardrey and The Naked Aye of
Desmond
Morris,
tive of
two
texts that
expound evidence
for the theory that
most human
behavior echoes the ethology of monkeys and baboons.
"I
was
The
Eye
Silicon
struck by the hypocrisy of the boys,
ested in science, in ing rites for
to
be
inter-
were animals perform-
like goats
seemed her best guide
butting heads." For
to life at Caltech.
and maneuvering grew intense. They "glommed"
jiggling
onto her and wouldn't
sum
spend more time with
me
All that platonic annealing
four hours in a day and
wouldn't give up.
The
"Glomming," so they
let go.
Caltech tradition." The to
who pretended
ideas, but in fact
my benefit. They were
a while, zoology
The
my
117
1
of
was
it
than
began
that "the guys
wanted
I
said,
to
all
was
"a
wanted
spend with them."
to get physical.
50 male nerds on the
Only twenty-
floor.
These guys
neural program was snagged on a high
energy plateau.
The
sex ratio
suffices to
For
all
war
"my
shell."
bit
much
She was
for
one
Misha. She began to retreat
a big girl with
but "there wasn't near enough of
me
my room with
Often she would
the lights
off,
flee Ricketts to
an expansive heart,
for all of
learn to slam the door in their faces, to
hide in
theory, seven to
her brains and heft, physical and intellec-
seven to one was a
again into
to one. In
overcome almost any armed and entrenched defen-
sive position. tual,
was seven
them.
scream
so no one
I
had
to
them, or to
at
knew was I
there."
Dabney, a nearby dorm dom-
inated by the hippy counterculture flourishing at Caltech in a
delayed throwback to the
when
sixties.
There she met Tim Allen, who
not in Carverland served as a leader in student "govern-
— ment"
"I
was
like
Ralph
in
Lord of the
Flies," as
him, Misha protested against the behavior of her don't think she really trusted today. "I
was focused on
me
or liked
me
he puts
dorm mates.
at first," says
electrical engineering,
it.
To "I
Allen
making things
George Gilder
us
work, and she wasn't. She was focused on learning about the brain."
But the
Allen says,
was
"It
for everyone.
real
The guys were not
sad.
wrong. There were just too the
same thing and the
easy
way
were caught
best friend,
and
As
doing anything
they were
in the
all
doing
middle with no
out."
became entranced with
Visiting Dabney, though, she
late
really
many of them and
girls
vision.
an awkward situation
really a case of no-fault,
was
It
problem was Caltech's unisex
Tom
artistic,
Tucker.
A
Allen's
glowing, charismatic figure, articu-
one of the most obviously gifted of
his class at
Caltech, Tucker was deep into everything that Misha's mother feared in "California lifestyles."
With Tucker, Misha discovered
some
and the downside of zoological
of the upside of zoology
males. Tucker turned out to fascinate
many
of that years girls at
Caltech. But Misha got more than her share. Together they took
her Disneyland hallucinations from physics onto chemistry and
pharmacology. sor
From Tucker, she
also heard reports of a profes-
named Carver Mead, whose course Tucker was
Allen,
taking. Like
Tucker was an enthusiast of Mead's adventures
in analog
VLSI and neuroscience. Misha
still
was pursuing physics. But the astrophysics was
not going so well.
Misha had been
a
math prodigy back
nesota, taking college courses as early as eighth grade.
read with exhilaration about
nearby Palomar, with kind of Capitol islate for
its
Dome
the universe.
Caltech's
giant
in
Min-
She had
observatory at
hyperbolic 200-inch Hale telescope, a
for the university's physicists as they leg-
When
her bus into Pasadena drove past
the service entrance to the Caltech Synchrotron Accelerator,
The
I
learn
what
This
is it.
is
where
119
it
really
speed of
light,
happens. This
she
where
is
really true."
is
She had always found math
easy,
prevail in the physical sciences.
But
math prodigy and she discovered est
Eye
particle collisions at the
where they contrive thought, "This
Silicon
and she saw at Caltech,
that her
math
as a
it
way
to
everyone was a gifts
were mod-
by comparison to many of her classmates, more than half of
whom
were physics majors. She became uneasy with the
assumptions they so readily upheld, the idea that
was a pyramid,
The
on
resting
all
of science
a substrate of physics at the bottom.
physicists implied that everything above
was somehow
epiphenomenal, that they were merely awaiting some more powerful digital
machines
come and subdue
to
the derivative disci-
plines to the underlying rule of physics.
Plunging into astrophysics, she discovered that the course that she
was taking offered
were scant
—but
authority, they
the total mass in the universe
and
forth, like
tune or
like
infinite
physics were
men
was
tractionary, issuing
an
and equations. Hard
made her
seemed
in
to
nervous. Estimates of
to swell
some remote
losing her faith.
astrophysical ride
into
reports from Betelgeuse
and
molecule responding
a Disneyland
was
—
shrink,
back
an accordion playing an inscrutable modernist
agenda contrived by adult, she
data
bristled with theories
on
follow, resting
little
to a
hidden
control center.
She could not
tell
As an
whether the
real or not. Inflationary, oscillatory, con-
from an ineffable singularity or multiplying
number
of parallel paths,
it
seemed
as
if
astro-
away
into space
from the supposedly solid foundations of the physics
at the bot-
all
math,
all
the time, flying far
George Gilder
120
torn.
She had wanted
like a
be an astronaut. Her head was spinning
to
nerdy propeller cap, but she was farther than ever from
orbit.
In her doubts, she
Caltech's physics.
Church. the
life
seemed
on earth
him
to
much
as the Catholic
emanate from males." Even
of the space program, a symbol designed to repreto all other galaxies,
forthrightly forward,
ing at
to notice the patriarchal tradition of
resembled nothing so
It
"All authority
emblem
sent
began
showed
man
a
engaged with the scene, and a
Was
reverently from the side.
looking
woman
gaz-
the universe itself a
male playground? In Caltech physics, tory biology course bristling
seemed
it
more
to be.
to her taste.
She found an introduc-
Rather than scarce data,
masses of equations, and mazes of theory, biology
offered reams of data, find patterns in
it.
measurements
She was an
galore,
artist,
found herself gravitating toward the
and
adept
need
a sore
to
She
at patterns.
living sciences. Science,
she
concluded, should not be a hierarchy or a pyramid but "an organ-
roscience,
and changing." Taking a course
in
neu-
she became deeply intrigued by the retina,
the
ism, continually growing
photons
sensitive thin outer layer of cells in the eye that receives
and transduces them into electrons of the brain. In
many ways
the crypts of the mind.
It
it
for passage to
seemed
was
to offer a
deeper layers
path of
light into
a regular structure, well
with accessible chemistry and self-evident functions.
It
could be
understood. She plunged deep into an effort to master
through
it
to
fathom the
known,
it,
and
brain.
In this pursuit, at the beginning of her junior year, she signed
The
up
for a course in
begun by Caltech
new
to achieve a biology, field
a
sciences.
121
A famous course,
Feynman, and Mead,
hoped
it
multidisciplinary synthesis of physics, neurojust
what Misha was looking
who had turned
for.
Hop-
to biology.
engineer and physicist with an interest in the brain
Feynman was
But early
tation.
titans Hopfield,
mathematical physicist
Mead was an
Eye
computational neurophysics.
and engineering,
was
Silicon
a legend exploring the limits of
compu-
he was stricken with cancer and
in Misha's year,
forced to withdraw from the course. That
left
the neurobiology and neural networks, and
Hopfield pursuing
Mead expounding on
the potential of very large scale integrated circuits for simulating brain functions and performing neural-network-like collective
computations. In the lectures, ing the
first
Misha found Hopfield's ideas
plausible path into a
inspiring, offer-
model of brain functions
—how
a simple collection of unintelligent neurons could achieve a
form of collective intelligence from the bottom up, with no central control.
But she was mostly baffled by Mead's dense expla-
nations of transistor physics.
"I
couldn't understand
she recalls. Although she demurred at tions of Hopfield's neural
project
some
him
at all,"
of the simplifica-
network model, she decided
to
do her
on the retina with Hopfield.
She went
to
meet with the teaching
David Feinstein. Shared and coveted by for his extraordinary gifts in
assistant in the course, all
the three professors
mathematical physics, Feinstein had
become an emissary among them.
Earlier
he had reproached
them
for asking their students to integrate the three disciplines
when
the three professors themselves
seemed
to
have
little
idea
George Gilder
122
how
Then Misha came
the fields could be joined.
paper she had written on the drawings that conveyed
its
the Millikan library, hiding
human
him with
to
a
retina, replete with large
operations.
She had spent weeks
in
away from the Ricketts scene, work-
ing out the detail and limning out the supporting
art,
showing
all
the biological links and signal pathways, horizontal and vertical.
Her model encompassed the
entire neural flow of light
and
energy, from the retinal cones that capture the colors in the eye
and the foveal receptors that focus on you are reading, through the
ters
communication
into the brain
fine detail
such
as the let-
triad synapses that begin the
and bipolar and amacrine
cells in
the "plexiform" layers that do vital balancing of energy and
dynamics down tex,
which
to the retinal ganglion paths leading to the cor-
integrates the
"fan-in" to 'Tan-out, "
it
image into a recognizable whole. From
was
misha was shy about
a coherent imaging system.
her work, but to Feinstein
extraordinary vindication of the course. field
and Mead,
Nothing
in
it
biological science
seemed
could be reproduced.
seemed an
To the physicists Hop-
offered any kind of agenda by
logical function
it
largely guesswork.
which the physio-
Mahowald was
a biologist
by training, but she had offered a model of the retina that was least suggestive of a
schematic diagram of a chip ... a chip that
could become a working model of a kind of camera.
at
mammalian eye
or a
new
Misha Mahowald shows Carver Mead a feature of the that she built in
teacher
and ended
Tobi Delbriick
Carver Mead's
laboratory.
as her collaborator
"silicon eye"
Mead began
as Misha's
on the frontiers of brain
science.
12 Misha
When
David Feinstein took Misha
seemed
Hopfield, the prospects
meet with John
to
favorable for an
enduring student-teacher link between the gray-haired professor and the ambitious
Hopfield tive
who had most
excited
young
Misha with
computation by individually
dumb
was Hopfield whose magnetized "spin
his
biologist. It
model of
neurons
glass"
tall,
was
collec-
in the brain.
It
concept seemed
to
offer the possibility of a physical foundation for neuroscience.
But Hopfield's neural networks were radically simpler than
Mahowald's concept. Crucially, the Hopfield
model was "feed forward."
the feedback loops that were essential to of a
human
lionfold
retina that
would have
to
It
lacked
Misha s understanding
respond
dynamic range, from moonlight
to
to light over a mil-
midday, that would
have to null out vast changes in ambient brightness to focus on the features of a scene. In talking to Misha, Hopfield too
125
was
all
feed forward and,
George Gilder
126
He seemed
said Feinstein, "in her face."
retina project only to the extent that
work algorithm. To Misha, retina,
if
couraged and decided
approach
with his neural net-
fit
the algorithm did not work for a
could not work for the brain
it
it
interested in Misha's
She was deeply
either.
dis-
drop the course as irrelevant to her
to
to the field.
But Feinstein would not give up. Increasingly intrigued with this
woman and
her retina, her vulnerable shyness and her pow-
and her quakes of laugh-
erful intelligence, her capacious smile ter,
Feinstein protested: "This course
synthesis
to offer a
between biology and engineering. You are the only
ogist taking
"Why
supposed
is
it.
If
you leave
it
will
be a joke.
show your paper
don't you
to
.
.
biol-
.
Carver Mead?" he
suggested.
"Oh, no," said Misha,
"I
can't
work with him.
I
can't under-
stand anything he says. All that electrical engineering.
I
want
to
understand the brain." "He's different in person," Feinstein said.
The
next
week Mead,
Feinstein,
"I'll
arrange a dinner."
and Mahowald met
ner at the Athenaeum, the old Gothic faculty center full
of leather chairs, science texts,
hundreds of
brilliant
at
for din-
Caltech
and men's club smoke. Like
students before her, she
fell
quickly under
the spell of Mead's quiet authority and polymathic knowledge.
Contrary to what she had expected, amid the smoke, he listened attentively to her views their significance.
himself. his lab.
He
on the
retina
He had been
and grasped immediately
contriving
"little
toy retinas"
suggested that she do a project with him, working in
The
So Misha found herself
part of the scene.
in the lab
127
surrounded with a mostly
though Allen, Tucker, and Feinstein were
different set of males, all
Eye
Silicon
Mead
gave her the kind of guidance and
support that she sought. "He was amazing," she said later with a wistful smile.
"He had an unworldly
or possible physically. his expertise
He
intuition of
could take a
and somehow
tell
phenomenon
whether
it
field like biology, full of conflicting data ability to tell
what was physically
was a dead end was
crucial.
I
likely
what was
was
far
likely
beyond
plausible. In a
and speculation,
his
and promising and what
couldn't have
done without
it."
Mead's central proposition was that you could not understand anything unless you could build
it.
The
offered models of the brain resembling a medieval
fully
biologists
map
of the
world, with arrows across the oceans pointing to bodies of
land
—
Broca's region here, area seventeen over there, dragons
lurking in a reptilian neural netherworld, and wildly speculative theories connecting one to the other sels of interaction
the biologists had no
The
way
to use
They could not define the
circuitry.
at all
They
electroencephalograms and nuclear resonance spec-
troscopy to yield crude patterns of energy flow and then
educated guesses that were impossible
As Feinstein different planet
up
ves-
of finding out
what was going on. They could not implant probes
the billions of neurons.
had
to the retina.
might as well have been clipper ships. To an
electrical engineer,
exactly
and
said, "Carver's
from the
to test.
view of a wiring diagram was on a
biologists'
a primitive brain, piece
make
maps."
Mead hoped
to build
by piece, beginning with the "trans-
ducer physiology," as the elder Delbruck had put
it,
all
the con-
George Gilder
i28
versions of energy and electrons in retinas and cochleas and the ionic channels that
who had
he had studied with the Xobel laureate
saw Misha's scheme,
He
said.
When
strode into his office fifteen years before. "it
looked
like a
lion
iUead
wiring diagram to me," he
could see the possibility of greatly enhancing the
sophistication of his electronic models of the retina
and actually
testing the latest findings of the biologists.
Mead
explained:
"When we
build the circuits and
them work, make them render images and to flashes of
light,
we can understand
brain's,
with
ambient
all
find edges
we make and adapt
with flows of energy that parallel the the retina in a
way
that the biologists
masses of murky measurements never can."
their
Nonetheless, for several months, they failed in their effort to create retinal devices that could simulate real functions. Carver's steady guidance,
Under
Misha was learning step-by-step
all
the soldering, and wiring, and "pin" connections onto an electronic breadboard. far
Becoming
a tyro electrical engineer, she left
behind her original notion of a transistor as mere
they had hit a wall in their understanding of
how
radio.
But
to translate
Misha's general schematic into a wiring diagram that would work.
Meanwhile, over dinners taking
Mead
Mead
recalls,
through the
at the
Sawmill Restaurant, she was
latest reports
from the
biologists.
As
"She was better than the biologists, because she
reinterpreted and embellished their findings as they related to
our project." They began by building on Mead's toy retinas.
"It is
always easier to start with a previous design than to begin with a
blank
slate,"
he
said.
The problem was
that the previous retina
The
structures had tell
Silicon
been designed
to
Eye
129
perform motion detection
—
to
the viewer what was moving in an image rather than merely
to present a fixed this function of
'
picture."
At the time,
motion detection was done
frontal layers of the retina itself.
deeper
Misha saw
level of the brain, in the cortex.
motion detection on the electronic
them
for
Mead
The
it
still
in
believed that
analog in the
as being
effort to
done
at a
accomplish
retinal surface distracted
two months.
They made key advances, though. Mead showed Misha how to
approximate the dynamic range of a real retina
—
its
ability to
deal with near darkness and bright sunlight, a difference of a factor of millions
—by keeping the
light
represented by currents
On
a chip, voltages are far
rather than translating to voltages.
more convenient, because they can be divided without strength, providing "fan-out" crucial for neural models.
picoamps
rents can vary from
losing
But cur-
to milliamps, a factor of millions;
voltages vary only from milliwatts to five watts, a factor of thou-
sands or
less.
Mead
figured out
how
to register the intensities
currents and then convert to the logarithm of the voltage.
and
Mead
model
by
Misha
incorporated this feature in their emerging retinal
that could operate like an eye or a usable
camera under
conditions of drastically changing light intensities.
At the same time, through Misha's junior the
dorm continued
to build up.
year, the pressure in
New troops
of baboons arrived.
They saw the curvy female semaphore but did not Misha's feedback. She finally got tired of fending
was an emotional woman and fied
by tears and
cries,
learn to read
them
off.
She
after a tussle at her door, ampli-
she had had enough. She rode the
George Gilder
130
LAX,
express bus to
got
on
a plane,
and returned
She was through. She was persuaded
lis.
only
when Mead managed
with
money
for
Minneapo-
to
Caltech
to return to
augmented
to get her scholarship
an off-campus apartment.
Carver did well by the arrangement. Misha became the
driv-
ing force behind his projects for a retina chip, neuron chip, and
see-hear chip (combining the retina with a cochlea). She astonished the others in the lab with her ability to envisage chip dia-
grams. As one of the leading chip designers, Allen often had to
work with Misha's better.
She was
layouts. "I thought
I
was good. But she was
As Feinstein puts
truly gifted with layouts."
it:
"She had an amazing faculty for spatial visualization: a kind of intuitive silicon compilation taken to a higher level of the sys-
tem, with
what design
The
tem.
the feedback loops. She realized what unit
all
primitive,
would elegantly generate the
wiring problem often
Misha was an
idiot savant
'beginner's luck' but she
lously
and
it
would map out
was her
... in a good way
we became
.
would draw the unit
She
.
.
cell
called
it
and miracu-
into a working system with reflections
inspiration.
fringe of the lab. Just Carver
the end
entire sys-
seemed insurmountable but
transitions, into a real chip layout."
of the retina
cell,
and
The hexagonal
As she I
and
said: "I
structure
began on the
a retinal model.
But by
the mainstream. Everyone was doing retinas
and cochleas." Mead's crowded laboratory became chiefly devoted
Delbrucks complex of "transducer physiology" sions
between input from the world and the
brain. In this pursuit,
Misha was
—
all
to
Max
the conver-
reflective parts of the
in the lead.
Sometimes vague,
The
Silicon
Eye
131
sometimes vulnerable, Misha, with the capacious Havahart trap of a mind,
glommed onto
ideas from
were converging around the lab
—
all
the scientific fields that
neurobiology, electrical engi-
neering, solid-state physics, neural networks, biosensors.
Soon the guys
were eager
in the lab
to share in Misha's proj-
Tobi Delbrtick collaborated on other retinas and on stereo
ects.
matching chips that could convey depth of scene. Tim Allen
went
work on
to
a device to display the output of Misha's latest
on
retinas in real time pixels serially
a monitor, rather than
and then looking
Mead wanted him
gram.
at
them
to build a
later
downloading the
using a plotting pro-
system that would scan
images from the retina and display them
in
NTSC
television
video form, so they could be seen as they happened and adjusted as needed, as in a digital camera. Allen
ous analog video techniques, "I
was
riding
Rose Bowl,
my bicycle
when
was contemplating
disaster struck.
on Linda Vista Avenue, alongside the
uphill, drafting
behind a friend when he swerved
My
head went through the back
around a parked
car.
I
window, leaving
me
badly injured and unconscious.
woke up asking
my
ambulance,
make
it?'
.
.
friend I
was with me.
I
When
all
I
remember
get into Stanford?' (Yes.)
Nervous looks
.
Am
I
around."
turned out most of the damage was facial and required
reconstructive surgery. cost
didn't.
him two questions: 'Did
going to It
in the
vari-
him three weeks
leisurely project for
gency.
The at
incident
the lab.
left
him some
By the time he
scars
and
got back, the
Misha and Carver had become an emer-
He lamented
his
plight
responded off-the-cuff: "Just use a
with another student,
who
ROM (digital read only mem-
i
George Gilder
^i
oi\
I
TV
ROM
he
could store a
conversion algorithm.
without numbers, with electrical patterns
As Allen put
It
all
numerical scheme of the
digital
didn't
all
have to be done
the changes performed in continuous
hard to store and control.
it:
"Instantly,
I
could see the entire solution. All
the complexity of the video scanning process,
waveform
itself,
could be encoded in a
cuit could read the
and
a
and even the
ROM. A trivial digital cir-
contents out and drive the retina chip for the image.''
He
built the
day and spent the next week creating a program to
generate the
As
ROM
few key inputs on the mixer
circuit in a
in analog,
ROM contents in NTSC video format. became something
a result, Allen
of a hero in the lab.
Video sources made on a breadboard are notoriously sloppy:
jumpy synch,
blurry pixels, lots of noise. But the
ROM
digital
approach made everything predictable with rock-solid timing, crisp pixel boundaries,
were both
thrilled
and
a clear picture.
with the results.
I
"Carver and Misha
had crossed the boundary
between 'peripheral character' and one of the Chosen lab."
As Carver put
it
later,
giving
in the
him the ultimate Caltech com-
pliment: "Tim, you really grokked cially significant that
7
it."
No
he had grokked
it
one thought by going
it
espe-
digital
—
by-
separating the computation from the changing real values of the
image and converting them
all
into
numbers synchronized
to a
clock pulse.
during this period
in her senior year, Misha's relationship
with Feinstein grew ever more intense. Lovers and intellectual
The
explorers, they
would
talk
Eye
133
deep into the
night, about philosophy,
Silicon
engineering, biology, the nature of the brain, the
power of
reli-
gion and spirituality. Driving in Feinstein's big Corvette, with
sumptuous black leather
seats, they
would return
to
its
park out-
apartment around midnight and sometimes not
side Misha's
leave the car until five, engrossed in each other, yes, but even
more entranced with the incandescence of pursuits. Leaving "Michelle"
new academic intellectual
behind
at last,
their intellectual
Misha had found
father in Carver, supporting her work,
and romantic companion
her service, she was feeling at
the
first
home
and an
in Feinstein, nurturing
and stretching her mind. With two of the best minds at
a
in the
at
Caltech
world for perhaps
time.
There was, however, a persistent canker of
dissatisfaction.
the world, the Caltech program of putting brain functions in
To sil-
icon was advancing with amazing speed. Mead's laboratory had
become in
globally
famous
in the field.
But
Misha, their progress
to
understanding the brain seemed agonizingly slow and she
continued to pursue
saw
—
against Feinstein's warnings
—what
she
as the parallel path of drugs.
Cocaine, she found, gave her an inspiring feeling of completeness, as
if all
come. Heroin she
and
it
her work were done, tried
all
once and gave up.
did indeed reproduce in an eerie
der and discovery that she
her challenges over-
LSD
was her
way the
favorite,
feeling of
won-
remembered from Disneyland. But
perhaps her most powerful insights came from the ingestion of
THC
(tetra-hydro-cannabinol)
brownies.
—the
essence of marijuana
—
in
George Gilder
1^4
Feinstein
remembers
sitting in the car
with her late
at night,
while she exclaimed that she could count the photons reaching
her eye. She had a vivid and overwhelming sense of the paucity of the information from
which her brain was creating
From
estry of nocturnal sights. less
few scant photons, mass-
messengers, her mind could build a scene
detail,
is
an achievement of an entire mind
beyond the
The
retina could not
do
it
alone.
completed over seven Mahowald's doctoral thesis
that
in
She would have itself.
years
it
reached
.
.
to explore
.
under Carvers wing,
1992 won the Clauser award
the best Caltech thesis of the year.
The
first
that prize at Caltech, she even relieved
life
—
models she was building.
retinal
the farthest shores of consciousness
fears
of mass,
full
and beauty. The experience strengthened her conviction
that vision far
just a
a rich tap-
woman
some
ever to win
of her mother's
about California by giving her "the most exciting day of
when
for
my
the award was announced, the last and most presti-
They dined
gious prize of the year.
with Carver
Mead and
that night at the
Athenaeum
Francis Clauser, the physicist
who had
donated the award. Misha reassured her mother that she had joined the John
Catholic
Henry
women who
Newman
also
Society, a gathering of Caltech
pursued yoga and spiritualism.
As Misha went on her triumphant way, writing chapters, books, she
1993 submitted
umentary on her in
May 1995
to
was called by
WGBH
some twelve hours
life.
of Boston
hour on PBS,
and
in
of interviews for a doc-
Entitled "Silicon Visions,"
for a full
articles,
it
was shown
as part of the science
The
Discovering
series
Eye
135
Women, and was
narrated by Michelle
Silicon
Pfeiffer.
Coming from
biology into the dense mathematical briar patches
of Caltech physics and engineering, she was sympathetic with
newcomers
other
Misha when ence.
I
I
needed a
clear explanation of
some
I
would consult
issue of the sci-
found her a dazzling mix of contradictions. Earnest,
— an experimenter with drugs — she pushed her
ambitious, and modest, engineer,
many
Writing Microcosm,
in the field.
directions that
But through
it all,
at
all
some
once
wistful,
an
a biologist, a physicist,
brain in so
seem
of the time she did not
there.
all
she was following the complex paths of Mead's
science farther perhaps than any of his other students did. I
remember walking behind her
brown braid bounce on her back,
for hours,
as
watching the big
we climbed
the sepia
hills
above Pasadena. Here she often retreated to ponder the conver-
gence between her biological studies and her physics, and here she went from time to time with
man
club.
It
comforted
logical creatures all
time."
her,
women
she told me, "to reflect that
my
perform amazing mathematical feats
She waved away
a curious
fly.
New-
friends from the
"Yes, this fly,"
all
bio-
the
she said, "can
see and react faster than the largest supercomputer while using a billionth of the power.
I
spent
six years at
Caltech learning the
Schrbdinger equation, but the merest electron can solve the Schrodinger equation
all
day long. That
every
movement embodies
know
its hard."
is
that calculation.
what It
it
in the
end
—
its
does not even
Through many permutations, Misha's
on the retina chip would lead
does
to Foveon.
early
But
it
work
would
be hard.
Misha was
a scientist
first.
While many of her colleagues
in
George Gilder
136
Carverland went north to companies Clara,
Misha would not become
her eye on the
fly.
When
Now
it
Palo Alto and Santa
a Silicon Valley
girl.
She kept
she was offered a leadership post in an
eminent neurobiological laboratory land, she accepted
in
eagerly.
at
Oxford University
in
Eng-
Misha had changed Carverland.
she was off to change the world of biology.
Federico Faggin at a desk at Synaptics with a
ment
that he
would eventually make
Cawer Mead?
obsolete.
lot
of electronic equip-
Did
Alvia? Courtesy of Federico Faggin
the rose
come from
13 Synaptics
was not only Misha who was impressed by the thinking
Itprowess tory all
the
and neurophysical charms of
Mead's labora-
flies.
was one of the few places on the face of the earth where
humans bowed humbly
Dipterolatry ruled, and
to the brain of the lordly Diptera.
was driving some of them
it
to drink
and
other distractions. In the lab,
Tim
Allen, for example, imagined that he could
deploy thousands of times more processors than a times more connectivity and hugely more power.
speaking of any
own mental
artificial
neural network.
faculties. Yet
and nervous system of
a
fly. It
swats aside
With ily
its
to
And he was
He was
not
referring to his
all
the notions of neu-
time were claimed in this
which was widely known
in the electronic
millions of
he remains baffled by the eye, brain,
romorphic insight that from time elite laboratory,
fly,
to
be leading the world
emulation of neural functions.
tiny brain
and infinitesimal metabolism, the
eludes Allen's hand, and only by mustering
*39
all
fly eas-
his concentra-
George Gilder
14:
tion
be
and
like
much
effort
can he sometimes
someone
ure, the
fly's
Application
is
less
Specific
with a swatter.
It
would
Building. Yet
Diptera gets away. By any obvious meas-
little
brain
it
him with the Transamerica
hitting
of the time
hit
complex than one of Aliens advanced
Integrated
autonomous power, the
fly
Circuits
(ASICs). Yet with
performs maneuvers that would be
the envy of any aerobatic team with the most advanced airplanes
commanding ASICs. The on
glass or
the computing power of scores of thousands of fly
can do flawless
flip
landings on the edge of a
a glass ceiling while scarcely slowing
find food out of thin air
down.
It
can
from elusive chemical cues. All the Cal-
tech computer firepower in Carvers lab
— Misha's
retina
and
Lyon's see-hear chip, David Gillespie's computer-aided engineering, David Feinstein's mathematical virtuosity, John Piatt's
neural networks, Tobi Delbriick's prowess and creathity with
analog electronic circuitry
—could not even scratch the surface
of the evident superiority, and continuing inscrutability, of the eye, brain,
and nervous system of the
As Allen finished up his career
hoped tive
to learn
how
the
measure explained
fly's it.
brain
The
fly.
at
Caltech
in
1987, he
still
w orked. No obvious quantita-
fly's
neurons were more densely
interconnected than any silicon chip. But they operate millions of times
more slowly And
a million neurons,
however heavily
interlinked, could not in themselves perform the
using any principles of algorithm
The
human
known
to
artificial intelligence or
fly's
repertory
neural network
computer science.
real target of the researchers in Carver's lab
brain.
Although a lummox by comparison, the
was the
human
The
Silicon
Eye
141
more formidable challenges.
brain offered far
Its capabilities,
Allen believed, must similarly stem from as-yet
independent of
tural principles,
since
it
struc-
substrate (or substance),
operates with continuity over a period of decades while
cells constantly alter in
its
its
unknown
so
an environment with ever-changing
temperatures and other conditions. For example, brain retain memories for
stance of the cells
is
Mead, Hopfield, had
that there
to
years or more, while the sub-
fifty
and transformations? Analog?
abstruse synthesis of or
the
replaced and the body endures thousands
of environmental shocks
Quantum? Or an
how does
three?
all
Digital?
Who
knew?
Feynman? Don't ask them. Allen thought
be some architectural concept, some principle
of neural organization, that
was eluding the group.
only they
If
could find
it,
what
Hoping
to
continue the quest, he applied to graduate school
at if
laurels they could win.
.
.
.
Caltech. Like most Caltech graduates, he was accepted only
he spent a year somewhere
seemed
that he
with Misha
would have
who
else, either
needed
ing from
its
a job,
was
still
in Pasadena. Allen
and Silicon Valley was
mid-1980s crash. Living
forty rejection letters
in his
just slowly
VW van,
on the bulletin board
his
lab,
inquiries.
Mead
emerg-
in the lab.
It
was
a
Mead.
followed a rule of never recruiting directly from
he would respond
Mead
now
he posted
cry for help. Finally, he turned in desperation to Carver
Although
It
to leave the brain science challenge
then, in 1987,
seriously
working or studying.
to
students
who made
told Allen about Synaptics
specific
and dazzled Allen
with the exploits of the legendary Faggin, creator of the
microprocessors at
Intel,
and, at Zilog, the Z-80.
first
George Gilder
i_|2
"The Z-80, wow," said Allen.
80
—
that Radio
Shack box
at least
learned to program on a Trash
that used the Z-80.
microprocessor." That was a
but
"I
start. It
was not
he could understand how
by the idea that the designer of the Z-80 parable advances could be
made
now
favorite
amazing
as
as a
He was
worked.
it
my
It's
fly,
excited
believed that com-
in neural networks.
synaptics was a fancy new neural name
for a small pile
Meno
of venture capital lamely launched the year before as
Corporation by aspiring venturer Lauren Yazolino. For help in finding a fashionable product, Yazolino called Faggin, then a
high-powered consultant,
who
suggested his old dream of a
microprocessor based on neural networks. Since the company
had no
CEO,
new world
Faggin was asked to take over and he did. In the
of neural processing, the monicker
Meno
did not
convey the cachet or the concept. Faggin suggested Synaptics,
and
it
stuck.
colleague
To find a neural network
Mead
in
expert,
he called
his old
October 1986, and together they began
assemble a team. In Synaptics
literature,
described as the founders, Faggin as the
Mead
Faggin and
initial
CEO
to
are
and Mead
as chairman.
Arriving at Synaptics in suite 106 in an office complex on
Zanker Road
in
Santa Clara, Allen became the
the reorganized company.
Glenn Gribble,
first
a software expert in
the Caltech lab as an undergraduate, followed later.
employee of
him
a
month
Joining Faggin in the small laboratory, Allen planned to
continue the neuromorphic pursuit. As time passed,
many
of his
The
Silicon
Eye
143
Tom
previous colleagues from Caltech joined him:
David Gillespie,
Adam
Tucker,
Greenblatt, Janeen Anderson, John Piatt,
Ting Kao, Lloyd Watts. Piatt had served as an intern
Schlumberger
dropped
also
petitor,
Artificial Intelligence
in
at the
Lab with Dick Lyon, who
once or twice, though he
first
became
com-
a
designing handwriting recognition devices at Apple. Tobi
Delbruck was a regular
Meeting from time
visitor.
time at dinners in the Valley with
to
and Faggin, Misha came by
nounced them to a Grateful
in the
summer
of 1987 and pro-
a "dynamite team." Allen took her in his
Dead concert
was happy and
at
home,
at
Laguna Seca
in
VW bus
Monterey. She
as at ease in the hippy scene
funny smoke as she was intense under the
Mead
rigors of
and
its
Caltech
graduate school and her emerging thesis on the retina. Carver
himself spent one day a
week
in the office. Synaptics
seemed
to
be becoming Carverland North. Allen never did get around to reentering Caltech. 'The
few years
at Synaptics,
Tobi, except
I
was well
I
was doing the same work
paid, better equipped,
as
first
Misha and
and got more
indi-
vidual attention from Carver." Yet Faggin
Mead had
yet to point to a marketable idea from his laboratory.
Carver was
were ing.
toys.
was growing increasingly uneasy. By March 1987,
still
The
coming
in only
one day a week. The imagers
neural networks could not perform real comput-
Although Business Week would identify the Synaptics tech-
nology as one of the five leading innovations of 1987 the kind of hype Carver could generate
had emerged. Faggin had
—no
—
that
was
serious applications
to return to the venture capitalists with
George Gilder
144
news
the
mode." stuff.
was reorganizing the company
that he
He
Not
explained:
going to take a long time, this neural
"It is
a lot of money, but a lot of time."
the annealing
in
in "research
model, every technology company needs
two kinds of leadership, coming from two kinds of men.
Let's
them, for short, the Meads and the Faggins.
call
One
is
mercurial, imaginative, philoprogenitive, conceiving of
something new and bringing
it
and precarious, but loving
as a
it
malformed
to life, often initially
mother does. He sees
potential in the balky bundle, in a fuzzy-faced
infinite
sophomore with
quirky design tools, in the noisy breadboard, in the oversized and
slow imager chip, in the large young bio major with granny glasses, in the snarl of wires
the
revenues
top
line
— running
—and often wins the
succeeds.
risk tak-
than punctilious, sometimes sentimental, he pur-
ing, often less
sues
and switches. Optimistic,
If it doesn't,
from wistful hope
credit of the
don't bother him;
crowds
he
off
is
if
to
fitful
the venture
on something
else.
This
man
could not consummate his ideas without collabora-
tors, also creative
tion.
Where
Where gin
is
the
the
Mead
looking for
and resourceful, but
Mead is
is
visionary,
lower level of abstrac-
at a
the Faggin
Where
nine-dollar sales.
peripheral vision, seeing the interconnections
bottom
line,
skeptical.
seeing zero-billion-dollar" markets, the Fag-
initial
in his universe, the
is
'
the
Mead
among
is all
everything
Faggin dons blinders and bends toward the
the goal line, profits
now
.
.
.
And
that
baby
in the
The
Silicon
Eye
145
bathwater, that recognizer chip with the blurry images,
it
was a
hopeless kludge.
Though needing each pretend to hold
when
all
other, the
the virtues in their singular selves.
—both Mead and our — they may not be
the claims are true
Renaissance
men
of
two types may sometimes
Faggin, after
era
able to
Even
all,
are
combine
the roles in practice during the course of eighteen-hour days.
When
these two types conflict, their
company
suffers.
But when
the two forms of leadership align themselves synergistically, they
can unleash an unstoppable force.
Carver Mead with
his protege
Tim Allen
at Allen's
ber 1990. Allen was Mead's student at Caltech
wedding
and
touch-pad project at Synaptics. Courtesy of Tim Allen
in
Novem-
the savior of the
1
The Analog Perplex
Tim
Allen saw the
first
two years of Synaptics
in
1987 and
1988 as an exploratory phase. Allen and Glen Gribble
had
set
up the
lab,
developed design tools for neural net-
works, and experimented with several neuromorphic analog cuits.
Tom Tucker
joined for a while and
made
contributions,
but seemed incapable of focusing on this long-term project. left
and did not
It
turned out, as Allen
resembled too
recalls, that "all
down our money. Understanding deep 1989,
for ten guys
we knew
it
much
for a
a
a Carverland lab.
we had done was run
the brain was just a subject too
and ten Sun workstations
was time
He
came and went. Faggin had
return. People
point. Perhaps Synaptics
cir-
to figure out.
By
commercial application of some
sort."
At that point, the company began a sharp expansion. Allen's
good friend Dave Gillespie saw
fit
to leave
Carverland south and
follow the others up Route 5 to Synaptics. Gillespie brought
tremendous sophistication
in
computer science and chip design
147
— George Gilder
14S
to the project.
He
of the
company
April,
John
replaced Glenn Gribble,
Piatt finished his
employee number
He and
Ph.D. thesis for Carver (on graph-
neural networks),
into an Oldsmobile,
out
an obscure but intense row with Faggin. In
after
ics applications of
who had stormed
and drove
to Silicon Valley.
on the way
six,
crammed
to
his belongings
He
enlisted as
becoming chief
scientist.
Faggin soon began collaborating on original research on
"Blind Source Separation," an abstruse field, which used complex time delay algorithms to improve identification of the sources of sensor signals such as sounds for
mixed up
in passage
example, the output of an array of microphones. Presented
NIPS4,
at
a neuroscience conference, these innovations led to the
creation of a mathematically esoteric research field
still
flour-
ishing today.
Although the company had yet
beyond closely
its
research, Synaptics
watched and celebrated
enues and
then
own unique
in
profits
1
to achieve a saleable
seemed
product
was one of the most
start-ups in the Valley. Large rev-
just a matter of time.
99 1, everything changed
for the Synaptics
band of
brothers. Verifone called, proposing a major application for the
technology.
No more
more
from Misha then finishing up
visits
flies,
no more abstruse neuroscience, no at
Caltech, no more
Carverland North.
With
a billion dollars of revenue, Verifone
company
in point-of-sale credit-card readers.
to introduce a device to read the
MICR
was the leading
Now
it
was aiming
(magnetic image char-
The
Silicon
Eye
149
on the bottoms of checks. Just as you
acter recognition) codes
currently swipe your credit card through a reader, so in the
future you could swipe a check. Joining the two key areas of
company
expertise of the
works
— the
new
—image
processing and neural net-
Verifone device would combine a fast image
MICR codes on
scanner with two neural devices, one to find the the check and the other to read the numbers.
be based on the retina work
identify edges,
and look
Caltech, including Misha's ingen-
at
ious edge recognizer circuit.
The imager would
It
enabled devices
The
for patterns.
neural nets would use
from John
"a fiendishly ingenious algorithm"
to detect light,
based on a
Piatt,
He
recent invention called a Support Vector Machine.
also
designed a training algorithm for the device that could process the equivalent of 128 thousand examples in 7.5 minutes on a
Sun Sparc would be
2 server. Storing the synaptic weights on the chip
floating-gate technology pioneered by Faggin.
Allen was in charge of developing the analog floating-gate technology. But
when
Faggin broached the idea to the company's
— UMC Taiwan, chips — the engineers foundry
in
which would have
to
make
the
just said, "No." Floating gates for flash
memory as a
cells that
one or
difficult
to
each could hold a single
zero, with
perfect.
endurance over
ability statistics
not going to
identifiable only
nothing in between, had been devilishly
UMC
had
fifteen years,
almost ten years to
bit,
make
it
to
and
reliable.
it
guarantee
it.
memory's
had taken the company
Now
and the working design
mess with
the
that they
at last, the
had the
reli-
foundry was
Faggin's idea of a floating gate that
could hold not a number (based on binary ones and zeroes) but
George Gilder
150
a
continuous fractional analog weight over time without leakage
or degradation
—
this
was beyond the
capabilities of the
UMC
process.
At Caltech, they had always been able tic
connectors
in a
among
to
program the synap-
the neural net's neurodes with memories
computer plugged
into the wall (that
was
real nonvolatility)
or with electronic potentiometers (like continuous light dim-
mers). But Verifone wanted a cheap single-chip system that
could be bolted to the side of the cash register and would keep its
weights
if
solving the
mental that,
you turned
it
off.
Allen explored the possibilities of
problem with some other device, but he had
barrier.
There were
amorphous
a lot of ideas
silicon circuits
—but
—
hit a
ferroelectric this
and
nothing was near ready
except a return to old-fashioned magnetic cores and no one
made
those anymore. Faggins basic concept for a programmable
single-chip neural net processor
seemed
Allen remembers driving up Route
Cruz
in a rental car
with
Mead
1
to
be
far out of reach.
home
7 from his
in
Santa
discussing the problem. "Carver
always had a rental car because he flew up from Caltech every
week and
stayed at a motel in Sunnyvale." Contemplating vari-
Mead
ous solutions. course"
—Allen
said:
liked that "of
"The simplest way
—"would be course"
weights as different transistor
That was the solution. but the of
MICR
numbers
marks
that
It
sizes.
in the bar
code
blips. It
a
do
it,
of
program the
can do that."
reprogrammable chip,
codes on the checks consisted
would not change
to nine, plus four
to
We know we
would not be
to
—
just have to read
from zero
would be an application-specific
device that could find the codes and recognize them.
The
During
this period, the
such projects.
It
Eye
151
company contracted
to
do a
series of
signed up with a bank to do a cash counter that
could both gauge the
amount
Silicon
number
of each one, and
it
of
bills in
committed
a pile
and
also read the
to contrive address label
readers for the Post Office and for United Parcel Service. Faggin
made
a deal with a scanner
tor for
company
called
Caere
to
do a detec-
an Optical Character Recognition wand. These projects
had baffled many companies
in the past,
but neural networks
and neuromorphic imagers made them accessible
at last.
Meanwhile, some of the limitations of the chip they had
made
for Verifone
were coming
codes were not the same after visited the
It
was
a
all
MICR
cue from Verifone, Allen
warehouse
bad checks. Allen got permission
clear right away.
idealized rendition of the to
On
a large one-story
checks through the machine
became
all.
Perhaps
bad-check repository of the 7-Eleven convenience
store chain. lions of
to the fore.
to
in Dallas. Mil-
swipe bounced
to his heart's content.
One
They had been working from
MICR
codes.
They were
all
thing
a highly
supposed
be the same, but in the real world there was a host of subtle
variations
had
to
and
machine did not work. They
go back to the drawing board and change the design to
accept more sistors
a lot of the time the
and
Where were
variety,
which meant changing the
also possibly increasing the
the floating gates
when
sizes of the tran-
number
of mistakes.
they needed them? Well,
standard floating gates were good enough to store a rough value,
and that would be good enough of neural networks, after
all,
for a neural network.
The
was dealing with imprecise
That was why you used them, what they were
for.
forte
inputs.
George Gilder
152
Employing the standard process used
for digital
memory
cells
programmable read only memories (EPROMs),
in erasable
a
an analog charge with the equivalent of
floating gate could store
of resolution (which in a binary system
means
perhaps
six bits
number
of two to the sixth, or sixty-four gradations of accuracy).
That was
was good enough
exciting; that
for the synaptic weights
of a neural network that could recognize codes.
one waiting trade, "if
I
for? "Well," Allen
mused one
charge with the same accuracy.
and
Digital converters
would not go standard
digital for
EPROM
cells
D
to As.
6 I
Of
course,
That
just the
one
A
tal stuff after all."
tions,
when
memory
to D.
I
I'd
just a
could store the
need Analog
measly 12
to
you
bits of
could store two to the twelfth or 4,096
bits of resolution. That's three orders of still
any-
a hassle. Perhaps
is
But with
bits.
What was
day, in the lingo of his
memory,
just took 6 bits of digital
a
Maybe
there
Analog seemed
to
is
magnitude more and
something
work only
in fixed applica-
the algorithm or inputs do not change.
or programming,
you
Nonetheless, as Mead,
still
needed
Piatt, Allen,
in this digi-
If
you need
to turn to digital.
and Gillespie
all
agree,
with altered transistor sizes the check-reading device began
working the
.
.
.
robustly.
As
Piatt recalls:
"We
really
sweated blood to get
chip working in time for Verifones shipping schedule.
only had a few people ... so everyone helped out and did that
was
lines for
far outside their
We
work
normal job description. To meet dead-
our silicon foundry,
all
of the engineers in the
company
pulled a simultaneous all-nighter, going through pot after pot of strong coffee, to
make
sure that the
first
version of the chip went
out on time and correct." Piatt, the neural network guru and com-
puter science Ph.D.,
became
"chip verification guy."
The
was
"I
Silicon
frantically entering
Eye
153
VLSI schematics
System while Tim Allen was laying the chip start-up," Piatt
Called the
into our
CAD
out. That's life at a
remembers.
I-
1000, the chip fulfilled
Verifone, reading the hardest
MICR
all
the specifications of
codes with an accuracy of
99.995 percent. Integrating a variety of leading-edge analog and neural network technologies on a single chip,
it
seemed
to
be a
dazzling triumph of the Synaptics vision. But what's this? Veri-
fone
s
specs were changing. That was not part of the plan. Veri-
fone wanted a cheaper chip. Verifone's customers wanted a
motor
to pull the
cisely
what the
checks through the machine, which was pre-
optical reading chip
was engineered
to avoid
(with a motor you don't need a smart optical imager; a motor can precisely pull the
check past an ordinary magnetic
Verifone executives
who had approved
ever reason, the deal
fell
the project
reader).
left.
The
For what-
through.
Faggin says today that
it
was a good thing
in retrospect,
because the analog settings on the chip would have deteriorated in use
and
it
would have been
"a bitch to
keep
one else on the project has any idea what he Faggin's growing
ently jaded his
in the field.'' is
No
talking about.
disenchantment with analog devices has appar-
memory
of his successes in launching them. In
any case, the chip used delta-based or differential technology that maintained accuracy
under any conditions that would not
also destroy a digital device.
author, the virtues of the
I-
But even by Faggin,
1000 are
the collapse of the market for
it.
less
its
ultimate
remembered today than
Retrospect
is
like that.
After the Verifone fiasco, similar frustrations afflicted the
other projects. Unable to create a general-purpose neural net-
George Gilder
i54
work processor with programmable synaptic weights, the Synaptic
devices
some two
became special-purpose custom
years to develop
chips, each taking
and each only good
one applica-
for
Even with changeable synaptic weights, the
tion.
difficulty
remained. For recognizing numbers, a ten-item matrix was optimal, but for letters you needed twenty-six, regardless of
how
variable the synaptic weights were. Similar problems afflicted
almost every neural architecture and application. that neural nets
been teaching
were
and
parallel processors,
It
turned out
as Carver
for fifteen years, "parallel processing
had
is
inherently
these obstacles to the analog ambitions of
Mead and
application specific."
But
all
Faggin were
trivial
compared
to the real
showstopper on Zanker
Avenue. During those two years while Allen, Gillespie,
Piatt,
and
the others perfected their special-purpose analog devices, the world's ten thousand digital chip designers
bled the capability of digital technology.
two hundred thousand new "Just think
against.
now, Intel has increased the clock rate of
Caere, the Post Office, or the bank
that
to
be
Faggin pointed out,
Between 1986 and
digital
from 50 mega-
By the time Synaptics would present
amazingly ingenious and low-power
pened
They generated some
digital designs.
what we were competing
hertz to 3 gigahertz."
had more than dou-
—had come up with
little
its
analog chip, Verifone,
—whoever the customer hapa high-power digital solution
was good enough and that could be simply plugged
in to
an
IBM PC. It
tells
was not only Synaptics the story of CTI, a
that
company
fell
that
before this cycle. Faggin
makes telephone terminal
_i
The
Eye
Silicon
equipment. During the same period aged to bring to market an analog
155
in the early 1990s,
memory
serial
it
man-
that could dis-
place the tape recorder in a phone answering machine.
It
was
cheap, robust, nonvolatile, and good enough to record fifteen
seconds or more of talk in
memory producers began tive chips.
memory
Then
the
tell
row,
markets for their defec-
to search for
was
a
killer.
the difference.
system might be,
it
But
for a voice recording,
If
no
However ingenious CTI's analog
could not compete with an analog-to-digital
converter and billions of nearly free reject chips from fabs.
DRAM
For most digital computing applications a defective
cell, or
one could
rendition.
its first
you needed
nonvolatility,
add
observed sardonically: "The world has
a
many
battery.
DRAM
As Faggin
cylinders."
Gillespie concludes, "This experience led to a lot of soul-
searching around here. For linking to the real world, analog indispensable. But trying to build analog circuits to do a
putation that could be done digitally are going to lose. If
you need
to
is
com-
to fight City Hall.
compute, go
to Fry's
is
You
and buy a
computer."
With
all
Silicon Valley trooping to Fry's, the
reached an impasse. By 1992, the road to failure. vice, the
The
cash counter
all
all its
company had
contracted projects were on
Post Office, Caere, United Parcel Ser-
were kaput. Faggin and Mead, though,
each had one further project under way.
Mead was
doing
CMOS
imagers, with the goal of developing his retina technology into a
new
analog camera that would collect
accurately and usurp film.
Mead
work was "extremely promising."
all
the color information
said Tobi Delbriick's imager
156
(
reorge Gilder
Having watched Carvers analog systems come over
previous
the
lous. Regardless of
$20
billion
six
years,
however,
camera industry seemed
had served
tion, the
nothing
was incredu-
any Delbriick miracles, competing with the
ambitious. Let's just say "impossible." gin
Faggin
to
for ten years
to
be more than merely
Or maybe
crazy.
But Fag-
on the board of Logitech Corpora-
computer interface device company. There, two years
before, he
had come up with
a better idea.
Dave ible
Gillespie at his office, with one of his Synaptics touch pads
behind
his right shoulder.
From
vis-
the iPod to the "teleputer," Gilles-
pie sees ever-rising markets for "haptic" devices. Dave Gillespie
15 Seizing the
As
Sword
Peter Drucker has written, in any industry the largest
profits
tend to go to the company which supplies the
missing element that completes a system.
Located Valley,
in
Freemont, California, across the bay from Silicon
Logitech Corporation was doing well as a maker of com-
puter mice and trackballs (essentially upside-down mice) that
completed desktop computer systems. however, Federico Faggin had
come
On
to
the Logitech board,
understand that these
technologies would not suffice for notebook computers. Pretty
soon notebook computers would be most of the business, and they
still
had no adequate input device.
Trackballs took too
moisture, and, likely to fall
cabin.
If
little
you
out and
space and depth,
tried to clean roll
— not
one
in
under the seats
a stewardess stepped
might go flying a
if
much
gummed up
with
an airplane, would be
to the other
on one, the
a useful cursor effect.
end of the
coffee, tea, or milk
IBM
had concocted
track-stick that avoided the moisture problem, but
i59
it
George Gilder
160
IBM
stuck up through the keyboard and gave you the
you thought that way
many people
—whenever you
made
finger
a mistake,
—
which
did as the cursor gyrated wildly across the screen.
With Logitech engineers, Faggin came up with the idea
of a
computer touch pad and had persuaded the Logitech board adopt fect
it.
if
to
After an exhaustive worldwide search, he found the per-
company
to build the chip to
run the device.
It
was
a
little
firm across the bay called Synaptics. Indeed, Synaptics probably
had the best
skills for
designing a system that could recognize a
and
finger position robustly
have
to
be analog, and
it
accurately.
would have
to
Such
a device
would
have significant powers
of pattern recognition such as were uniquely afforded by neural
networks.
To pursue
this opportunity,
Faggin proposed a partnership
between Logitech and Synaptics. Logitech already had the and firmware
ers
force to market
invention was
to
it
connect the device
to the
PC
dampened when he
and found 170 examples of to
industry.
to the
PC
and the
Although Faggin's
driv-
sales
thrill
of
investigated the patent record
"prior art,"
no one had yet managed
put the touch-pad idea into practice. After
all,
there had been
plenty of "prior art" ideas for the microprocessor also. Early in 1992, Faggin signed a contract with his friends at Logitech, Per Luigi Zapacosta and ian
connection"
Borel. tion,
—and
The group
settled
their
Giacomo Marini
"the Ital-
Daniel "Bobo"
Swiss colleague,
on price parameters
—
for a
two-chip solu-
including a Motorola microcontroller, with plans to
a single device
veteran
when volumes
who had begun
move
to
rose. Steve Bisset, a Silicon Valley
at Intel
and defected
in
1976
to
form
The
Silicon
VP
in 1989.
161
had joined Synaptics
chip-tester pioneer Megatest,
ing
Eye
as a market-
Faggin put him in charge of coordination with
the Logitech engineers.
was the idea of dispensing with
Bisset's greatest contribution
the need to attach the touch pad to the computer as another
He
layer of gadgetry.
suggested that they
move
the touch-pad
capacitors into the back of the computer's printed circuit board itself.
become
Circuit-board layers had
accommodate the
so thin that they could
entire sensor array, enabling an entirely inte-
grated touch pad. Taking virtually none of the precious space inside the computer, one side of the
pad would bear the chips
and connectors, surface mounted on inches square. face space
The other
side of the card
where you place your
a printed circuit card
would provide the
sur-
finger.
Faggin led the planning and assigned the design job to Bissets
new Intel
hire
Bob
Miller, a burly six-foot-four
who was uncontaminated by
ing with
Caltech graduate from
the earlier failures. Consult-
Mead, Miller learned the magic
of analog.
He
a system with analog potentiometers, floating gates,
novel circuitry. says, "It
The chip
was close
fit
designed
and other
Synaptics technology well. As Faggin
to the teaching of Carver, with analog sensors
and recognizers and compensation using floating-gate devices." Faggin, Bisset, and Miller got patents on the device.
Soon plunged
into a bitter divorce, however, Bisset
Faggin and asked to leave the project.
and nurse
his
wounds.
He
could not accommodate him Carver,
seemed enough
He needed
came
to
time to reflect
asked for a part-time job. Faggin at
—but
Synaptics
when
—one key
part-timer,
Bisset suggested that he
George Gilder
162
move
to
Logitech and manage the project part-time from the
other side, Faggin saw a "win-win" for both companies.
vened
make
Logitech to
at
the arrangement. Bisset
He
inter-
moved
to
Logitech and was put in charge of the touch pad. By early 1992
came back from
the prototypes
enough
to signal a
the foundry and worked well
go for the project. Faggin went to Logitech to
finalize the deal.
was then
It
They
that his friends at Logitech
Bob Miller they wanted
told
would take nearly puts
had to
it,
"It
was the
this device
make
all
began
to get shifty.
a drastically lower price that
the profit out of the product. As Gillespie
classic situation for a joint venture to
which had
a lot of
We
added value and we wanted
had
a profit, while Logitech
fail.
all
market connections and wanted the chip
the key interfaces and at cost."
Faggin thought he could straighten out his longtime colleagues. After
all,
it
had been Faggin who had conceived the
touch-pad opportunity for Logitech in the Italian
place.
But
his
connection had frayed, as Per Luigi and Giacomo had
resigned from over.
first
On
management and Daniel "Bobo" Borel had taken
the Logitech board, Faggin had long supported
and gotten along well with him. As a to pull out all the stops
last resort,
Bobo
Faggin decided
and resolve the issue man-to-man. Meet-
ing Borel for breakfast at the Stanford Court Hotel in Park, Faggin thought they
Menlo
were near agreement. Faggin would
just patiently explain the cost situation,
and
as
soon as Bobo
understood the deal would be done. Every time Faggin tried to pin
would
shift
down
the terms, however,
Bobo
ground. Like Oliver Twist, he always wanted "more."
The
Silicon
Eye
163
Borel believed that Synaptics had no recourse.
"We were
nered," said Faggin. But as creator of the
first
microprocessor
and an industry paladin, he was not going
run a profitless con-
tract design crazy.
But
house
to
He knew he was
for Logitech Corporation.
He
his pride got the best of him.
cor-
stood up at the table
and announced that the two companies would have
to go their
separate ways. Yes, he was resigning from the Logitech board.
He would Back
figure out
at
some course
for Synaptics. Bye, bye,
Zanker Road, Faggin summoned
his
all
company meeting. After much wrangling and many he called
team
for a
questions,
an entire touch pad, com-
for a crash effort to create
plete with drivers, firmware, ergonomics, that could be designed, built,
Bobo.
and even economics,
and sold by Synaptics.
If
they
could not win by joining Logitech, they would win by beating
them. This meant a complete
life
change
for the
company. For-
get understanding the brain, forget exotic imagers, forget the
comforts of Carverland. Faggin was leading the company into the fierce, unforgiving crucible of competition in the global personal computer business.
He had
leave.
If
you
didn't like
the board on his side.
it,
well,
And one by
you could
one,
many
of
the previous stalwarts did indeed leave. Faggin's decision might not have
seemed
as
foolhardy as
Mead's idea of competing with Kodak, Nikon, Canon, and Sony, but
was
it
close. Logitech
had Steve Bisset and
his intimate
knowledge of the Synaptics design. From long experience
mouse
in the
business, Logitech had the crucial and time-consuming
firmware and software drivers to connect to any personal computer.
And
Logitech had close
ties to
every major manufacturer
George Gilder
164
of personal computers, including the inscrutable
webs
wanese fabs and contract manufacturers and obscure but niche
PC
of Taicrucial
players elsewhere in Asia.
Alps in Japan was already well on the way to producing a
touch pad for PCs. Moreover, one of those 170 patent holders that Faggin in Salt
had found turned out
Lake
City,
to
be a company called Cirque
and they had already
tied
Apples new Powerbook. Synaptics seemed
up
to
a contract for
have almost no
chance. But in the face of the definitive failure of
its
other
designs and the quixotic prospects of the imager project, Faggin
could present "almost no chance" as a tempting
new
horizon of
opportunity. Battered by long failure, the board did not balk.
Nearly everyone was willing to "bet the company" on something,
and
if
Faggin thought touch pads were the ticket, they would
buy another
ride.
At the
least,
touch
pads meant competing
with Logitech and various companies they had scarcely heard of rather than with
Nikon
or Kodak.
Allen and Gillespie also supported Faggin's decision. But until this
meeting the touch-pad project had been a fringe operation
at Synaptics.
"An ugly duckling,' "
as Gillespie said, "without ele-
gant neural nets or imagers." Allen and Gillespie were focused
on speech recognizers, pattern matchers, tions. Enthralled let
by
all
Faggin, Bisset, and
retinas, neural func-
this
high science, everyone was willing to
Bob
Miller,
all
old Intel hands, have their
But when Allen
way with
this low-rent project.
nized
the gory details" of Miller's design, his stomach began
to
"all
churn ominously. As he thought about
ready to scream, "Not again!"
it
more
He concluded
would bring the company down.
finally scruti-
deeply,
he was
that Miller's design
The
Eye
Silicon
165
Through the Verifone episode, the Post Office Caere scanners
ments
—Allen
— through
a
set
boundaries between analog and all
had
to
of rules for marking the
The
digital.
direct interfaces
be analog. So the finger would have to
be sensed and located through analog circuitry capacitor matrix. insulator.
A
UPS,
the gay launches and gory denoue-
all
had annealed
with the world
fiasco,
capacitor
is
— through
a
two conductors separated by an
A finger is essentially "a bag of water," a conductor, The
the sensor array would be conductive.
finger
and
would move,
and the capacitors would discharge and issue a rough pattern of data that could ultimately define the point on an x/y capacitive
would be analog.
matrix. All these sensory functions Allen's rule
stants
circuits
away before you had
was
fine for signals with time con-
to
use
it.
and
fast.
placement
A
But
if
and comparison
the time spreads toward a second, you have
touch pad would have to remember the fingers
for as long as a
on the pad.
It
minute or more
—people
rest their fin-
could not forget where the finger was.
could not forget the residual pattern
Bob
Filtering
could often be done in analog, since they operate pas-
to go digital.
gers
that analog
on a scale of milliseconds. The charge would not degrade
or leak
sively
was
when
It
the finger was absent.
Miller intended to store these values on analog floating
saw
gates designed into electronic potentiometers. Allen gates as nothing but trouble.
They would
drift
and
floating
distort the
output, assuring failure.
Tim Allen was the
felt his
job was on the
line.
But forming
in his
mind
entire circuit diagram of a device that could succeed, a
device that might even prevail in the market. Gillespie
and invited him
to
come up
He
called
Dave
the next day to his place
George Gilder
166
near Santa Cruz above Route 17 to help him define the details. Allen himself had conceived and helped design the house.
mostly glass structure on a ridge 3,000 feet high
Soda Springs Road, not change see from
it."
it
was
"all
analog," as Allen put
Through the windows on
Monterey
to
good
a
A
at the top of it.
"You could
day,
San Francisco, from Bay Bridge
you could to
Pebble
Beach. But this was not a good day. There was a wild storm outside, rain beating in torrents against the glass.
Even an ordinary
storm in the valley would bring 80-mile-per-hour winds top of the ridge. Allen's
anemometer ran
other occasions he had seen Gillespie
showed up
at
it
to
120 mph, and on
top out.
9 a.m., running up the rocky dirt path
through the rain to get into the house as Allen's wife work. Huddling up in the living room, they
was under
siege.
at the
"We were standing on
left for
felt that their
world
this little island, this
beautiful analog structure that Carver had designed at the top of
the mountain" from which you could theoretically see the city in the distance, the froth of the ocean, the glow of San Francisco, the
jeweled for at
lights of
Monterey
— symbolizing huge
potential markets
neuromorphic transducers and elegant neural networks. But
the time
all
you could actually
of creative destruction winging against the
see, feel, or hear
were the gales
up from Silicon Valley and Japan
windows of the house.
Nonetheless, hunched over that kitchen counter of gray granite
in
glass,
the eight-faceted
180-degree curve of the living-room
they found an eye in the storm. Totally absorbed in the
project, blocking out the world, they created a design for a work-
able touch pad based on Allen's concepts and on Gillespie's
.
The
Silicon
shrewd ways of squeezing the cheap
Eye
167
computation out of
last bit of
They could do much
eight-bit digital microcontrollers.
of
the calculation on special-purpose digital circuits on chips. But if
you were going
microcontroller,
"Dave did a
it
to
move
all
the rest of the computing to the
had better be able
to
do the job
in real time.
brilliant job," said Allen, "of partitioning the algebra
so the right answers dropped out naturally."
So
it
went. Again and again they chose to replace clever ana-
log solutions beloved of Carver
devices.
and even Federico with
As they worked through the lessons of previous
digital
projects,
they found themselves shunting aside, one device at a time,
most of the some
thirty
patented analog circuits that the com-
pany had concocted over the
years.
For Allen and Gillespie, stalwart graduates of Carverland, this
was
painful.
to leave
"Going
stand
it.
It
dragged
it,
"was
So we began
me down
it
worked pretty
well.
It's
It felt
In this touch pad, there
is
couldn't
just a sim-
empower-
analog. All digital processes ape
would be no analog
electronic potentiometers,
zona,
I
was analog concepts that they were simulating. But
analog functions conducted by the mostly analog
A
deciding
into the details too
to simulate stuff in digital.
everything in the real world
ers.
like
slowed the rate of experimentation. In the end
ulation, right? Yeah, but ing." It
put
home. To run away from your parents. But analog was
too hard, too slow. quickly,
to digital," as Allen
digital
human
brain.
floating gates,
no
no clever analog adders and multipli-
microcontroller from Microchip of Chandler, Ari-
would perform many of those computations.
Allen and Gillespie finished their work just as the storm
George Gilder
i68
relented and the sun began to peer through the clouds and the
image outside their window began
As they stood up and looked exultant.
They had
to take
shape from the murk.
each other, they
at
a design for survival as a company. But in
electronics, as they knew, there
were many designs, not many
They were
products, and most companies ultimately go broke. just
two engineers
and they lacked
in a very flat organization,
any clear channel
were
at first
to
impose these last-minute ideas on
their
company. Neither Allen nor Gillespie was used to company neither
knew
the next step. But whatever
a severe attack
on the work of
was,
it
a colleague.
it
politics,
and
would mean
As the senior
figure
Allen would have to do the job, and he had no such experience. Miller and Faggin were close, he knew, and Millers design well
down
the pipeline.
would not be easy over, Allen
The
to stop.
train
had already
left
With the heavy use
was
the station.
It
of digital, more-
doubted he could gain Mead's support. "Mead's not
going to like
this,"
he
said.
But Gillespie insisted, "You have got to have lunch with Carver." Filled with trepidation, Allen arrived the next for the Train restaurant
have lunch with Mead.
was cruising
"We have
on Middlefield Road "I
fully
thought
I
day
in
at the
Menlo Park
was already too
to
late.
I
for trouble."
a
problem
the problems with
at Synaptics,"
Bob
Miller's
Allen said.
He
explained
touch pad and the
dependent solution he and Gillespie had
crafted.
digital-
Then he
waited for Mead's reaction. After an excruciating pause, finally spoke:
Late
Mead
The
"It's
like
it
a
lot.
think you've grokked
I
169
telling
it.
He had been
him what But
cial place.
it,
I
Tim."
immediately grasped the validity of the new design
and endorsed
He
Eye
an adaptive analog circuit doing collective computation.
Mead had was
Silicon
if
it
had
you want
told Allen to
listening to the technology,
told his students
to
— analog has
and
it
a cru-
do a computation, buy a computer.
work out the
details in a
memo
and
fax
it
to
Faggin in Japan, where he was attending a conference. In the
course of writing the
memo, Mead
says, Allen
significantly
improved the design. Faggin was also ready for a
and
new
Gillespie, Faggin himself
design.
Unbeknownst
to Allen
had become deeply suspicious of
analog computation. By the time he arrived at Zanker Road,
he was ready Miller to a
to
new
put Allen in charge of the project and move role,
managing
relations with
the contract
manufacturers.
Then
in
May 1994 emerged
Cirque touch pad on four
its
the
new Apple Powerbook,
digital
ASICs.
It
with
brought wide-
spread acclaim as a major advance over trackballs.
The PC
industry immediately began to look for suppliers for a touch pad for its
PC-compatible notebooks. Alps was soon
own two-chip
ately resolved to
in the
market with
solution. In a critical decision, Faggin
adopt the Alps form
nectors and power levels.
He
factor, all the
immedi-
same con-
purchased drivers from the same
contractor used by Alps. Alps's device, however, suffered
from
sensitivity to
ambient
electromagnetism and moisture. Synaptics's pad did not. After the experiences with previous Synaptics products, Allen and Gillespie
had incorporated on-chip error correction and redun-
George Gilder
170
dancy and used redundant memory technology pioneered
mainframe computers. These decisions increased also
enhanced the immunity of the pad
programmable read only memories)
the Synaptics touch pad
PC
that the entire
company wanted ware could be 'Timing times
is
was
ready,
three
was precisely the device
fit
for the
a
data into a pot,
it
helped to be prepared. Full Synaptics had
moment
for five years.
company's expertise, but a perfect
mass of ambiguous capacitive is
'Where
stir it
answer. By taking a
is
the finger?' You put
bunch
—
it is
of
crummy
as
it
all this
murky
estimates and have
it
It is
not
them liter-
has no Hopfield algorithm, for exam-
domain of messy real-world
computer domain of all
and the
a similar kind of collective computation, trying to
bridge the analog
With
touch
around, and hope to get a solid dependable
neural network
—but
A
fit.
signals,
vote you can get a very high quality response.
pad
If a
a slightly different device, however, the firm-
for this
answer you want
ple
later,
everything in this business,'' Faggin says. "Some-
pad converges
ally a
months
Gillespie reflects, "Touch pads turned out to be not merely
good
all
All these
altered.
been preparing
a
firmware over
ROMs.
of analog-sensor and neural-network experts,
As
(eras-
notebook computer industry was seeking.
helps just to be lucky." And
it
When, it
EPROMs
for the
the significantly cheaper but unchangeable
decisions turned out to be crucial.
costs, but they
to outside interference.
Allen and Gillespie chose one-time changeable able
in
signals with the
digital precision."
the backing and
filling
and to-and-froing across the
settled for the correct path, the processor
had
to per-
The
Silicon
Eye
171
form a kind of cognitive annealing. Settling on a correct path the company,
it
was
time to plunge the Synaptics sword
finally
computer market. Over
into the icy turbulence of the personal
the next six months, with deceptive ease, the
some 60 percent
to capture
Holding most of the
rest
Then Logitech came Bisset's guidance,
least
would be twice
the price of
own
its
it
sold.
In the price
pad. Designed under
The
war
was
it
a one-chip
using two chips, which in theory at
30 percent below Synap-
at $6.50, nearly
would
of $9.00. At $6.50, Synaptics
tics s price
every pad
forth with
in Japan.
as expensive to manufacture. Logitech set
system
its
was Alps
still
company managed
of the global touch-pad market.
cultivated at Synaptics,
was
device. Synaptics
for
battle
seemed
to
be
lose
money on
over.
that followed, Logitech did
make some
small
inroads into the market. But Allen and Gillespie were already intensely at
work on
a
new
single-chip solution incorporating
all
the advances of the two-chip system. Faggin put Gillespie in
charge of the task of designing a microprocessor exactly cus-
tomized for
all
the special
demands
of touch pads. Gillespie
demurred. Designing a microprocessor titans
of
the
industry
—people
like
is
something that the
Federico
— accomplish
through legendary feats of genius and tenacity. But Faggin per-
suaded Gillespie that
it
did not take a "god or a Faggin to
make
a microprocessor."
With
Gillespie leading the processor design
and Allen contriving the
emerged
in 1997.
and the firmware
overall design, the one-chip solution
At that time one of Logitech's main customers,
the contract manufacturer of
Compaq's notebook, came
to Fag-
George Gilder
172
gin to report problems with the Logitech device, sitive to interference.
ready to plunge
adaptable
in the
alternative? Faggin
sword again. With
EPROMs, and
uct proved far year,
Did they have an
which was sen-
its
neural algorithms, the Synaptics prod-
more robust than any other touch pad. Within
Logitech was virtually out of the market, and
withdrew Faggin.
was
error correction,
entirely.
'They threw
in the sponge.
We
in
had
1999 it,"
a it
said
Facing defectors from Synaptics, former partners
at
Logitech, a continual stream of analog frustrations, Faggin had finally
uct.
triumphed
in a
new company
with a radically
new
prod-
His vindication was sweet.
Meanwhile, amid
all
the excitement over the touch-pad
breakthrough, few noticed that Carver and his team had taken a
remarkable picture of a
cat.
IN
A
scarcely risible cat lurks behind the blue pixels in the cover photo-
graph article
for
Carver
Mead and Misha Mahowalds
on the future of analog imaging. Could
threaten
Kodak
7
Scientific
American
this technology
ever
16 The Cat
While on
Faggin early in the 1990s was taking Synaptics
his
happy
ride into haptics
imager projects remained
—touch pads — Mead's
far short of the
But perhaps because they tapped a deeper stream of
market.
scientific
research, they continued, ever so slowly, ever so relentlessly, to
gather
momentum
both with Mead's
with Misha and her colleagues Fruits of
at
team
little
at
Synaptics and
Caltech.
Mead's collaboration with Mahowald spanned the
entire history of Synaptics, beginning a
launch in 1986,
few months before
when Misha was embarking on
its
her career as a
graduate student, through her departure to Europe in 1992, and
through the power of her design ideas afterwards. In 1987, in a Caltech classroom, his conviction that
first
publicly declared
what he termed "neuromorphic analog VLSI"
offered the possibility of a radically sor.
Mead more
effective
image proces-
For Mead, this assertion was not new. But later in the class,
he presented the
first
actual example of such a machine, a
175
sili-
George Gilder
176
con retina chip, based on Misha's design. Modeled on the
human
eye,
could follow a rotating fan without aliasing
it
do
(reversing direction as spinning wheels also adapt to drastic it
changes
in movies).
The images
in the intensity of light.
offered at the time were crude. But unlike
all
could
It
the other crude
imagers of the time, including the popular digital imagers used in the robotics industry to inspect
work
in progress in factories,
these images reflected a deep study of brain functions and processes.
Mead
explained:
"When you examine how
visual information, several facts are clear. as
system by as
meant
many
that the
as nine orders of
tem
is
a fly
[billions]."
He
—comparing the power
—
of the
fly's
neural reflex sys-
incomparably superior to anything computer science can
build or even model.
does
magnitude
"power delay product"
usage with the processing speed
it.
And
if
know how
"We have almost no
we cannot the far
The operation
separate memory.
work
in
real
figure out
how
idea of
the
fly
how
does
the
it,
more complex human brain does
fly
we do
it."
of the brain shares virtually nothing with a con-
ventional digital computer.
to
Even the brain of
eludes a swatter outperforms any real-time computer image
it
not
biology processes
It
It is
not serial and
it
does not have a
does not have a glossary of symbols. Forced
time
—creatures
with von
Neuman
imagers
would have been eaten by creatures with real-time brains
—the
The
brain
key neural constraints are energy and connections.
had
to
work
in a
remorseless environment of scarcity, in time,
power, and communications.
Examining the
brain,
we
find
that
unlike
a
computer
The
schematic, which wire
—adding up Less than
local.
percent of
its
to 1
Eye
177
dominated by several
some seven miles
percent of
energy,
80 watts used transmitting
is
Silicon
which
its
layers of long-distance
— the brain
in turn
between
supremely
wires are remote. Less than is
less
than
1
in a leading-edge microprocessor,
signals
is
different
parts
1
percent of the is
of
devoted to the
brain.
Although some experts have spoken of the "huge communications resources" of the brain, involving "millions" of axons (the
long-distance links), these millions compare to local billions of
The
millions of connections.
millions of axons represent roughly
one-billionth of the total synaptic lines,
order of quadrillions (10 of
15
The
brain lavishes
the neurons.
Though McCullough and other neural-network
champions have termed figure implies an active
this a passive process, this
70 percent
and transformative kind of communica-
resembling a data flow or systolic array more than a von
Neumann
computer.
Calculated in the 1980s by Margaret versity of ics of
Wong
Reilly of the Uni-
Michigan, these percentages reflect the energy dynam-
the only image processor that demonstrably works.
overthrow nology.
way,"
some 70 percent
energy on dendritic processing, the input trees that link to
its
tion,
).
which number on the
"I
all
the prevailing assumptions of passive imager tech-
do not imagine that
Mead
understand. efficiently
They
says.
The
than
I
understand the brain in any
"But our retina chip result
digital
is
that
is
full
based on what we do 4
it
operates at least 10 times more
imagers do and
it
can simulate some of
the functions of vision on silicon in real time." It
was
a significant first step
toward creating a real-time
George Gilder
178
imager on monolithic
silicon. In the future,
devices could
silicon
Mead
believed, such
empower improved forms
vision or superior solid-state
machine
of
cameras that could process images
as well as merely take pictures.
To the argument
cumbersome
bulk
CMOS
Mead
to build,
silicon
used in
To the argument that
CMOS
would require
and thousands of discrete devices that would be
exotic designs
too
that such an analog system
it
all
offered his device in the plain
the digital chips in your PC.
would use too much power, he ran
transistors at exquisitely
and P
tive) electricity of electrons
low power. Both the
N
(nega-
(positive) electricity of "holes"
(where electrons were missing from the material) he ran subthreshold voltages like the micropower systems in
at
digital
watches. Even the light-receiving photoreceptors needed no
complicated addition to the Every every
CMOS designer faces a fundamental problem.
cell's
positive
two
devices
transistors
—
—the
a potential
is
latch-up or parasitic device. a
CMOS process. complementary negative and bipolar transistor,
space charge region between the two
Where
positive
area termed a
and negative forces meet
P/N
CMOS
is
called
transistors.
in a device is
junction: a scrimmage line
the
called
can crop up across what
It
Between
an active
where contend
the negative electrons and the holes (electrons missing from the structure of an atom). Every
some way
CMOS
process has to figure out
to neutralize the potential parasite "latch-up"
the transistor that
the transistor that
is is
switched by a positive voltage on
its
switched by a negative voltage on
But rather than neutralizing
it,
Mead
between gate its
and
gate.
listened to the technology,
The
Silicon
Eye
179
and actually enlarged and enhanced the bipolar latch-up tor that the silicon
wanted
to provide.
He made
it
transis-
into a
com-
bined photodetector and signal amplifier. Collecting light on the
P/N junction
transistor amplified the signal in the
and
emitter.
With gain
at its base, the latch-up
channel between
its
source
of over 1,000, the latch-up outperformed
ordinary photodiodes that would have to be placed on a separate
chip from the image processor. Integrating the photodetectors
onto the
CMOS device,
systems that scaled
Mead showed
the
way
to create analog
systems in accord with Moore's
like digital
Law.
The latch-up photodetectors
play the
same
essential roles as
the 6 million cones and 120 million rods in the single
photon with an energy
will create
human
larger than the silicon
eye.
A
band gap
an electron-"hole" pair that causes swarms of elec-
trons in the base of the transistor to cross the channel. Govern-
ing the amplifying gain of the transistor
electron hole pairs
is
the
life
span of the
—how many holes can pass through the base
channel before they recombine with electrons and lose their charge. In this case, the transistor multiplies one photon into
hundreds of electron charges.
Similarly, in a
human
retina, a rod
converts a single photon into a million electrons worth of charge in the
span.
form of sodium
Though
ions,
with a gain dependent on their
life
radically different in detail, both are analog cas-
cades of amplification that use the lapse of time as a crucial guide in generating an image.
The human eyes and mind embodied
in
the
see and sense by the analogies
neurochemistry and electrical patterns
in
George Gilder
i8o
the brain.
They
don't even
know
it's
By
hard.
contrast, a high-
resolution digital imager takes billions of steps every second
consumes scores of watts of power and heat
and
to calculate a single
picture.
Only with agonizing
VLSI
would Mead's analog
delays, however,
gain adherents outside the circle of his
senting his ideas to an audience at
MIT
own
students. Pre-
few months
a
classroom revelation of the analog retina chip in 1986, unfriendly classroom guffaws and endured
ment provoked by Mead's new
insights.
some
1
after his
incurred
of the resent-
The MIT
professors
were proposing optical computers that used photonic devices
modeled on
borrowed from Mead, that the absence of
assertion,
optics in the brain
processor
They laughed out loud
digital transistors.
—
—the
at
my
digital
world's only fully successful image
cast serious doubt
on the prospects
for
image pro-
cessing with digital optics.
Only slowly did the analog years
later, in
May
retina chip crawl forward.
1991, after between twenty and thirty itera-
tions of the device in Carver's lab,
American. Representing
was the face of
But four
this
it
made
the cover of Scientific
technology to the world of science
a cat registered
by
nal camera. Inside, the story by
Mead and Mahowald's
Mead and Misha was
reti-
confi-
dent: "The behavior of the artificial retina demonstrates the
remarkable power of the analog computing paradigm embodied in neural circuits. ...
tations are based
on
A
neuron
is
an analog device:
ones and zeros. Yet neural systems
work with basic physics rather than against
it.
.
.
compu-
on smoothly varying ion currents rather than
bits representing discrete
."
Its
trying constantly to
work
The
As Mahowald explains for Stereoscopic Vision,
that
world.
.
.
measured much ration
—with an
log system
An
Analog VLSI System
same
level as the
as part of
system [the
interacts directly with the visual
It
responses are real-time electrical signals
[Its]
.
181
"The analog system functions
emulates.
it
Eye
her book,
in
physical world, at the
the real, retina]
Silicon
.
.
.
as are the signals in a real physiological prepa-
oscilloscope.
and the
The one
biological
similarities
between the ana-
comparison between
facilitate
the two and create a milieu of serendipity."
Mahowald was
expressing her delight with the similarities
between analog electronics and biological processes. In contrast to the
immediate translation of
numbers
be crunched into an image, the analog and biolog-
to
light signals into digital
ical
imagers shape the light into a cascade of related chemical
and
electrical radiations.
to
They do not cut
off nature
and
resort
mathematical abstractions. They use the amplified signals
from the
light
and allow them
to flow as a pattern across the
neural plane.
The ican,
result, as
Mead and Mahowald
was huge energy
wrote in Scientific Amer-
efficiency: "In digital systems, data
and
computational operations must be converted into binary code, a process that requires about 10,000 digital voltage changes per operation. Analog devices carry out the
step
and so decrease the power consumption of
by a factor of about 10,000.
They pointed that
same operation
would prove
"They respond
in
one
silicon circuits
." .
.
to a little recognized virtue of
pivotal in the
analog devices
development of future imagers:
to differences in signal
amplitude rather than to
absolute signal levels, thus largely eliminating the need for pre-
1
George Gilder
82
cise calibration.
.
." .
As Misha pointed out
only sees changes, or by changing. objects or
sees
it
still
movement
is
movement
treat
moving
movement
is
frozen in one place,
as noise,
becomes
it
of
While
itself.
the key signal and guide for a biological imager.
the eyeball, for example, It
down and
her book, the eye
detects the
It
objects by constantly
imagers are bolted
digital
in
If
blind.
moving and focuses on movement. Vision machines
sees by
that try to
remove the noise of movement defy the lessons of
nature.
Misha concludes: "Because
[in
Claude Shannon's theory of
information] only changes and differences convey information,
constant change
source of
is
a necessity for neural systems
difficulty, as
it is
for digital systems.
of this venture will give rise to an entirely tion processing that harnesses the
tems
to
digital
All
solve
— .
.
.
rather than a
The success
new view
power of analog
of informa-
collective sys-
problems that are intractable by conventional
methods/' this
was
theory,
however. Regardless of
its
academic
significance as evidence of an advance in understanding the
nature of vision, the cat on the cover
blue
—
actually
was
a
downer
for
—
a
monochrome
most observers.
It
blur in
belied both
the confident assertions inside and the grandiose claims of
the Synaptics's business plan. Captured in only 2,500 pixels, the
image seemed
to
pose no significant threat to the Moore's
juggernaut of digital electronics based on devices
machine
that
was already propelling
a
charge-coupled
thriving
vision for manufacturing applications.
"bucket brigade"
CCD
Law
industry of
With
a single
chip holding millions of pixels,
many
The
company
laboratories
Silicon
Eye
183
were experimenting with
that offered resolutions far higher than Mead's.
digital
cameras
Few were awed
by his claim that he could scale his device to densities a hundredfold greater than the early rendition.
chrome
pixels scarcely
Some 250,000 mono-
endangered Kodak or Sony.
17 Teleology at Oxford
After Misha's
Scientific
American breakthrough and her
where she would
thesis triumph, she left for Oxford,
confront the world of biology in
its
own
Unlike
lair.
Carverland, congested with computers and oscilloscopes and other scientific instruments, the Oxford lab was filled with
books and prowled by
real cats.
Her
colleagues,
Rodney Douglas
and Kevan Martin, though pursuing the neuromorphic were nervous about
As Misha put goal.
They
it
its
implicit critique of biological practices.
at the time: "Biologists don't
call that 'teleology'
measure everything
vision,
and think
it's
want
bad.
to
have a
They want
objectively, all the infinite details,
to
without
understanding the purpose. But in order to understand something,
you have
to take things out. If
cat, here's a cat"
— mind
"it's all
—
a black feline
there. Perfect."
"But to understand a
judgments. Have a
cat,
you want
to
understand a
jumps onto the
lap of her
She holds out her cupped hands.
you must reduce
goal.
185
it
somehow. Make
George Gilder
186
"Making models of the
almost everything out that the biologists think out the chemistry, the In
making
DNA,
a silicon model,
we have
retina with Carver,
we
take out everything but a few elec-
—amazingly the same —
same
the biological properties and signal paths. But salt
'The
water
it
important. Take
is
the shape, the salt water, the lipids.
tronic properties that look the
chip in
to take
reacts differently.
if
dump
you
as
the
will short out.
It
do the same thing every time they do a meas-
biologists
urement. They take everything out they don't measure. But they don't admit
it.
So they never develop a clear agenda and a
set of
goals for their research." Biologists at Oxford that year in fact
were studying the neurons of
live cats
by probing them with a
sensor that could measure their electrical activity on an oscilloscope. But these crude measures yielded
and were very bad
for the cats.
understanding
little
Misha conceived
vision as a
com-
plex collective feat of billions of neurons, rendering the blips of
one neuron essentially meaningless. But
at
times she thought
that even this bootless torture of cats offered
the brain than her
own
efforts to create
During the long winter league ogist life
Rodney Douglas,
a genial
moved
my
work,'' she said.
bogged down and other
in
silicon.
with her col-
and accomplished English
and neuromorph, recently divorced.
from
insight into
neurodes on
Oxford, she
in
more
"I
cannot separate
biologists at
Oxford outside their
little
when ETH
Zurich established and endowed an entire
interest or support.
institute to support Misha's
new
building.
It is
my
As time passed, though, the work
group showed in
biol-
The team was
work and housed
it
in a
little
exultant
new
spanking
called the Institute for Neuroinformatics.
The
Silicon
Eye
187
1995, Misha and most of the group
In
moved
to Zurich,
including Douglas, Martin, and a team of ten scientists. Their goal
is
make to
familiar: "to identify the
computational principles that
the brain so formidably versatile and powerful, and attempt
embody them
into a
new and
innovative type of computer
architecture." Their chief achievement
is
the "canonical cortical
microcircuit," a kind of minimal electronic neurode that theoretically
could be ganged into groups to perform neuromorphic
computation.
meanwhile back
in
San
Jose,
Mead was
pursuing his
imager project in a corner of the Synaptics Lab. In
late
1994,
impressed by claims of progress on the imagers, National Semi-
— revenues — made conductor
a Silicon Valley
a bid to
behemoth with some $2
billion in
buy Synaptics. The purchase did not
go through because of resistance on the Synaptics side, but National
made
a $5 million investment.
Sanquini on Synaptics a joint nies.
development
To Mead,
s
put
its
COO
project, to
be negotiated by the two compa-
this proposal offered the
development be devoted
appealing chance to
He
fab.
suggested that the
to imagers.
In charge of the joint venture,
Mead
hired Tobi Delbruck to
build the imagers and National assigned two engineers.
up
a lab at Synaptics
Dick
board of directors and agreed to pursue
have ready access to a leading-edge joint
It
They
set
where they contrived small cameras based
on imaging chips that
Mead and Delbruck had
developed.
Carver began having fun again because "we were doing fun
George Gilder
188
stuff."
But as time passed, he discovered that despite weekly
Fri-
day meetings with National executives and engineers, he lacked leverage with the National fab.
Mead was
near the end of his rope
when Mark
charge of Nationals patents, showed up
meetings
in April 1996. After
and asked him how
He to
said
he
didn't
is
with
know, because
he approached
over,
We
was so hard
way your
will
Mead
at the time.
to get National
replied: "You
know, Carver,
have cross-license agreements
the other semiconductor companies.
come up with on imagers only
it
Grant
silicon.
not really the issue.
all
was
one of the Friday
was going. Mead was seething
it
put anything into
that
it
at
Grant, in
be available to
ideas will have any value
is if
Any
patents you
The
the others.
all
they are spun out
into a separate company."
Mead was son
why
exultant.
Grant had given him a perfectly valid
National would benefit
if
the imager project
rea-
became
independent. The company would not want any of the newintellectual property being
away
to Intel
accumulated on Zanker Road
to slip
and Sony or any other major microchip producer
in
the world.
The group was sensors.
already developing the
first
CMOS
imaging
Using Mead's invention based on bipolar latch-up
sistors that
could be
made
into photodetectors in an ordinary
icon fab, they could achieve densities
advance of
CMOS
tran-
technology.
sil-
bounded only by the
Doubling transistor densities
every eighteen months under Moore's Law, the process was already scheduled to produce a billion-transistor
the next century.
DRAM early in
The
Next
to
Eye
189
be hired, in August 1996, was Nick Mascarenhas.
particle physicist
he had devoted trinos,
Silicon
who had joined Carver
much
A
as a postdoctoral fellow,
of his academic career to detecting neu-
massless particles that pervade the universe but interact
with almost nothing, making them incredibly hard to catch.
If
he
could detect neutrinos, photons should be a snap.
from the outset
of the
new independent imager
Carver upheld two key principles.
He knew
could not succeed making imagers alone.
ence had confirmed
whims to
company
The Verifone
his resolution to avoid
of a large company.
that the
project,
experi-
dependence on the
The imager joint venture would have
produce entire cameras, from shutter
to software.
Even
if
they
never marketed a camera, they would need to build one in order to prove the capabilities of their imager.
was
to create a color
camera without
The second
arrays of filters
resolution
each dedi-
cated to just one color. Eschewing digital guesswork, the
camera would have
One day
in
new
to register all the colors at every pixel.
September 1996 Mead showed up
an old tube-based video camera.
It
used a prism
at the lab
with
to separate the
three colors of light for processing into a full-color image by
three tubes. Carver suggested that they extract the tubes from the prism element and do a color imager with three imager chips,
one
for
each
cumbersome, but all
the light from
color,
it
all
attached to the prism. This design was
possessed the supreme virtue of collecting three primary colors at every pixel.
By Jan-
uary 1997, Tobi Delbriick completed a working prototype, based
George Gilder
190
on chips manufactured than two months.
It
was the
first
camera they made.
clean and vivid images of 640 by 480
cheap PC, or some 300,000
With in
a tiny
took him a
at National. It
pixels. It
VGA
was
more
little It
produced
resolution, like a
a start.
team and few resources, they had demonstrated
two months the
ability to
make
a
camera with images of
higher quality than the prevailing standard in digital photogra-
phy
in 1997.
But they could not proceed further without support
from National. The key event was a demo done by Mead, Toby Delbruck, and Nick Mascarenhas
at
Friday meeting with
a
National in April 1997. Assembled in a National boardroom
were
CEO
Brian Halla, Dick Sanquini, and technical
including an imager and wafer fab expert
named Dick
staff,
Merrill.
Tension was high. Delbruck and Mascarenhas had never seen
Mead
so nervy
and focused. "Assembled with duct
device consisted of a cuit board
and attached
Mead
set up,
told Tobi
The National looked
bad
at
TV
camera lens mounted on
to a
executives
themselves
all
live in
sharply, "Don't
quiet.
touch
it
was
it."
paraded by the machine and
VGA
color.
pixels, the device offered a clean
uration.
a single cir-
computer color monitor. After
and Nick
the
tape,
With few defects
image with good color
or
sat-
Everyone seemed impressed, although Merrill was
The high
resolution,
evident. National
low noise, and high potential were
began exploratory negotiations
for a deal with
Synaptics.
Mead
then met with Sanquini and patent counsel Grant, and
together they
made
company and put
a proposal: National all
their
would
invest in a
imager patents into
it.
new
Synaptics
The
would
Mead
way they add value
summed up
in the lawyers, to
is
hope, Sanquini called:
.
Let's
"Isn't
months
six
to close
they have to add value.
change the deal
does not work anymore/' But just as
communicating?
and
the deal that was finally
"The process took
recalls:
because once you get
.
191
namely Mead, Delbriick, and Mascaren-
Although those terms
negotiated,
.
Eye
also put in their intellectual property in imaging
assign key personnel, has.
Silicon
in
The
such a way that
Mead was
it
near to giving up
we
the problem that
are just not
have a another meeting, just you and
me
and the key lawyers, of course." Mead gulped ("key lawyers,
of course"), but he actually exactly
which lawyer
to bring.
Sanquini came with
Mead
was pleased, because he knew
Mark
Clark, the National counsel, and
Law Group,
brought Craig Johnson of the Venture
leading entrepreneurial lawyer in Silicon Valley,
man
of Garage Technology Ventures. Johnson
is
now
the
also chair-
known through-
out the Valley as a wizard at structuring win-win technology deals.
As they began hammering out the agreement, he began
talking reflectively about the structure of arrangements
seen work and those that out
is
neers
that the big
room
for
guidance, they
didn't.
"The key
for
any good
he had
to
come
companies leave the entrepreneurs and engi-
independence." With Johnson's authoritative
made
Mead and
a deal that gave
his
team the
space to innovate, and National and Synaptics a bountiful share of the upside. At the time National held
49 percent of the equity
and Synaptics 16 percent. Before launching the
new company,
however,
sure of one thing. There were two people he
Mead had
to
be
knew he would
George Gilder
192
have to sign up.
and teacher,
One was
at the
Richard Lyon, former Caltech student
time resting on a gilded six-month severance
package from Apple. The other was Nationals leading inventor of imagers and silicon processes, Dick Merrill. Merrill
suspicious of
Mead
understand quite
as a pointy-headed
how
side the hothouse
still
was
academic who did not
to design a device that
worked well out-
environment of a Caltech
laboratory.
But
Mead was
quick to recognize Merrill as "the most creative engi-
neer that
had ever met."
I
"He was extremely the visit
elusive, though,"
Mead
said.
company could be consummated, Mead would Misha
And first
before
have to
in Switzerland.
J
18 White Rabbit Sometimes
I
doubt the
consciousness.
reality of a separate
—Misha Mahowald
September 1996, Carver Mead got
Inwas
in serious trouble in Zurich.
to focus
Misha
She had moved out of Rod-
ney Douglas's apartment and was
been able
a report that
She had not
living alone.
on her work. She needed "space/' She needed
"time."
As everyone it is
discovers,
a hard problem.
step, like
bionic
murky
when
they take brain science seriously,
Mead had modest
goals of
moving step-by-
an engineer. Misha was more impatient. She wanted
breakthroughs,
psychedelic
Beseiged by
revelations.
data, dendritic loops within loops, neuronal
mazes within
mazes, a susurrus of analog noise, a buzz of bugs and
flies,
howls
of dying cats bristling with metal probes, she felt trapped.
Seeking the doors of perception, she kept hitting the walls of recursion. In confinement, no system can be fully understood.
No
logical
scheme, mathematical code, or algorithm can be
proven without resort to premises beyond defined the problem.
itself.
Kurt Godel had
Max Delbruck had couched
J
93
it
in his
Mun-
George Gilder
194
chausen epigram. Misha resorted mistic prayer.
She even returned
She resorted
to yoga. fitfully,
but recurrently, to the
forms and symbols of her original Catholic to extract her
mind from the swamp by
to ani-
faith.
She struggled
pulling on her
own
long
braid.
Like so ley to
many
Timothy
brain researchers before her, from Aldous Hux-
Leary, she
had sought
in her desperation a facile
Disney-like epiphany, an artificial ride into the mind, and out into a higher domain.
and
THC —gave
At
first
the drugs
her a sense of
new
—
particularly the
insight
LSD
and power. But over
the years they began to evoke frightening specters beyond her control.
She then halted drug use
enhanced form of
ing even an for a
entirely, to the extent of refus-
aspirin offered her
by Feinstein
headache. But without drugs she found herself prone to
depression. She had wanted to transcend her limits, enter a universal consciousness
own mind from phantoms
fly,
Mead and
beyond the cerebral
down on
her
a universal eye. But there, knowledge stops,
and demons his
trap, look
lurk.
Checkmate.
companion, Barbara, flew
to Switzerland to see
from
her. Arriving at
her apartment with flowers after a
California, they
were greeted by Misha with a strangely cold and
flight
she were not really there and they were
dismissive shrug, as
if
not really there, as
the world had receded and she were a waif,
diminishing
down
if
a long tunnel,
all
Disney magic gone, and they
were inspecting her from a microscope above. They did
their
best to reach her, without success. In
late
December, Misha summoned David Feinstein
to
Geneva. Her old Caltech lover and most enduring friend and
The
Silicon
confidante, he had been the one
map and
nal to
grasped
its
Eye
who
promise.
It
first
195
contemplated her
reti-
was he who introduced her
Carver Mead. Although in 1992 she and David had amicably
broken up to pursue their divergent careers, they had continued to talk
Misha
weekly on the phone, Feinstein in
at
Boeing
in Seattle,
Pasadena and Oxford. Feinstein came now
to try a
rescue.
Misha wanted him
on Christmas morning, and so he
to arrive
did.
There were
ing.
Churches beckoning. Swiss burghers thronged the
amid
St.
Nicholas decorations everywhere. Bells ring-
the lights and consolations of the ordinary
all
Misha had once wanted:
streets
love, family, children, revelation,
She greeted him bearing
a
King James Bible and
that
life
God.
restlessly
shuffling rosary beads. Pointing to scripture, she told
him she
was being invaded by creatures crawling up from the base of her spine.
They were beginning
on her mind. They were
to prey
relentless. For distraction, she insisted
by a friend from the
carol party given all
in
German. She did not seem
her. Feinstein called
that this
Rodney Douglas
had happened before
effect in six hours
to
—
on attending institute.
okay.
Feinstein lay sleeping in her living room,
Feinstein was
He
Panic seized
assured him
that her medication
and she would be
Ten hours
Misha
Christmas
The songs were
know anyone. for advice.
a
would take later,
while
left quietly.
awakened the next morning by policemen
knocking on the door. Surveying the apartment, they opened the
bathroom floor ful
was
door. a
The tub was
ruddy kitchen
full
and red with blood.
knife. After
On
the
an apparently unsuccess-
attempt to cut her spinal chord, Misha had
left
the apart-
George Gilder
196
jumped
merit for a nearby train station. She to
Geneva. She was thirty-three years
feinstein wrote to Misha's
old.
friends:
myself exclaiming, 'How could you do really a mystery.
chosis,
eternally
if
damned, whereas
however she conceived
scripture,
youth
—
Misha?' But
it
isn't
I
if
if
a psy-
hope she found redemp-
it."
—contrived from from her
figmentary recollections
she died before Christmas ended at the place of
her birth in Minnesota, seven hours behind, she could find vation.
to
own
she died by her
told Feinstein of her conviction
imagination,
that
find
madness triumphed, then she
hand, then she might be redeemed.
Misha had
I
madness was coming
to believe that
take her. She explained that
tion,
this,
"Sometimes
She was beset by demons within. She had
and had come
would be
in front of the train
sal-
Otherwise the demons would capture her mind and she
would be
lost to
madness and
perdition.
She did
arrive at the sta-
tion before six o'clock.
Feinstein continued:
"I
find Misha's story eerily like the story
7
of Carl Jung. However, Jung's psychotic period in his thirties
ended
finally,
releasing and
empowering him
do
to
his
life's
work.
"Misha was more connected than anyone
I
to the collective
subconscious
have ever known or known about, save perhaps
Jung himself. That connection intensified abruptly during her 'long gray winter' in
ued
to
Oxford two and
a half years ago
and contin-
deepen, increasingly threatening her sense of herself.
My
The
understanding of her death her
when
Silicon
is
Eye
197
that suicide
became
possible for
her fear of loss of self surpassed her fear of death."
In 2000,
Tim Allen went
to visit his old friend
Tom Tucker. At
Caltech, Tucker was the incandescent one, as intellectually sparkling as
was Misha, with
whom
he tested the doors of per-
ception and the elasticities of the mind. Allen found Tucker ing on his mother's
expression and
couch
in Littleton, Colorado.
for
leader on neural networks and then start-up called It
seven years as a technical
became
chief scientist at a
MicroMonitors that created liquid
crystal dis-
Now
failed during the millennial technology crash.
white house propped up on the side of the
lives in a
social worker, a
huge dog, and several
mathematics as a volunteer
cats.
to
On
he
over-
hill
looking Portland, Oregon, with a no-nonsense young
teaches
a blank
little to say.
feinstein worked at Boeing
plays.
He had
liv-
woman
Thursdays, he
troubled youths in
prisons.
Through
his
company, Analytic Animations, he consults on
the intractable problems of applied mathematics with compa-
when some
scholars write of the
assert that all the major
problems are solved,
nies around the globe. At a time
end of science and the frontiers of sides.
He
human
learning
still
beckon
to Feinstein
on
believes that "the horizons of knowledge are no
than a week away. Within a
week
a smart researcher will reach
its
all
more
of deeply delving into any field limits
and begin asking ques-
tions that the leading thinkers in the field cannot answer."
George Gilder
198
Delivering "Feinsteins" of ever-increasing density and elo-
quence, he gives mesmerizing sermons before large crowds local
churches on the analogies between the findings of physics
and the nature of the
down and limits).
in
at
soul.
"Waves are neither
local (pin
them
they disappear) nor infinite (to propagate they need
But
all
the action happens at infinity."
some way Misha's waves flow on
He
believes that
still.
Carver Mead, meanwhile, never solved the perplexities of the
mind. But he did find a project
man who
—however transformed—
could bring Misha's retina
to a practical climax.
Part Four
FOVEON
—
19 Merrills Magic
Walking through
the Foveon Corporation parking lot
with Dick Merrill on a cool February day in 2001, did not
know what
to expect. Merrill is
I
an analog
chip designer, and analog people are different from digital people,
who
favor neat streams of ones
and earnestly entertain physical layer,
journalists.
where everything
skewed and connected
in
Analog people work
teams at the
noisy and intimate
and
to everything else in the universe.
This
sense of exposure and connection firewalls
and zeros and work
is
—
of a lack of thresholds and
and the clear numerical boundaries of digital systems
makes analog people defensive. They often become of black arts
and trade
secrets, volatile
loners, full
resentments and spiky
prejudices.
One
of the greatest,
worked
some
twenty-five years for Analog Devices of Nor-
for
wood, Massachusetts, from
Mead had met Lion
a
named
Barrie Gilbert,
cabin somewhere in Oregon.
with Merrill four years before, in 1997,
& Compass
has
at the
Restaurant, not far from Foveon's offices, in an
George Gilder
202
effort to enlist this silicon
manufacturing sage
start-up.
Mead was
industry,
and they were meeting
Valley.
in his
new imager
probably the most important intellect in the in
one of the
elite eateries in
the
But Merrill had given the impression that he would rather
be chomping on a ham-and-cheese sandwich in a cabin in
Vermont.
for several years, Mead had urged me rill,
whom
to
meet with Mer-
he compares with the industry's most famously
feral
analog chip prodigy, the late legendary Robert Widlar. This was less
than reassuring. As
researched Microcosm,
I
semiconductor industry,
for
Widlar had led
a hero,
Lawrence Station
which
me
on a
Sunnyvale
in
I
my book on
the
merely wished to make him
futile
wild-geek chase from
to the Hilton in
Puerto Vallarta,
Mexico, without yielding a single interview.
On
this bright Silicon Valley
intense,
and
quiet, as
But Merrill was or shoot
it,
as
if
noon, Merrill seemed
shifty,
he was planning a Widlaresque getaway.
at least unlikely to
be drunk, or
Widlar had been known
to flaunt a gun,
to do, in efforts to attract
the attention of his National Semiconductor bosses ("Always
worked," as he
said).
And
the trip to the Lion
be short. Merrill would not
likely
escape on the way. Mapquest
estimated a seven-minute jaunt, just one exit
which was only
a
& Compass would down Route
101,
few hundred yards up San Tomas Expressway
from Foveon. But Merrill had a shortcut. I
steered
parking
lot,
my
rental car, an ungainly gray Lincoln, out of the
squeezed across a small bridge and into a series of
adjacent parking
lots,
snaking past the Intel Santa Clara plant
The
and companies with names such onto what
Eye
Silicon
203
as Swixx, Svee,
and Luminant,
thought was Central Expressway. Merrill was mut-
I
and assured us
tering directions
would be no time
it
at all,
what
with the short cut, avoiding the jams and terrors of 101. Chiefly
new
engaging Merrill, though, was his denunciation of the
Nikon D-l
digital
Nikon on
ital
camera.
Using standard
said Merrill, the
camera
Nikon was
know you have
Mekong
this
CCDs
River with
photography
with
all
lec-
their noise,
horribly flawed, inflicting ghastly
where there should have been whorls, and
forcing delays and wasting
only
the
was fascinated by
I
digital
whirls on the pictures
taken the top-of-the-line dig-
down
a recent trip to Asia
his Laotian wife, Sang. ture.
He had
power (no ready plugs
a real
in Laos). "You
camera when you want
to take
it
mountain climbing," he explained. After several turns, in the midst of a shameful capitulation
Route
toward
announced
that
from
101
Lawrence
we had missed
Expressway,
Merrill
the key turnoff and overshot the
mark. Turning onto a backstreet by a company sign with three Xs in
it
—whether adult
or
nervously that perhaps
pre-IPO
I
could not descry
we should
much lunch
the Lion just off total of
anyway." But Merrill
& Compass
any
moment
a
Nikon
digital
Foveon had arrived
By
thought
we
"I
don't
could find
now, somewhere across 101 to thrash
around
for a
twenty-five minutes.
By the time we reached the buy
suggested
reassured him.
I
still
Magdalena, and we continued
some
I
return to Foveon and eat in
the cafeteria. "That will be fine for me," eat
—
his shifty
restaurant,
I
camera, but Merrill had at its
knew enough still
not to
to explain
how
miraculous alternative.
manner, Merrill conveyed the impression that
George Gilder
204
he
really
wished he could return questions
intimate
distractingly
on
arrival at
work and escape
all
these
about semiconductor P/N
and charge-coupled devices. But
(positive/negative) junctions right away,
to
the restaurant, he clarified Foveon's
route to a revolution in image catchers.
He
said that our cir-
cuitous path to lunch offered a perfect analogy for the road that he,
Mead, and
new imaging
his colleagues
had followed on
their
way
to their
product.
Foveon and Synaptics had indeed traversed a twisting and swiveling route.
They narrowly missed
a possibly fatal collision
with digital camera monster Sony. They overshot key turnoffs
toward simple solutions. They pursued lengthy roundabout paths and kludges devoted to complex three-chip camera sys-
tems with exquisite precision optics and color supplied by epoxied prisms,
handcrafted, during the course of which the
all
embattled team frequently debated whether they wanted to go to the destination
anyway. Perhaps Foveon should have merely
exploited the niche market of professional cameras that cost
more than $70 thousand apiece.
amount would
Maybe
curve.
few thousand Bell.
in
.
still
.
in
a
be cheap tuition for a
them
trip
for half that
down
the learning
the end they could lease the things for a
month,
like a T-l line,
and get
as rich as Pacific
.
the late summer
Mead
Selling
of 1997, preparing to launch Foveon,
also set out to recruit
Dick Lyon,
his old student, friend,
teaching assistant, and a camera buff to boot. Inventor of the optical
mouse, master of signal processing, and creator of retinas
_
J
The
and cochleas
who
at
Silicon
Eye
205
Caltech, Lyon was the kind of practical scientist
could bridge the realms of laboratory and production
imager component and camera system.
Mead had
to
line,
have him
aboard. Living on his golden severance from Apple, he readily
succumbed
to
Carvers invitation. More challenging was Merrill,
with his secure and lucrative role at National Semiconductor.
As much
as Merrill valued his National position,
intrigued with the chance to launch
the historic lunch with
Mead
new imagers
at the
a great guy,
himself. But at
& Compass,
Lion
harbored suspicions toward these Caltech types.
was
he was more
he
still
He knew Mead
but he feared that he might have undue pride of
authorship for his inventions and excess loyalty to the Caltech
crowd and
their
sometimes figmentary
ence.
"He wanted
Mead
said.
to
make
sure
artifacts of laboratory sci-
we were
going to do
By the end of the lunch, Mead had persuaded him
that despite forty years teaching physics, computation, trical
engineering at Caltech, he
ing for
mass production
become
right,"
it
He
too.
and
knew something about
elec-
design-
would
also agreed that Merrill
technical leader of the effort.
When
Merrill decided to join,
pany on the road.
On
August
7,
Mead knew he had
his
1997, the imager effort
com-
moved
out of the increasingly cramped and busy Synaptics facility on
Zanker Road and into new offices on Bubb Road Arriving on August
7,
Mead
in Cupertino.
took along Delbruck, Mascarenhas,
and Orion Pritchard from the Synaptics imager project and signed up two more National engineers: later
became Foveon's VP
of engineering,
Rich Turner,
who
and layout master
Milton Dong. Responsible for
all
the manufacturing of chips, Merrill had
George Gilder
2 o6
one major concern as he joined the company, namely those bipo-
between every two
phototransistors. Already latent
lar
in every
CMOS device,
the bipolar transistor photodetector was
supreme elegance and com-
a beautiful idea, with the virtue of
Crammed between
pactness. tors,
enabled dense retinas
it
alone
still
had feature
transistors
CMOS transisCMOS transistors
the two ordinary at a
when
time
sizes far too big for a high-resolution
imager. Pushing the technology for two years, Delbriick
Mascarenhas would
exploit
its
efficiency to create imagers with
a full nine million pixels in 1997. Still
cameras eight years
digital
later, it
design. But, as both Delbriick
cations
it
CMOS
had
beyond the reach of most
was
a dazzling feat of circuit
and Merrill knew,
for future appli-
fatal flaws.
geometries by the mid-1990s had dropped below a
half micron. This itself.
and
was around the wavelength of
With Moore's Law taking care of
density, the
visible light
new
imper-
was ease of manufacture. Creating bipolar phototransistors
ative
with consistent behavior in the midst of an industry standard
CMOS
process was going to be a bear, Merrill thought. At the
least,
would require
it
fab teams
A lag.
would
risky process
changes that the National
resist.
more crippling problem of the bipolar Because
worked
it
was part of the
like the eye: It
CMOS
had no hard
transistor
was image
pixel, the bipolar
reset.
device
This meant that each
subsequent frame was haunted by a brief ghost from the previous one. Though perfectly acceptable for a lag
would be
era that
a
showstopper
Mead had
for the hybrid
still
still
camera, image
and motion cam-
targeted from the beginning.
The
knew
Merrill
Silicon
Eye
207
Meads
that in order to displace
he would have
tricky bipolar transistor,
of a design of his own. Just two
months
elegant but
prove the superiority
to
after the
founding of the
company, he delivered. Merrill came into the weekly Monday
meeting with
test data
from a new imager
—the
3-T
—
that
he
had gotten back from the National fab two weeks before. Based on
be instantly reset after
a photodetector design that could
each exposure,
it
with low noise, no image ities.
"Merrill
up with,"
said
... for the
Mead
showed us Mead.
first
2K by
provided a
"It
lag,
it
2K, four-megapixel imager,
and dual
was
still
and motion capabil-
a lot better than
was time
to bet the
what
I
had come
company on
Merrill
time."
resolved to use the 3-T to
make
a professional high-end
studio camera with a prism that directed light to three separate
image sensors. Saving
two of the
tering out
perform
all
all
the light at every pixel rather than
colors, this
other digital
fil-
system would eventually out-
cameras
in
color
accuracy and
luminosity.
But
this goal
would be achieved only
science from Dick Lyon.
after
some arduous
color
20 Lyon s Den of
At
Many
Colors
the heart of the headquarters of Foveon Corporation
are
two emblematic spaces, reflecting the two different
paths to the future pursued by the company's chief technologist,
Dick
Merrill,
and
Dick Merrills space evokes ing in the middle test gear that
is
the
chief scientist, Dick Lyon.
its
his time-travel trappings.
aluminum
he attaches
like
chassis for the rasterfarian
an electrified dreadlock wig of
wires and lasers to the pins and sockets of his
chips
—applying merciless
Gleam-
little
imager
tortures until they confess every alias
and aberration, pixel-leak and pico-capacitance, of their photoelectric behavior.
(Hey,
it
drift
matters
all
and
—
jitter
this
is
analog.)
Dick Lyon's space, by
dim corner of the second lamps, photographic
contrast, floor,
it
is
a library.
offers plush chairs, reading
relics, esoteric slide rules,
209
Occupying one
and archaic cam-
George Gilder
210
eras surrounded
books
—shiny
fraying backs,
on four sides with a
new books
and color images. The
full
range from canonical works by
titles
Graham (RWG,
William
The
Reproduction of Color.
hundred
of black-and-white
James Clerk Maxwell and Hermann Helmholtz tions of Robert
five
with jackets, moldy old books with
photography catalogs
tall
some
total of
some
to
to you)
five edi-
Hunt's classic,
shelves bulge with several
MIT
Handbooks of Colorimetry and an array of Kodachrome manuals and stacks of handbooks
for
Kodak Ektachrome development
procedures E-3 through E-6. Discreetly horned into the corner of a bottom
row
is
a
copy of Roundup
that memorializes Lyon's third
1
970, the El Paso yearbook
camera and
his
promise of a stun-
ner of a fourth.
Both of these emblematic displays are dead serious, as
Foveon
itself.
Merrill's ability to extort detailed
is
performance
data from every device has been critical to the fast progress of the
and
company
as
constantly experiments with the
it
most of those books, and
science
—was
designs
he concocts. Lyon, meanwhile, actually read
circuitry that
labyrinthine rune
new
his
knowledge of
and wrinkle
like Fill
—every
in the often cryptic lore of color
crucial to transforming
imager consultant,
their contents
Foveon from an innovative
Factory in Holland, into a possible
supplier of comprehensive camera and imaging solutions, hard-
ware and software,
So
let
treatise
like
Kodak
of old.
us retreat to Lyon's den and curl up with an exciting
on color theory
.
.
.
Technically referring to a set of electromagnetic wavelengths or frequencies, the
word
color tempts physicists
and engineers
to
The
imagine that they understand
Running from
Eye
Silicon
when
it,
211
they have barely a
glint.
infrared through visible light to ultraviolet, with
ever shorter wavelengths and higher frequencies, the spectrum of "colors" climbs a rising slope of energies (the formula for elec-
tromagnetic energy
is
Thus the frequency of
The confusion
the frequency times Planck's constant). a color
is
proportional to
energy.
arises in relation to "brightness''
an image most striking
erty of
its
—the prop-
humans. Frequency only
to
vaguely relates to brightness. In physics, brightness signifies the
number
of photons arriving per second, while frequency
measures the number of waves per second
in
each one of those
photons. All the regnant imagers in the industry
CCDs
and
CMOS
ness, photons per second.
basic physics. it
will
Even
— measure
photo-sensors
if it
But no imager can
all
the
physical brightrestrict itself to
gauges energies as well as intensities,
produce graphic gibberish without any "colors" that
humans
will
To build
deem a
accurate.
useful camera entails entering Mishaland and
acquiring a mortifying knowledge of wetware" '
to
—
—
biology.
You have
master not only the physical spectrum of frequencies but also
the
sensitivity
medium, and
spectrum of
short receptors
"blue," the six million
human
(LMS
eyes.
Divided into long,
curves) for "red," "green," and
cones of the
human
eye will detect zero
"brightness" in radiant infrared, zero "brightness" in sizzling ultraviolet,
and a peak of brightness
at
around the 500-nanometer
wavelength of yellowish green on a forsythia bush. will see different materials as the
same
color in
some
Human lights
eyes
and
different colors in other lights. Things that are objectively
as
and
George Gilder
u2
such as dollar
spectrally different
same
be the
color.
common
anyone who wishes
to
and maple leaves seem
Heck, they are the same
Called metamerism, these
problems
bills
to
to
color.
anomalies pose baffling
reduce photography to mere
photo-detection. Things that look red to the
human do
not
have to look red to the camera. Digital processing can adjust the
hue
as long as the
color.
camera and the eye both
But no amount of
digital
register the
same
processing can correct for an
imager that reports as different two items that the eye sees as the same, or reports as the
The eye and less
same what the eye sees
must
the camera
same way
err in the
an image than a splash of expressionist
At
its
as different.
or
you have
art.
mad
photoreceptors, an eye or an imager receives a
lightspeed rush of unfiltered frequencies, reflected
through, and diffracted around objects in
its
off,
refracted
ken, traveling
diverse distances through the atmosphere in different phases,
illuminated by sunlight, fluorescence, tungsten bulbs, lasers, stars,
and candles and bearing no obvious signs or signatures.
Taking
this nearly infinite
radiation ors
is
the crucial
"How all
and reducing
time.
first
that step
You
MIPS and
can't
all
is
it
dimensional space of electromagnetic
to the three
dimensions of primary
col-
step in designing an imager.
performed," as Lyon observes, "fixes
undo
that.
the king's
You
can't get
DSPs cannot
it
it
for
back." All the king's
retrieve the colors in a
scene once they have been inaccurately represented. Then once
you have the primary colors, you have play technology and to
human
to
adapt them to your
dis-
spectral sensitivities by applying
the right coefficients in a process called matrixing.
When
Mer-
The
rill
invents a
new
Silicon
Eye
213
device for registering pixel values, Lyon must
then do the work to adapt them to humans. At Foveon, this translation did not always go well.
arriving at the new Pritchard, on
what
August
offices with
1997, Lyon had no clear guidance on
7,
his role as chief scientist
seemed
to
mean
Mead and Orion
Carver
would
be.
At
janitorial aide, setting
first
"chief scientist"
up desks and
chairs,
computers and white boards. After the setup was done, Lyon
poked around awhile, looking rance behind
On
first
all
for dusty areas in
shadows of igno-
the shiny Ph.D.s.
would
glance, knowledge abounded. At Foveon there
soon gather experts on the neutrino, authorities on retinal design, masters of microchip layout
Carver
Mead
and
fabrication,
and image representation
circuitry,
to the esthetics of power-line insula-
the enigmas of neuronal and dendritic processing, the
topologies of at his
in
himself there was a polymathic repository of infor-
mation on everything from quantum tunneling, VLSI
tors,
and
computer systems, and the husbandry of hazelnuts
farm in Oregon.
One
thing he didn't
know much about
was color science, beyond the theory of the physics. But he
nudged Lyon toward the canonical book by Reproduction of Color, saying:
'Why don't you
soon figured out that the path through the thickets of
As he began, the
this
first
to his fourth
R.
W. G. Hunt,
look at this?" Lyon
camera would run
science of color.
challenge was the endless prism
Like everything that happened to Foveon,
it
was nearly
crisis.
fatal
and
— George Gilder
214
nicely fortuitous.
turned out that projection prisms of the sort
It
that Carver brought to the
ductor were
company from National Semicon-
but useless for color accuracy. As Lyon shortly
all
discovered in the pages of R.
W. G. Hunt, prism
metrical. If a prism projects accurate colors,
torted colors. Inc.
(OCLI)
But their supplier
—was
first
Why
the leading optical modules
JDSU,
would anyone want
a receiver
in
prism for
making
visible
that.
Delbriick tore apart a camcorder and extracted a solid-
prism from
it.
But as the supplier of that prism, Collimated
Holes Corp., pointed out with alarm, 'That patented by that
to sepa-
wavelengths in wavelength division multiplexing
anyway? They weren't interested
Then state
cathode ray tube in billions of computer
optical systems for
company. colors
— Optical Coatings Laboratory,
and then with the better idea of using prisms
rate infrared
(WDM)
will receive dis-
entranced with the idea of using projec-
tion prisms to usurp the
displays
it
colors are asym-
IBM
IBM
—
can't use that."
had patented
bar Foveon from using
it
it
is
an
IBM
prism
Lyon investigated and found
for projection only,
as a receiver.
which would not
So they proceeded. In the
end, the Foveon prisms would evolve into "a color-separating, beam-splitting prism assembly"
made
of five pieces of glass,
including two "dichroic" mirrors that divide light into two frequencies.
The contraption could be
help of a glass rated into a
company
in Russia
fabricated only with the
and then could be incorpo-
camera only through an exquisitely exacting
involving epoxy and
craft
picomotors that could pick and place
imagers with submicron precision in three directions for gluing
onto the prisms.
The
When light that
all
that
Silicon
Eye
work was done, you
215
had three streams of
still
could be rendered as fully authentic colors only with
the help of ingenious software from Lyon that merged a glint of violet into the red.
That was a patent, and
ing color accuracy, but
made
it
it
led to world-beat-
the manufacturing challenge
even harder. Any error got manifested as a purple end, they achieved
some
of the
fringe. In the
most accurate colors and most
—
expensive cameras in the industry
great cameras, endorsed by
Hasselblad, buying in and briefly branding them. But for two years Hasselblad, despite increasingly harried visits from
could not get itself,
its
lenses in line.
Making
a
Mead,
few hundred units
Foveon became an exotic niche supplier of $50,000
cameras, which was not what
Mead had
in
mind
as a serious
market. Beyond the near-perfect colors, the chief appeal of the
instrument was to studio photographers
who
could
triple their
output by avoiding the time-consuming hassle of Polaroid
first
takes.
While Lyon juggled the hues of the prism camera, he contin-
ued
push on through
to
books of color science. piles of
his reading agenda.
He
He
scoured the
pored through the ever-increasing
One
proposed inventions and other ideas from Merrill.
day he stumbled onto Merrills old concept of a vertical color ter
based on the wavelength-dependent absorbency of
didn't
know about
this,"
he told Merrill. "Will
it
fil-
silicon. "I
work
for a
camera?" "I
doubt
it,"
Merrill said.
"It is
hard to make good colors out of together.
Ask
Carver."
a real physical effect, but it.
The
colors tend to
it is
merge
George Gilder
216
Lyon went
to
Mead, who
referred
him
paper written by
to a
Delbriick reporting on their previous experiments with the idea.
The
results
were
lousy.
The
Lyon was discouraged.
seemed
colors
smeared
Beyond the
into a delta of
prisms,
exotic
be running out of options. Would
to
mud.
Foveon
be another
this
—ingenious analog technology
episode from the Synaptics story
ultimately dwindling into an academic stunt?
Then Lyon began
He
more
closely.
depth
at a time.
to scrutinize
Meads and
Delbriick's data
noticed that they had tested the colors one
The
charge from nearly
result
all
was that the junction would
of whatever color light
collect
was focused on
it.
But Lyon speculated that the results might be different using a stack of three junctions at once. three junctions should divide
it
Competing
for the light, the
in accord with the principle of
wavelength-dependent absorption depth. More of the highenergy blue should gather in the junction closest to the surface, while the green and red would sink a micron or so down.
asked Merrill to test
this proposition.
In a process that took six weeks, Merrill
workable
test pixel
inserted
it
murky,
but
and run
into his
the
Delbruck
it
it
pattern
offered
results. Merrill
managed
to design a
through the National fab. Then he
torture chamber.
telltale
detectable. At least
He
The outcome was
of absorption
still
depths was
more promise than the Mead-
thought he might be able to tweak
it
to
separate the colors more.
Before Merrill accomplished the superior separation, Lyon resorted to his
data
—the
Mathworks program
to
combine the relevant
three stacked junctions in Merrill's pixel and the spec-
The
tral sensitivities
of actual
Silicon
human
Eye
217
eyes. Correlating the differen-
tial
capture of the three electronic junctions with the differen-
tial
responses of the long, medium, and short cones in the retina,
he noted that the error curves
—the
difference
—was below 10 percent. The
Merrill
between the two mix and the
retinal
mix were both messy, but they were the same essential mess.
The neat Kodachrome tate
were nice
separations that they were trying to imi-
for projections
—and the
slides
—but
Lyon
recalls:
"Everyone looked
can't
make
camera with
said
strikingly different
a
rich saturated colors of
from the curves of human
that.'
I
prism.
I
could
make
a better
said, 'You
looked at Merrill's results and
could make a better camera with that than
I
and
at Merrill's results
vision.
I
could with a
camera with that than
I
could
with film." This finding,
if
true,
promised a supreme fulfillment of
Mead's career-long romance with
silicon
con had become the perfect substrate
and
its
for
computation
properties.
Sili-
—and
you got the PC. Silicon had become the perfect channel light,
to
and you got
fiber optics.
Now
for
Lyon was opening the way
an ideal silicon single-chip imager. Using the physics of the
material,
form
all
formed
one
silicon sliver
some
five
microns thick could per-
the feats of photodetection and color separation per-
in other
systems by at least three chips plus millions of
tiny color filters that
blocked out two thirds of the
pixel, plus a "blur" filter to
smear out borderline
light at every
errors, all
cou-
pled with digital signal processors to retrieve the lost colors by
guesswork. Lyon graphed his results and showed them to Mead. For the
first
time,
Mead
believed that Foveon was near a single-
George Gilder
2i8
chip color imager.
Then
The
silicon
camera was declared
March 2000, Foveon received
in
feasible.
from on high.
a call
Kodak
Intrigued by the color accuracy of the prism camera,
Rochester
summoned
New
of upstate
moment
the Foveon team to the bleak gray reaches
York to discuss
they had been waiting
ing chief Rich Turner flew east
Mead
technology,
said they
showed them the new and promised
to
for.
its
innovations.
make
now," he said. "This
is
had
and arrived
much
pany
will
will license
late," said
Foveon's
it
The Kodak
when you can make
it."
Mead. "Some other com-
Kodak was unimpressed by
the meeting deteriorated,
He
working on
the future of photography."
be too
it."
on the way.
better. "Merrill is
pictures like these prism photos with it
of the prism
still-murky colors,
its
people laughed and then said, "Call us back
"But then
was the
in Rochester. After
a still-better idea
Merrill chip, with it
It
Mead, Lyon, and engineer-
Kodak engineers the advantages
explaining to the
in
the threat.
As
Rich Turner offered yet
another of Merrills ideas. National had just completed the conversion of
its
Portland, Maine, fab to the .18 micron industry
state of the art. Extrapolating
how many more
pixels the smaller
geometries would enable him to put on the Foveon black-and-
white imager, Merrill calculated he could pixel device
on one chip,
then maxing out
be interested
at
far
ahead of the
make
a
16-million
rest of the industry
around 4 megapixels. Perhaps Kodak would
in leading the industry to
16-megapixel imaging.
Again, the Rochester colossus was unwilling to buy a license for a technology that did not yet exist.
Foveon's assurance.
But they were impressed by
The
Silicon
When Mead heard the idea, Foveon
itself to
create this
"Why do you want customer need. There
Eye
219
he was also impressed, and urged
awesome 16-megapixel
product.
to
make something with no market? No
is
no camera that can use
it,"
one of the
Foveon marketing executives protested. "That has never stopped us in the past,"
Mead
said. "If
we can
build the world's best photon catcher, the world will beat a path to
our door."
A month
company
the
Mead's decision began
later,
got a call from a
said
portraitists,"
when
Greg Gorman of Los Angeles.
Who's he? was the general response. most acclaimed
to bear fruit,
"Just
one of the world's
Foveon's marketing chief,
Eric Zarakov "He's got great shots of such stars as Leonardo
diCaprio and Sharon Stone .
.
.
.
Richard Gere
.
.
.
Kim Basinger
Drew Barrymore. Heck, Arnold Schwarzenegger. Who do
.
you want?"
An
Gorman was was
.
in black
white),
old friend of Foveon sales chief Larry Stevenson,
"always ready for something
new
—
particularly
and white" (90 percent of his work was
and he had
a
male model ready
in
in
black and
San Francisco.
In the photo journal Eyestorm, Zarakov read that
Gorman's
more informal pictures ranged from "intimate glances Hollywood demimonde let,
to a
if it
sweaty Jeff Koons,
sitting
of the
on the
toi-
flanked by two leather-clad strumpets."
"Just imagine
how
they would look with Foveon," Zarakov
said.
"But what are
we supposed
to
do with
this
guy
this after-
noon?" asked Turner. "He's already seen our prism camera."
"Get on the stick and show him that 16-megapixel camera you
George Gilder
220
were playing with the other
map
of the U.S.
the paper
it's
day.
and zooming
You know, photographing the you could see every
in so
fiber in
printed on."
"Hey, that's not a camera!'' Turner protested with a note of desperation.
'That's just
enhanced through
a
The images
prototype chip.
a
Photoshop hack." But Gorman had
prism camera and found that
it
produced
of noise, "like a perfect negative."
files
are
tried the
wonderfully free
Now he wanted to try this new
Foveon chip he had heard about that could produce the world's highest-resolution black-and-white images. "Well,
he's
When?
already on the plane from
L.A.,"
sales
chief
Stevenson interjected. "From there he goes to his house in Mendocino and then to Europe. This
is
the only day he's got."
"You guys are crazy," said Turner. But he and Dick Lyon rigged
up the 16-megapixel chip on
breadboard and created a crude
a
camera using the chassis from the prism product, with attached on a cardboard cylinder. the door, they tried to interest
prism camera, but he said he
When Gorman
him
knew
new
in
all
about
a filter
appeared
at
refinements of the that.
He'd come
to
see the 16-megapixel black-and-white imager. "We've got a few chips,"
made "I
first
Lyon confessed, "But
a picture with
want
it
isn't in
to use that,"
Gorman
said. "I
want
16-megapixel picture." They set up the
unbuttoned
No one
has ever
it."
traption in the old prism studio. hat,
a camera.
his shirt,
to take the world's
new "camera"
The model donned
a
con-
cowboy
and assumed a brooding pose of a
Rodin "thinker." Gorman was at the electronic viewfinder,
thrilled.
But when Turner looked
he saw nothing there. The scene
The
Silicon
Eye
221
was apparently fuzzed out by infrared radiation from the smoldering model. While sion,
Turner taped a further
which removed more of the the chassis and
looked on with a bemused expres-
Gorman
Gorman
filter
onto the cardboard cylinder,
light. Finally,
with Turner handling
and
triggering the shutter
a strobe light,
they took a series of portraits.
Turning over the raw tions.
files to
Lyon, Turner had low expecta-
But over the next three days, Lyon extracted from the
some 20 megabytes
of
monochrome information one
of Gor-
man's best portraits, a riveting black-and-white image of the model, with amazingly sharp definition and vivid
blew up the image
to a lifesize
ceptible degradation even ity
detail.
Lyon
photo eight feet high with no per-
when viewed from
inches away. Vital-
shone from the model's features, bringing
to
mind
a classic
photo of actor James Dean from the movie Giant and recalling "Pope of Trash" director John Waters is
the only person It
I'd let
was indeed a
photograph
s
comment: "Greg Gorman
my
corpse.''
historic picture, offering
white pixels than ever before in a
digital
denser black-and-
image. Beyond that one
photo extracted by Lyon's alchemy, however, Foveon had nothing
more
to offer
Gorman.
A
cameras, he ended his return to Foveon.
The
fervent devotee of leading-edge trip at his studio in
early years at the
Canon
L.A. and did not
company seemed
to
be
repeating the pattern of the early years at Synaptics: great laboratory stunts ucts.
and
scientific
advances but no marketable prod-
Foveon had no color system beyond the prisms and no
market
for a
16-megapixel black-and-white imager.
Meanwhile, there would surely be
a
market
for a
one-chip
George Gilder
222
color imager. But the effects of the device
improve
it,
and
it
Kodak people had
more work Lyon lavished on the color
from Merrill, the
less
he found he could
was nowhere near good enough. Perhaps the a point.
Lyon had miscalculated the interplay
of the silicon physics with the particular parameters of the
CMOS
fab process. Silicon might
theory, but there
was no
silicon
make
imager in
a great color
semiconductor process that
could yield an effective camera chip. Perhaps that was the reason that Kodak, Fuji, Canon, and Nikon had
all
previously
patented the one-chip vertical imager and then abandoned
The
CMOS
process entailed crucial steps that nullified the
advantages of the material as a vertical per.
in
filter. It
As they approached the key industry
Cologne, Germany, in
late
for
them even than
Then Dick another idea.
Merrill
was
a showstop-
exposition, Photokina,
September 2000, they
color process to sell but prism cameras,
market
it.
still
and there was
had no
a smaller
for exotic black-and-white imagers.
emerged from
his
aluminum cave with
21 Photokina 2000 and Beyond
a
huge industry such
In
that a
new company
like
as cameras, rare are the occasions
Foveon can
arrest the attention of
the relevant world of experts, media, and rival engineers.
Held every two years
in the vast reaches of the
Cologne, Germany, with several
total floor
KolnMesse
in
space 800 meters long and
hundred meters deep, with tens of thousands of
exhibitors
and hundreds of thousands of
thousand from the media, Photokina
is
visitors,
including five
Comdex for
most intense commercial showcase and crucible
in
cameras, the
photography
Every company, every shutterbug and photo-bird on the planet, it
seems,
is
there and on stage, testing the son et lumiere, tout-
ing prototypes and photo-types, waving his arms and amplitudes,
swinging her hips and chips, flashing her bulbs, and zooming his lenses, in front of towering multistoried exhibits
shaped
like
cameras.
As Photokina 2000 approached September
— scheduled
— Foveon had impressed people 223
like
for
late
Greg Gorman
in
as
George Gilder
224
of technical leader with
some kind
its
prismatic contraptions,
but you wouldn't want to commit your career to one of their
cameras. passed,
And Foveon had nothing would be
it
else to
sell.
When
September
two years before the industry would
a full
again gather in such frenzied force.
Nonetheless, just one month before Photokina 2000, while the
company was
filter.
preparing to push
a
his alu-
chassis with yet another concept for a single-chip color
Added
to his original idea
was
a simple but revolutionary
process step entirely compatible with the Still
prisms to a mostly
its
camera market, Dick Merrill emerged from
indifferent
minum
still
company
secret today, the
showstopper inside out.
What had
ble obstacle to a vertical filter
and improvement.
An
new
CMOS
Merrill
process flow.
scheme turned
the
previously been an impossi-
became
a radical simplification
invention as elegant as Carver Mead's
exploitation of the parasitic bipolar transistor as a photodetector,
was
Merrill's idea
still
more commercially promising.
"But you can't do that," said Lyon. "That's crazy.
been able "Don't
done
tell
it
if
Merrill
me what
I
can't do," said Merrill., "I've already
it."
you can," said Lyon.
had made
an existing
change. There were to
"It
sure would work."
a basic invention to render the
CMOS
pany would have
one has
before."
a lot of stuff like
"Well,
in
do
to
No
process.
The output
many imponderables. put
its
X3
possible
of the chip
would
In essense, the
faith in Merrill's
magic
com-
in the fab.
Even some of Foveon's own executives were doubtful. At that time in late 2000,
just
Jim Lau, the hero of Synaptics touch-pad
The
Silicon
Eye
225
new CEO,
replacing
for professional
cameras.
manufacturing in Asia, came aboard as the
Ray de Moulin,
Kodak VP
a former
After a nearly fifty-year career in the industry, de Moulin had
announced
a well-earned retirement.
By following de Moulin,
the master of high-end professional photography, with an expert
on low-end manufacturing, Foveon signaled readiness in all
new
directions.
and make
set to go off
camera.
I
But what directions?
said that
a
Mead
recalled: "Jim
more automated version
by the time you get that
to
—and
amazed
Mead and
that
was
of the prism
happen we
have the vertical colors working." Having seen the chip results
move
to
heard of the Kodak rejection
first
will
messy
— Lau
was
Merrill could see anything there. But for
experienced silicon sages such as
Mead and
prototype was no cause for dismay at
nearly every experiment
Merrill the
murky
all.
in electronics starts within a
shroud of noise, a mass of overlapping curves and spiky anomalies.
Only with long experience can the engineer nurse
signal
— the meaningful
and amplifier ronmental
result
distortions
fuzz.
—from the
in
froth of "dark currents"
and electromagnetic skews and envi-
The prototype
Advanced Techology Group fab duced data
forth the
that Merrill got
from National's
in Santa Clara late in
2000
no obvious way better than the output of
his
pro-
two
previous vertical pixel designs. But looking at the results, apparently a
swamp
Merrill
was pleased. Scrutinizing the mountain ranges of arrayed
of noise
mixed with
curves on the oscilloscope, the
a turbid chromatic overlay,
summer
storms, and electrical
George Gilder
226
turbulence, he could descry a workable retina.
The mountains,
they could be laid low, one at a time; the storms could be tamed, step-by-step;
and the
valleys raised up, oh, so delicately;
and
above the wilderness could arise a rainbow of perfect colors. Merrill
Doing
was a past master his
own
and so was Mead.
at this process,
calculations "so
could grasp what was really
I
new
going on in the physics/' he was also convinced that the
approach would work.
"we bet the company on knows when something off
and doesn't
it
up
the data.
as
the fab and
in his lab to
it
it's
will
have his
Foveon's friends in Rochester.
largest,
CCD
and had incorporated
era called the Pro-Back.
Kodak had
in his bones.
Goes
testers
little test
will
chip. He'll
and come back with
right."
photokina approached,
created a 16-megapixel
It's
"He
refrain.
and then some run of wafers
some twenty
And you know
was Mead's
going to work.
talk to anybody,
come back from hook
is
Merrill"
At
the
big
story
Kodak announced
imager, the world's it
into a
first,
that they
had
best,
and
first,
new $20,000 mosaic cam-
Mead and Lyon
stolen Foveon's thunder. But in fact
in
feared that
Kodak had put
Foveon on the map, by defining the focal feat of the
By the time Photokina opened
came from
year.
Cologne on September 23,
2000, the only actual image from a 16-megapixel chip loomed over the front of the Foveon exhibit area.
It
was Greg Gorman's
The
Silicon
Eye
227
incandescent black-and-white picture of the cowboy model. eight-foot-high image that revealed
nized one foot away,
it
no
announced
visibly
when
distortion
An
scruti-
a major breakthrough
from the tiny new company. Imager experts quickly calculated that Kodak's
CCD
was
a
monster device that would cost some $7,000 apiece, suitable only for the most costly cameras like the Foveon's
duced
in
Pro-Back, while
CMOS chip was one quarter the size and could be provolume
decision to
for
make
under $300. Massively vindicating Meads
the 16-megapixel device was a huge flood of
visitors to Foveon's display area.
Mead had
contrived a gigantic
The people came
game
of mega-bait
and switch.
16-megapixel imager and to be pho-
to see the
tographed by the high-resolution prism cameras, and de Moulin, formerly chief of professional photography at Kodak, was there to push the advantages of prisms to serious inquirer
was steered
Lyon, and Turner pitched a existed chiefly in their
Dick Merrill back
One
Compared
room
comers. But any
in back,
where Mead,
single-chip color imager that
minds and hearts and
in the
hands of
Santa Clara.
to
Canon
was Michihiro Yamaki of Sigma
or Nikon,
Sigma was
a relatively
company with fewer than one thousand employees and
revenues of $140 million.
But the ostensibly tions.
new
of these serious inquirers
in Japan.
small
in
to a
all
still
He wanted
camera company.
to
shy,
It
had previously specialized
in lenses.
dapper Yamaki harbored higher ambi-
break out of Sigma's niche and build a major
When
he heard of the 16-megapixel
imager, he insisted on visiting the Foveon exhibit.
CMOS
George Gilder
128
Conceding camera,
X3
Foveon did not yet possess a 16-megapixel
that
Mead
steered Yamaki into a back
which was even farther from
prototype,
smoke and mirrors and team
in Rochester,
could soon be
it
filter.
first
was
It
immediately saw
Meads
assurance
Before the end of the day,
Mead and
trusted
Yamaki shook hands on an agreement create the
show him the
realization.
his engineers
X3 design and
built.
to
But unlike the Kodak
a sprite of silicon.
Yamaki and
the significance of the that
room
launch a joint
to
effort to
commercial camera based on the Foveon
vertical
Sigma would do the mechanical camera design with expo-
sure and
ISO
and would create the
controls
single-lens-reflex features that allow
you
lenses, with the
preview the image
to
through the same lens that captures the picture. Foveon would
do the imager hardware and software, the readout process, the display preparation, to software
and
file
and the
storage for an entirely
the two companies faced an
down the hall
from
From chip design
file-saving code.
awesome
Merrill's
new
kind of camera,
challenge.
office
Foveon
at
is
what
appears to be a heap of colorful Navajo blankets, with a sheet of glass resting
do you
on the
realize that
hue and
fiber
top.
Only when you
try to
remove the
glass
a photograph, utterly indistinguishable in
it is
from the
pile
below
it.
Fooled by
this device,
you
can understand Merrill's impatience with his mere Nikon in Laos. But this trompe
l'oeil trick,
done with prisms,
is
designed
only to grab your eye, not to grab the entire imager market.
On
the wall as you enter the Foveon building
is
a
more
sig-
,
— The
Silicon
Eye
one with a story that portends
nificant photograph,
it
likely suc-
Taken by Merrill
cess in the company's imperial quest. spring of 2001,
229
in the
depicts a vividly colorful totem pole in Van-
couver, Washington, shot against a perfect blue sky. Sky, Merrill
points out, offers the best visible index of the noise level in an imager. In Merrills picture, the sky
normal course of events
is
impeccably blue. In the
in the silicon business, a
new
chip
based on an original design and a novel manufacturing process goes through scores of iterations before
Many never do. But
designers hoped. the
first
product of the very
was based on the very tion,
from the
first full
first
run
first
this
it
works the way
its
impeccable image was
single-chip camera,
which
itself
chip to emerge, under Merrill's direcat the
National Semiconductor wafer
fabrication plant in Portland, Maine.
Thus, Foveon's revolution begins to take shape. Yes, the totem pole picture
supreme
is
flawless, with exquisitely authentic
resolution.
But more
crucially,
the
is
it
hues and
first
working
photograph made with a single-chip, full-color imager. The
Foveon camera
is
no mere
"digital
camera,"
microprocessors and mirrors and shutters state
it
is
of chips
still
(like
the
and
a fully solid-
machine, based around a single analog chip and
no mechanical paraphernalia, capable both
—
full
human
virtually
retina) of
and moving images.
Single chips have a singular virtue:
manufactured
in
volume
at a cost as
They can eventually be
low as 80 cents a piece
about the price of the packaging. In other words, Foveon's X3, as the marketers have colors at each pixel
dubbed
—
will
it
—
make
for "times three," signifying three
possible throwaway cameras and
George Gilder
23
surveillance devices with a resolution today's
and accuracy approaching
most costly Hasselblad. Merrill simply plugged
new microchip
into a circuit board, installed the board in
his
an old-
fashioned nineteenth-century camera chassis, snapped the picture,
and potentially transfigured an
industry.
Between the invention and the marketplace, however,
lies a
long shadowy slough: a treacherous competitive realm, where
dragons
lurk.
Clayton Christensen, author of
The
Innovator's
portentous challenge to Foveon: "Foveon cannot lose.
"
Glenn Davis
Dilemma,
issues a
make cameras.
It
will
22 Foveon's Disruptive Challenge
Greg
Gorman had
a world
moving
sisted as a
long held out as an advocate of film in
cameras and then long per-
to digital
champion
Hollywood scene garish with
of black-and-white images in a
But
color.
in early
2003, this virtu-
oso photographer finally found an electronic camera that fied his exacting
satis-
demands.
Although confessing that
"I still
love shooting
my own
and-white images on film and then scanning them into
new
his
he
says. "It far
exceeds what film can
ity,
more
"It's
digital,"
fabulously sharp and fast,"
he raves about
camera.
black-
realize. It
color range, smoother transitions. Its
has more tonal-
more
sensitive in
dark scenes and more responsive to different color temperatures.
It
has
16 bits of resolution and creates
files
of 63
megabytes. That's huge."
The challenge
to
Foveon
single-lens reflex called the
mosaic
filters. It is
made
is
that
Gorman's new camera
IDs based on
entirely
CMOS
a
imagers and
by Canon Corporation
233
is
in Japan.
George Gilder
234
With ordinary camera, do the
CMOS performing such feats in a professional CCD champions skulk away into the night?
Hardly. Early in 2004, tronics paragon
Sony Corporation, the consumer
elec-
and champion of CCDs, would emerge with
DSC-F828 Cybershot based on
a
new
its
CCD,
four-color filter
adding emerald to the usual red, green, and blue, to reduce color errors by 50 percent
and enhance reproduction of treacherous
borderline blue-green and red colors where Foveon's technology excels.
With 8 megapixels,
a Karl Zeiss
optical
and the capability of doing either high-resolution
new Sony
quality movies, the It
stills
Framing" for accurately composed photos
mode
when
a
dominant force
in
to sink
soon
using flash, and
ahead
at a
down
an
for
all
below $1,000. is
dominant
in
are both pushing
tremendous pace. With volumes
millions of times greater than Foveon's,
careen
far
cameras, Sony
consumer electronics and CCDs, and they their technologies
"Night-
for black-and-white picture taking in total
announced price of $1,200, sure is
DVD-
or
light,
darkness without flash, using infrared illumination,
Canon
lens,
offers a featuristic photocopia.
provides a hologrammatic laser focus for low
"NightShot"
zoom
Canon and Sony
the learning curve, harvesting gains of scale and
scope, making their leading-edge imagers and the cameras
based on them ever better and cheaper. Carver warriors at Foveon faced the
Mead and
complex business challenge of
attacking an incumbent industrial establishment.
they do
the
How
could
it?
In every
new
era,
companies
prevail
by grasping the
cance of a new regime of abundances and
scarcities.
signifi-
Foveon
— The
would have
to exploit
Eye
Silicon
an amazing
235
new abundance
in
photo-
graphic technology.
Advancing through an interplay between what
volume and what
is
They use
maws and middens
ter to preserve the scarce arable
silicon to
abundant
oil
to save
human
of fossils in Earth's cen-
spaces on the surface. They use
economize on humans unwilling
abundant bandwidth
available in
not, entrepreneurs exploit the
resources to relieve the scarcities.
muscle, the interior
is
to tote
up
detail, or
to obviate scarce local data storage (or stor-
age to compensate for choked connectivity).
men and
If
scarcity,
nations ignore the abundance, and obsess on the
they plunge their companies and economies into an
abyss of ever-narrowing horizons and possibilities. nations
who
Abundances end
a
promised land.
with a price of zero. Scarcities
in infinity,
in zero, with a price of infinity.
Power, for example, has a
and
glimpse the glow of abundance beyond the deserts
and culs of scarcity can lead the world toward
end
Men
become an abundance with
few cents per kilowatt hour
from batteries
—has
become
a price of
for electricity. Wireless
power
a scarcity with a price of a
few
thousand dollars per kilowatt hour.
Bandwidth has become an abundance, with gigabit per
a
price
per
second almost free on worldwide webs of glass and
light.
Connectivity has
become
scarce, with a price of
megabit per second per month down a T-l Storage has
become abundant, with
per gigabyte on
new
disk drives.
$600 per
line.
a cost of
around 50 cents
George Gilder
236
Wireless storage for cameras has
become
scarce, at a cost of
scores of dollars for a gigabyte on a flash card.
Both entrepreneurs and economists
live in a
world of
scarcity.
Only entrepreneurs see the abundance beyond.
foveon possessed the
secrets of a
new
canonical abun-
dance: high-resolution low-power imager pixels. In theory,
it
could combine this abundance with complementary abundances of bandwidth, storage,
and wireless connectivity
theoretical.
Foveon could find markets
tapping using the
its
for a
It
could be realized only
huge volume of
potential, Foveon could
X3 s high
resolution
and accuracy
Sigma.
Since
But the
relatively small
trade.
That is
is
mar-
the strat-
a
leading
may work for them.
high-end market would leave most of the
new abundance untapped by Foveon,
CCDs
chips.
to invade the
Sigma
high-volume producer of lenses, the strategy
its
if
attack from above,
kets for professional single-lens-reflex cameras.
egy being pursued by
new
cheap imaging. But Foveon's
global paradigm of ubiquitous
abundance was purely
to create a
while the producers of
and cheap black-and-white imagers took the volume
With volumes thousands of times
their costs
and even
their
performance might eventually become
better despite their relatively
After
all,
the
Canon
1
larger than Foveon's,
cumbrous
designs.
Ds celebrated by Gorman was based on
an advanced 11-megapixel
CMOS
imager and mosaic
filters.
The
Silicon
Eye
237
Foveon's Rich Turner acknowledges that the entire system
is
beautifully designed. For a high-end camera, the actual imager is
a relatively small portion of the price.
this
Canon could dominate
market without ever confronting the technical superiority of
the X3.
Perhaps then Foveon could attack from below, using chip cheapness and dual function,
still
and motion
its
one-
versatility, to
enter the mass markets for throwaway cameras, cell-phone imagers, and camcorders. That full
—
company
albeit with the world's highest-resolution imagers.
That
not be easy to accept. Otherwise Foveon could perhaps
attack from the side, pioneering surveillance applications.
by chasing
all
new markets
Or should they
for high-resolution
risk diffusion of focus
these markets at once?
As Foveon emerged slowly from its
a
of photography connoisseurs into a pure low-end imager
house
may
means changing
stealth
and began
to
market
product in 2003, these strategic issues remained unsettled.
for many years, camera company,
the leading competitor for any microchip
digital
believes that this era theatrical films in
all
is
or
analog,
was
essentially over for
their forms,
from
But Gorman
film. still
images. Only in
70mm IMAX
to
Cin-
erama, does film continue to hold on against the persistent incursions of digital, and that too
may soon
change.
At the high end, Foveon's key competition film or even from
from Moore's
Law
Canon and itself,
as
it
will
come not from
the other Japanese producers, but steadily improves the
performance
George Gilder
238
and lowers the price of every
whether
Foveon then
will face the
new system cannot as the
camera based on
silicon
uses Foveon technology or imitates
regardless of
it
digital
grim challenge of Drucker's Law:
prevail unless
it is
it.
A
times as good
at least ten
incumbent system.
Repeating the military defender's edge in an entrenched position,
Drucker's
incumbency.
It
Law
reflects the
momentum,
scope, and scale of
feeds on the cumulative investment and tacit
learning of both consumers
tions. Similarly in
merely
is
cameras,
of manipulating lenses,
suppliers. If
Windows, you
intricacies of Microsoft
operating system that
and
if
five
you have learned the
are not ready to learn a
many
times faster for
you have mastered the
filters,
and
f-stops to shape
your picture, you will not welcome a
new
applica-
intuitive lore
and enhance
system, no matter
superior in resolution and simplicity, that relegates
new
many
how
of these
functions to post processing in software. Counterintuitive to most people, Drucker's
Law
explains the inventors angst (and lawyers'
glee) suffusing every conference
on patents,
as the creators of
things "at least two times better, five times better, eight times better"
than the prevailing 'junk" concoct conspiracy theories of
incumbent crime and malice
to explain the facts of
life.
Subverting the military analogy of Drucker's Law, however, the counterexample of guerrilla terrorism. stances, the attacker has the advantage.
is
Under some circum-
He
can select one vul-
nerable point to assault, while the defender has the impossible
challenge of preparing for not
know what
power of
all
to expect or
surprise.
eventualities.
where, and
The incumbent does
may underestimate
the
The
"do you think me
a
Silicon
Eye
239
me
dog that you contest
with sticks and
stones?" roared Goliath, the Philistine giant of biblical times,
among
expressing the contempt usual
established elites toward
a disruptive challenge. Confronting rivals from
observe the rules of "the road ahead," as first
fail to
Gates entitled his
Bill
book, powerful incumbents belittle the tools and talents of
their upstart rivals. David's slingshot
was what Clayton Chris-
tensen, the author of The Innovator's Innovator's
Solution,
"disruptive
a
calls
Dilemma and now The technology." Defying
Drucker's mandate of tenfold superiority, this
is
traption that could even be tenfold inferior by
standards and
still
blow away
Disruptive technology
is
all
the prevailing
a treacherous concept, easy to grasp
surprising effect on a seemingly fixed status
Hugely
a slingshot con-
a regnant Goliath.
too quickly. Akin to David's disruptive sticks
car.
below who
and stones
in
its
quo was the motor
inefficient at converting oats to energy, inept at hik-
ing the available
trails,
prone to constant breakdowns, and sup-
plying no fertilizer for the garden, early automobiles practically
required their users to uct was that? Perhaps
become mechanics. What kind it
was
a disrupter.
With
of prod-
their obvious
advantages in speed and comfort, though, cars seem too clearly superior to equestrian technology to be a clear case in point.
Some
upstart technologies are so vastly better than the incum-
bent that they blow
it
away by raw
ruptive stealth. Perhaps Foveon
Gorman's enthusiasm attack will not
work
for his
superiority rather than by dis-
fits this
model, but judging from
Canon, an overwhelming
for Foveon.
frontal
George Gilder
240
A
more familiar
Based on Faggin's microprocessors and their descendents,
puter.
early
and pure disrupter was the personal com-
PCs could
not execute the graphics, computer-aided engi-
real-time
neering,
transactions,
Fourier transforms, and
fast
database access that comprised prime functions of established
computers.
And
to
use one, you had to
computer technician. Yet ers,
and then
early
PCs found
become some kind a
market among hack-
companies one by one
infiltrated large
of
for use in
running spreadsheets. Within ten years between 1977 and 1987, they
moved from under
ket for
percent to over 99 percent of the mar-
1
computer power. That's disruption.
disruption becomes possible nological paradigm.
the old paradigm rate
and
versatile
market wants a
—
When
a
in the face of a
new paradigm
faster horses, bigger
cameras
—may
flyswatter,
arises,
new
tech-
advances
in
mainframes, more accu-
offer technology overshoot.
The
but industry keeps improving the blun-
derbuss. Every advance pushes the incumbents deeper into a
zone of disruption, what Christensen
calls a "failure
using gunpowder in ever-more sophisticated ways to In the context of the radical
new paradigm
framework," kill flies.
of portable instant still
and
conventional digital imagers
may
cellphone cameras or teleputers that can do both
motion images, advances
in
bring scores of impressive companies into jeopardy. Heading for
the zone of disruption today at a rapid pace
panies as
Canon and Kodak.
may be such com-
The
Silicon
Christensen explains for the
even against giants. against
what he
first
Eye
time
He shows how
241
how disrupters can win
inferior tools
calls "sustaining technologies,"
can prevail
such as ever-
mainframes and minicomputers or ever-faster and more
faster
cornucopian mosaic
digital
cameras. Analyzing a wide range of
from disk drives
industries,
from excavators
to steel mills,
to
pharmaceuticals, from motorbikes to electric cars, he brings rare
new
light to the
No
matter
predicament of the dominant firms
how
superior your technology, you
in a
may
market. well be
attacked from below by vendors of cheap convenient devices that your
The
own customers
disdain.
become
alleged success of inferior technologies has
misleading cliche. antitrust lawyers
Unleashing aggressive class-action
a
suits,
promise to save our economy from a blight of
'market failure" and bad product "lock-in." Cited as cases in point are such defective items as
QWERTY
videocasettes, and IBM-Microsoft standard
edly
won
economies
to foster
monopolies. In
was superior
QWERTY
equal in performance.
tapes,
that suppos-
Apple
lis,
fact,
each of the allegedly
in the relevant features,
record an entire movie
(IBM). Even the tially
PCs
VHS
over better designs because of a sly bent of free
inferior products ability to
keyboards,
(VHS)
or
16-bit operation
keyboard turned out
None
to
of the competitors
and Dvorak keyboards
such as
be essen-
— Betamax
— could begin
to
meet
Drucker's requirement of tenfold superiority.
Christensens theory has nothing to do with such fables of "market
failure." Instead,
to foster not
he cogently explains
monopolies but their overthrow.
how markets tend
George Gilder
242
In part, the inferiority of disrupters
an optical
is
The new
reflecting the vantage point of the establishment.
ucts are inferior only by the prevailing metrics.
ures they
may be hugely may
established tools
or
may
how much
faster
its
By other meas-
not be able to function at
hold or
prod-
or even infinitely superior, since the
market. For example, no matter disk drive
illusion,
how
how many
superior
is its
all
in the
new
gigabytes a 5. 2 5 -inch
seek-and-rotate time
data throughput rate,
it
could not supply
room only
storage for a laptop computer, let alone a camera, with
for 3.5-inch or smaller disks. In that market, the superiority of
became suddenly
the larger disks '
irrelevant.
The
smaller, slower
were competing not against another technology
'inferior" disks
but against the alternative of doing without. As Christensen says,
"Competing not against
existing competitors with existing
customers but against nonconsumytion cameras, and
cell
—
against
teleputers,
phones with no internal hard drives
at all.
Christensen shows that in the history of technology tern recurs over
and over
again.
The
this pat-
inexorable dynamics of
innovation render dominant companies of one industry para-
digm feckless Since most their
digital
cameras and camcorders currently receive
images through a silicon array of
CCDs
now
are
some $5 Sony
in another.
billion,
still
a fast-growing,
CCD
photodetectors,
Japanese-dominated business of
and the pride of wafer fabs from Sony
cherishes high hopes for
its
to Sharp.
future. Shigeyuki Ochi,
Sony's venerable semiconductor chief, predicts the ubiquitous
use of
CCDs
agrees: "As
for video teleconferencing. Hiroshi
we move
into the multimedia era,
Inoue of Sharp
CCDs will be used
The
Eye
personal computers will have
all
Wherever
human
there's a
eye,
Carver
Mead
need
for
electronic,
if it's
CCDs
something it'll
X3
Tokyo early
visited
CCDs
device. Japanese engineers
very politely to his case.
Then amid
.
.
"When you make
your chip behave
same power
as a
like a
CCD
in the
could be
would
listen
various permutations of
same
smiles and bows, they would respond in the
uses the
.
to play the role of the
century to bring the good news that
it
them.
built into
thus did not cause dancing in the clean rooms
replaced by Foveon's
so
lines,
be the CCD."
and design centers of Japan when he
new
243
images and send them down phone
to capture all sorts of
and
Silicon
essential way:
charge-coupled device,
and plugs
into the
same
CCD and emits the same kinds of signals as a CCD, exhibits the same noise level as a CCD — and you make
sockets as a
and
it
significantly
cheaper
—then
you come back and we
will talk
to you."
The source
of the stubborn opposition
is
no conspiracy of
threatened incumbents or blindness to opportunity. Drucker's
mandate of the tenfold advantage other Japanese Goliaths, with
all
applies. Facing their sales
Sony and the
momentum,
learn-
ing-curve progress, knowledge base, cumulative investment, and
customer
familiarity,
Christensen, unless
Foveon has "no chance" it
to prevail, declares
adopts a disruptive strategy.
goes up head-to-head with the incumbents,
"it
If
Foveon
will lose,"
he
says.
In his entire Harvard database of technological innova-
tions,
he has catalogued no case of an outsider displacing an
incumbent
in
what he
ing core business.
calls a "sustaining"
—
technology
its exist-
George Gilder
-44
"Foveon cannot make cameras," he
must pioneer new markets.
camera customers but tions.
done
It
asserts.
To win, Foveon
must compete not
for entirely
new customers and
Foveon can meet both these requirements. But it
a
applica-
it
has not
yet.
Returning from Japan without any significant the
for existing
X3 beyond Sigma's high-end camera, Mead
problem,
it is
an opportunity.
nearly all the
We
are going to
new
orders for
said: "This is not
make cameras."
world's executives have learned the apparent
lesson of IBM's decision to adopt Intel microprocessors and
Microsoft operating systems for the
IBM
personal computer.
Therefore, companies such as Nikon, Kodak, Sony, and will mightily resist pasting
Canon
"Foveon Inside" on the fronts of their
cameras. According to Christensen,
Mead
will fail if
he
tries to
capture these existing camera markets. Yet following his trips to
Japan
to find
customers for the X3,
Mead
many
believes that
in
order to demonstrate the superiority of his chips, he will have
to
make cameras. The
solution to his problem
the Japanese camera journals full
significance of the
is
now
emerging. Early in 2003,
became the
first to
recognize the
Foveon breakthrough. Three leading
Japanese photography magazines put on their covers alone or with one
SD9, made by
rival
a lens
—the image of
company
that
a
—
either
new camera, the Sigma
had never before manufac-
tured a digital camera. Like Intel's logo on a personal computer,
the camera bore the distinctive
X3 svmbol
of Foveon.
The
lead-
— The
ing Japanese
Eye
Silicon
245
camera publication
digital
Digital
Photo
—
devoted no fewer than twenty-eight pages to a detailed comparison of the Sigma
SD9
with Canon's
EOS- IDs
single-lens-reflex
Canon
camera. The journal concluded that although the
pos-
sessed certain advantages in ease of use, the resolution and verisimilitude of the
Sigma were
essentially equal or superior in
every category. Operating near the peak of their resolution, the
Canon
CMOS
with some five
$1,000 "blur"
imagers suffer from the distortions inevitable of every pixel. Included
filters in front
filter to
smear away color
even a
is
artifacts that result
from
the inability of the chip to define colors near borders between
two different hues. Without any way sor to re-create information that has alternative
is
for a digital signal proces-
been thrown away, the only
to blur the colors together.
Although camera technology connoisseurs such as Merrill
and Mead can detect these
flaws,
so brilliantly performed that
it
is
Canon's invisible
digital
even
processing
is
to professional
photographers.
With camera man.
all
the very
is
It is
cence of that the
the other features
honed
same machine
that so captivated
the device that prompted
his cherished film.
Canon IDs
photography can
get.
is
to perfection, this
him
it
Greg Gor-
to assert the obsoles-
Even Foveon's own experts agree
a superb camera, about as
But
Canon
costs
some
five
good as
digital
much
as the
times as
Sigma camera using the Foveon X3. Inside the pages of Digital Photography
was
a full portfolio of
photographs from the two cameras. To the lay person, there
seemed
to
be no discernible difference
in quality.
The Sigma
George Gilder
146
camera proved that the Foveon technology was and effective
erful
in its initial
form
be introduced near the
Canon camera used
top of the market. But the expensive plex multifiltered
to
pow-
sufficiently
com-
a
imaging system, while the Sigma camera
image reception and processing
in just
one chip. That chip might ultimately be manufactured
in vol-
embodied
ume
all its
crucial
for as little as a dollar.
As
Merrill explains,
"The fundamental advantage of the
cal color filter technology
is
verti-
two-and-a-half-to-three-times
its
greater efficiency in the use of any particular silicon area or lens
quality
With
a lens of given quality that can resolve a spot of
more information
given size, Foveon can measure three times
than can a
CCD
with a plastic color
Silicon area
filter."
is
a
rough index of unit cost, so three times the efficiency in the use of silicon area
means one-third the chip
cost.
Lens quality com-
pounds
this advantage, since for
quality,
Foveon can use a lens with one-third the focal area
(which means
far less
any particular
than one-third the cost).
As Foveon's volumes increase, cheaper applications. Because of be incorporated ability to
market of
in every cell
render both billions, cell
still
its
its
imager
is
moving
supremely low power,
phone handset, imparting
into
it
can
a joint
and motion images. With a potential
common
digital
for imagers.
From
phones are now the most
machines and represent the cell
image
level of
largest
market
phones, Foveon can move to every computer screen for
video teleconferencing, which
is
the next major broadband
Internet application. Multiple Foveon chips can be deployed for surveillance in every convenience store
and gas
station
and can
The
guard every automatic
teller
Silicon
Eye
247
machine. Foveon imagers can sur-
vey every security point in every airport. Fulfilling the
eventually
Christensen mandate, the Foveon device can
become ubiquitous without
incumbent camera company then
—not before —
into the
This
that the
ever challenging any
for its current customers.
It
is
Foveon chip might move massively
mainstream camera market.
is
the disruptive path of Foveon
vator's solution" that
Christensen
compete not
is
s
triumph.
It is
proposing in his
the "inno-
new
book.
chiefly against existing products but
Foveon
will
against
nonconsumption of imagers. By making the imaging
function cheap and robust, Foveon can open huge
new markets
for its innovation.
Foveon s ultimate achievement onrush of
translates into a digital file of
up
characters; this
book
Foveon pictures
to take over the
less
depend on the continued
Each Foveon picture ultimately
digital electronics.
is
will
to
40 megabytes (millions of
than one megabyte). In order for
world of photography and film,
these dense images will require broadband communications and trillions of
bytes of storage in a combination that
"storewidth."
Meads
I
have dubbed
analog technologies will change the world,
but the world will also have to change to accommodate these
new
capabilities.
Intel
and Microsoft are among the leaders
in
moving Foveon
toward mainstream applications. Intel has staked
much
of
its
future on the success of wireless broadband communications.
To transmit Foveon's streams of images from surveillance devices or cell phones will entail ever
more
wireless bandwidth.
George Gilder
148
Now
spreading through "hot spots" across the country from
Bryant Park in Manhattan to Dallas-Fort Worth Airport, from the top of the
Mark
2.4-gigahertz
and
to
Starbuckses everywhere, WiFi in both the
5 -gigahertz
bands can play a key
role in public
places and offices. However, in accord with Merrill's rule
camera ing to
satisfactory unless
is
—photography
Foveon
that can cell
will
be
is
you want
it
mountain climb-
not restricted to "hot spots."
Qualcomms
More
capabilities
everywhere that
are deployed.
Preparing for the Foveon revolution in software
When tem
Bill
Gates launched
at a trade
relevant
CDMA 2000 EV-DO technology
endow broadband imager
phone systems
to take
— no
his
is
Microsoft.
XP-network-based operating
sys-
show, this king of the digital age did not begin by
entering a password or clicking icons in a pop-up window. Instead he put his finger firmly on a glowing biometric touch
pad that recognized Gates, loaded
him access
to his personal files
pany that supplied But
its
and
digital
open sesame
is
innovation in pattern matching
cal leader,
another
this
his personal settings,
Vance Bjorn,
fruit of Carver's
a former
campaign
and gave
kingdom. The com-
called Digital Persona.
is
analog and
Mead
its
techni-
student, represents
to transform the
world of ana-
log interfaces to the digital world.
Microsoft has also developed a
new
standard for representing
photographs on the net. The GIFs and
World Wide
Web
JPEGs
of the current
are grossly inadequate to capture the vividness
and verisimilitude of Foveon pictures. As new standards are perfected, the world of the
image of a cat
—
Net
will
be able
to
render yet another
a cougar, in the act of leaping off the screen
The
Eye
Silicon
249
toward you. Mead's revolutionary camera
make your com-
will
puter brim with the intense dynamic colors of a
new
era.
Like
the apparent glass on the top of the pile of carpets, your screen will
disappear into a luminous
Still,
most poignant
the
new world
is
cat. It is
first
Mead and Mahowald. But
a three-foot-high
It is
based on the same essential
con technology used twelve years before ican cover heralding the
in the Scientific
photography of any kind.
sili-
Amer-
working retina chips made by
rather than the blurry
matic sketch offered on that old magazine cover, the offers a full-color vividness
color.
the picture on the wall
outside Mead's corner office at Foveon.
image of the face of a
and
symbol of the amazing
feline
advances achieved by the company
of art
and verisimilitude
Some
resolves every hair, whisker, glint,
six
square
lifelike
new image
rarely excelled in feet,
and gleam of the
renders the eyes of the cat with a
monochro-
the image
feline fur
and
glow that gives the
viewer the distinct and disturbing feeling that a formidable ani-
mal
is
To
watching him. Yet the picture fulfill
would have
of a kitten.
the promise of kitten or cougar, however, Foveon to find
down market would be
is
live
new
leadership that could take the
into the realms of disruption
cougars galore.
where
company
in its
path
23 Holy Kammol
On
his daily
dawn walk
the Arastradero Preserve,
in
intently chatting in Italian with Alvia, his wife of thirty-
Federico Faggin met a cougar in his path.
six years,
Early in the twenty-first century, cougars, terrorists, class-action lawyers,
begun
and other
to gain
feral creatures all
"had their rights" and had
an edge on humans in Silicon Valley Faggin had
home
seen several big cats around his
Admiring the sleek
feline,
Los Altos
Hills.
Federico and Alvia waited for
cross over into the sepia grasses side of the road.
in the
and twisted
Then they made
their
trees
way
it
to
on the uphill
briskly
down
the
mountain. Entering his large predators,
with a jaunty
sixties,
Faggin was no longer inclined to confront
whether sepia or
gait,
the
investor." "After all," so I
Graying
at the
temples,
dimply smile, and a modest paunch, he was
slowly retiring into
Altos Hills,
silicon.
Silicon
Valley
role
of "gentleman
he says with a gay chortle,
could hardly be a gentleman farmer."
251
"in the
Los
George Gilder
2^2
By
dint of the insistence of Alvia,
who
corrected any journalists or historians icles of
gently but persistently
who
her legendary husband, the world
erred in their chron-
now
increasingly rec-
ognized Faggin as the true father of the microprocessor and the silicon-gate process that
proud of their also
done
artist
made
it
possible.
at
a biological father,
daughter, Marzia, and their two sons, he had
well. Eric, twenty-three,
philosophy
As
was
a graduate in physics
and
Santa Clara, making a leisurely survey of the uni-
verse,
and the other son, Marc, twenty-four, had done
major
in
math and chemistry and then plunged deep
a
double
in pursuit
of a doctorate at Cornell, in nanotechnology. "Ill leave that to
him," Faggin said with a smile. In the Los Altos Hills, he
sides by local venturers
found himself surrounded on
and investors devastated
technology crash and then by the in
public offerings.
initial
new
all
by the
first
century's long doldrums
But Faggin had mostly emerged
unscathed. After scoring handsomely in Synaptics, which had the
first
invested ates
and only major Silicon Valley IPO of 2002, he had
some $400,000
in
Foveon with
(NEA), the eminent venture
Kramlich.
He
New
Enterprise Associled by Richard
capitalists
looked forward to the success of this promising
technology in the midst of a booming market for imagers. Later
he joined the board of Blue Arc, the preeminent success ware
for storage networking, a British
and financed by
Italian friend
company
in hard-
initially
guided
Walter Allesandrini and run by
another Italian friend Gianlucca Rattazzi. Faggin also became a director of the Allesandrini telecom
equipment
start-up, Avanex,
which soared during the optics boom, collapsed
in
2000, and
The
Silicon
Eye
253
then morphed into a successful scavenger
among
the ruins,
picking up optical divisions from Alcatel and Corning.
Becoming call
a laureled lion in the Valley, Faggin
from Jim Thornburn of
Zilog, the
even received a
company begun by Faggin
and Masatoshi Shima when they defected from later
returned to Intel to head
its
Intel
Japanese research
(Shima facility).
Faggin had broken entirely with Zilog in 1981 and, until the mid1990s, the
company had
subsisted on sales of billions of Z-8s
and Z-80s, microprocessors embedded
in
kitchen white goods, and microwave ovens.
automobiles, toys,
Then
Corporation decided to cash in on the telecom the
company
for
network processors
private
and relaunching
it
Texas-Pacific
boom by
into the hot
— microprocessors used
to
taking
new market
run networks.
Milking cash from the Z-80 to support entry into the com-
munications chip business where Zilog built a
roots as an
Engaging talgia,
some
had no special competence,
$700 million wafer fab and nearly went bankrupt.
Thorburn asked Faggin its
it
for
guidance in returning the company to
embedded microprocessor
in a rare Silicon Valley
firm based on the Z-8.
success in the enterprise of nos-
Faggin enjoyed helping in the revival of his design from thirty-five years before.
In the late
summer
of 2003, the Faggins left for a vacation in
China. Both looked forward to more exciting travel as the years passed.
foveon, meanwhile, was
in the throes of a series of tur-
bulent board meetings. Under a search team composed of Dick
George Gilder
254
Kramlich and Forrest Baskett of
New
Enterprise Associates
(NEA), Dick Sanquini, formerly of National Semiconductor, and the absent Faggin, the board was attempting
CEO who could
take the
company
across the
to find a
chasm from
high-end cameras to the promised land of billions of
cell
new few
a
phones.
Under prompting from NEA, the Foveon board hired Lonergan Richards, a Silicon Valley headhunter, to conduct a nationwide search. Faggin
would presumably be covering China.
Previously chosen by Faggin as his chief financial officer at Synaptics, interim Foveon
porate leader. liked him.
An
CEO Jim
Lau was
a
competent
cor-
engaging man, with a friendly smile, everyone
As the organizer of Synaptics s
intricate
web
of Asian
manufacturers, Lau had been crucial to the success of Faggin's previous company.
CFO who
Lau would continue
also possessed a sharp
to serve
superiority for assert that pixels.
a
its
was cautious.
Foveons imager. Under Lau, Foveon did not pixels
were worth between two and three
master of the "no comment."
company was being passed by
company
to journalists,
He was no rivals
that
cougar.
most
digital
Lau was
And now
his
increasingly alarmed at
Lau had been
leading.
Full of photographers indignant about the artifacts tortions inherent in
CCD
with inferior products. As
members met, they became
the condition of the
and
dis-
cameras, Foveon had targeted
the top of the camera market, where Carver earlier at
He
by making grand claims of
That would be aggressive. Talking
the board
a
understanding of the manu-
facturing mazes of Asia. But in marketing he didn't like to offend competitors
Foveon as
Mead
three years
Photokina 2000 had uncovered one avid customer,
The
Silicon
Eye
255
Sigma. But in the subsequent years under Lau's leadership, no further buyers of the Foveon chip had emerged.
many customers era,
for Sigma's
which Sigma treated
much-honored but
Nor were
there
still-exotic
cam-
chiefly as a loss leader, a
Sigma's estimable line of lenses.
On
way
to sell
the board of directors, Brian
Halla of National Semiconductor brought everyone
down
to
earth by comparing Foveon's search for customers to "a fellow
with a metal detector combing the beach."
foveon's chief hope
breakthrough
for a
at the
bottom of
the market was Sanyo, the major independent Japanese
manu-
facturer of mass-market digital cameras, including Nikon's pop-
Ask anyone
ular Coolpix model.
beyond Sigma and those to
know would
cite
in the
Foveon about customers
at
know
talking to those with a right
Sanyo, sure to sign any day now. Foveon's
marketing department under Allen Rush had been negotiating with Sanyo for nearly two years. All the reports were "highly promising, deep interest, excited by the technology," but some-
how no
deal was closed. "You
know how
Mead
panies are. Takes them forever,"
Now
in the fall of
companies
meeting
cell
in
Lau went
to
Sanyo
in
told
me.
phone imagers from other
selling in the millions, the
close a deal. a
2003, with
those big Japanese com-
board mandated Lau to
September 2003 and secured
Tokyo with the top executives
top Sanyo executives turned out to care
in the
little
company. The
about Foveon. Lau
reported to the board that while the lower-level engineers were enthusiastic about the Foveon sensors
X3 and XI 9,
the senior
George Gilder
them
executives were open chiefly to using
down that
CCDs
prices from rival vendors of
fit
CCD
snugly in
When
imagers
Faggin returned from China in September, he began to
tains with Alvia,
ture?" he asked.
On
visible results.
He was
"What
Sure
CCD
reminded him of the experience
ing
still.
the
company perfected
moun-
at the
slow
this pic-
beginning to doubt what he had been
the light at even pixel, but the
It
in the
wrong with
is
told about the superiority of the technology. all
walks
his
he expressed increasing frustration
development of any
it
was better
to
was not standIntel
at
when
dynamic random access memory
its
(DRAM). Bob Noyce and Gordon Moore
DRAM
CMOS
and
slots.
hear distressing reports on Foveon.
capture
as a foil to force
projected that the
would displace the regnant magnetic-core memories
two or three years. But challenged by the
DRAM,
core
in
memory
engineers improved the cost effectiveness of cores tenfold over the next five years.
decade.
Now to
reduce their pixel
pixels
size
Challenged by
Foveon's X3, however, to
triumph was delayed by half a
and Merrill declare that
square microns).
managed
D RAM's
Faggin at board meetings heard
CCD inventor) be able
The
CCD
reduce their pixel
5
CMOS
size to
(himself a
CCD makers would not
below
engineers
Mead
at
microns square (25 imagers,
including
Sonv and Canon soon
4 microns square.
were soon not only smaller than Foveon
CCD
scientists believed
possible but smaller than Foveon itself could then
make
at its
National Semiconductor fab. Dick Lyon insisted that the smaller pixels yielded
More to
errors.
worse images. Less
light
was collected per
But the fact remained that the
CCD
pixel.
was on track
put three pixels, together with admittedly tricky plastic color
The
filters, in
a space that
tical pixel.
Eye
Silicon
Foveon could put one three-transistor
So what was
this
Foveon
huge advantage?
would be
and
taste. Yes, the
less
power, and so on. But Foveon
pixel
problem persuading anyone nology in favor of a
257
new
to
still
A matter of design more
elegant, use
would have
throw out their old
a serious
CCD
tech-
device that did not even yield smaller
and thus did not even allow Foveon
pixels
far
ver-
to claim leadership in
the heralded "can-you-top-this" megapixel sweepstakes.
But then, under Lau and Allen Rush, Foveon was not even claiming more pixels, except by the backhand "times 3" after the 3.6 megapixel
number (even though
the "times 3" was always
elided in press reports). Foveon's marketers had allowed themselves to be intimidated by the Japanese
from even asserting that
with three colors at every spot a Foveon pixel was worth 3 onecolor pixels
on
a
CCD
or ordinary
megapixels were actually said the
tive,''
1 1
CMOS
imager. Foveon's 3.6
megapixels. "Better be conserva-
Foveon marketers. "Better be wrong, and
lose,"
said Faggin.
Faggin argued that the board could not find a promising until
it
tigate.
determined what the problem was.
Then
as
He
offered to inves-
he began his inquiry, he realized that he could
not succeed in diagnosing the problem unless he had to
probe anywhere in the company.
Lau
by
as interim
CEO
He
full
power
suggested that he replace
CEO.
the next board meeting
in
September, Faggin was ready
to
take over. Before leaving on his search for powerline artifacts in the Appalachians,
Mead had one
final
meeting with Faggin.
— George Gilder
"Faggin chose to have one of his hissy
Carver
fits,"
recalls.
"According to him, no one had done anything right since the
founding of the company. Before he took
show
else
the day.
He
—
knew what they were doing built himself up.
up the pressure, the do the job. that
he wanted to
everyone that the situation was nearly impossible
to
no one
over,
Much
of
But
that
that only he could save
he was building
in the process
intensity, the
competitiveness he needed to
what he
was exaggerated.
Sigma was not the
said
final solution.
We
We
all
knew
knew we had
all
to
enter the cell-phone market. But that said, there was no doubt that he
had a much better view of the competitive
situation than
any of us had." Faggin's investigation
who had
passive chairman
"Mead was
had persuaded him that Mead was
allowed the company to go off track.
He knew
infatuated with the technology.
advantages, in principle.
a
He was
with the theory.
in love
all
the
When
you are infatuated, you cannot see the flaws. Faggin concluded that in principle the Foveon process was
indeed superior imagers. But in
And
in the
to its
charge-coupled devices and existing
form
it
CMOS
was not superior
digital
in practice.
imager business, as opposed to the image business,
the market for principles remained small.
from time to
time over the years. Carver
had clashed. Carver was always listening while Faggin was listening to the market. to
open up technological
possibilities,
Mead and
to
Mead
Faggin
the technology,
normally angled
while Faggin sought to
anneal them into a sword. But just as Faggin had previously
— The
deferred to
Mead
in
Silicon
Eye
259
shaping the technology,
readily deferred to Faggin in running the it
now needs
Faggin's
Faggin
is
mused on
needs his own technology
it
no slouch on technology
Reflecting on
company, believing that
marketeering insight and strategic
resourcefulness more than
And
Mead
at this point
either.
Mead
twenty companies,
experience with
the Foveon saga:
insight.
what happens when you
"It's
start a
company. The unlimited potential of your new technology.
huge high it
just thinking about
becomes
thing.
.
.
.
a product,
It's
it's
it.
But once
a
manifest, once
it is
not a myriad of anything
happened with every company
It's
—
one
it's
worked with
I've
they get to the point where they're successful, they're on track,
and
there's less
and
less that
You actually become
someone
a distraction
—
like
questions. That's
when
Mead
observed:
"He
these interesting
all
right
man
to lead the
doesn't take a lot of input.
always seeking win- win situations, but Federico loves a All the
good
And
today Foveon
Japanese want to knock
no one can "I
fight.
am
fight
having the time of
Faggin was focusing
over,
Kammoli!"
my
in the
a fighter.
midst of a
None wants
am
I
He
fight.
to give in.
And
"If
life
not having to do this," says
in better hands."
fiercely.
runway," he commented. is
off.
it
is
is
com-
more resourcefully than Federico.
Mead. "Foveon could not be
way
and
time to leave."
it's
Concluding that Faggin was the pany,
can contribute.
they're trying to focus
you're wandering around thinking about
new
me
"A small company has
you don't gain
lift
you see the fence coming ahead
.
a limited
before the run.
.
and
it's
Holy
Dick Merrill
tells
the Telecosm conference about the myriad
cations for cheap, high-resolution imagers. Glenn Davis
new appli-
RAW in Vegas
Foveon
couldn't
happen
to a
guru more
digitally
Delphic than John
ItDvorak, beetle-browed futurist, nerd supreme, PC magazine doppelguru, cynical seer through the industry's swell and jig-
raw
gle to the
new
crazy
silicone below, virtuoso italicizer, brazen imbolder,
talent at
the
Consumer
Electronics Show, serial
keynoter at the Photography Marketing Association, maturing as a
vamp
Teller
in
and proud owner of a
From ics
Vegas as familiar as Cirque de Soleil or Penn and
Show
free
Sigma 10 camera.
the record-breaking January 2004 a
month
earlier, full
of
Consumer
Electron-
HDTV specs and scams,
so goes
the story at Foveon corporation, Dvorak went on to the wedding of a friend.
When
his chance.
the photographer failed to show, Big John
He would
become: John Dvorak, ier
I
can
bravado
tell
take the official
wedding
pictures.
photographer, which
He would is
a lot eas-
you than expectorating two pithy columns,
in bold,
and i-pinions
in italics, every
magazine.
261
saw
two weeks
full
for
of
PC
George Gilder
262
Under
a
canopy on
a
sunny
blithely with a dark foreground
images.
If
day,
the wedding unfolded
and bright background
all
the
there had been a professional photographer, he would
have been gnashing his apertures and didn't
for
notice.
He
clicked
just
away
filters.
But Big John
relentlessly.
Just
like
columns. Write enough of them, with enough pungent wild-
eyed predictions, and
if
you are Dvorak many
prove to be
will
More
amazingly on target and your column will be a gas to read. is
more.
them
More
will
better.
Take enough photo shots and some of
be great, right?
era bearing the all
is
And
with this super
Foveon X3 imager that collects
new all
digital
cam-
the light and
the colors at every pixel, he couldn't go wrong. His host sure
was lucky
to
have
all
that extra
luminance and chrominance of
a
beautiful day and lucky to have Big John to memorialize the
event for the ages. Arriving
home, Dvorak booted up the Sigma Photo Pro
soft-
ware, linked the camera to his computer with a universal serial
bus (USB), rendered his harvest in
JPEGs
(the familiar
pression standard of the Joint Photographic Experts is
com-
Group
that
ubiquitous on the Web), and prepared to bask in the magnifi-
cence of his work. But what
is
this
garbage on his screen? Dvo-
rak was shocked. Every picture he took
was under the dark
canopy with the bright background and every picture was dungeon dark with a dazzling halo. In botching the scene
same way each and every and
time, the
Sigma was
in the
entirely robust
reliable.
With JPEG images was no way
to fix
it.
—
a global
Change the
compression algorithm settings
and you
—there
just get
new
— The
Eye
Silicon
permutations of bad pictures. You darken
263
enough
it
to mitigate
the blinding brightness outside, and the bride's face disappears
under the shadowy canopy. Brighten up the side scene
face,
whites out" along with the wedding dress. Ready to
'
write the experience
he made a plaintive
No problem,
off,
call
perhaps with a column dissing Sigma,
back
to Eric
says Eric. This
Zarakov
at
Foveon.
the digital era, John, just as you
is
said last year giving your friendly keynote advice to crafty analog photographers,
etors of all
cramped
just skulk
was time were
little
away and use
Once
Foveon
moldy
all
fig film retailers,
the artsy-
and
propri-
photography shops that they might as well their stores to rent
to sing a Hallelujah
right.
and the out-
chorus of
digital fiber alles.
again a prophet in your
you have
DVDs, because
own
it
And you
time. But for the
use the correct
digital for-
mat. For redeeming botched exposures, you can't use
JPEG. You
full-scale
have to do
it
effect,
Vegas
style, in
image frame.
just
one simple
light sensors
shadow
is
ness will be
made
lambdas and
be well.
early February,
totally
an
accu-
And
FillLite slider bar in
to a level of perfect balance
days, or years after exposure it
and the brightness
lines will lie
By
it
and serve
and the luminance all will
Hubel
shift of the
RAW data
light
it is
in
in
there.
—even hours,
you can cook up the
on the camera
but don't panic,
your Sigma Pro software, moving of light and
undressed,
and bytes from the scene arrayed
will look awful,
information
rate; all the
With
It
RAW. The pure
the
uncompressed output from the digital form. All the bits
to
au will
down with so
it
point.
The
dark-
be made mute,
the chrominance
was.
when he rekeynoted
the Photography Mar-
George Gilder
2^4
keting Association bash, also in
member
ecstatic
Las Vegas, Dvorak was an
of the Foveon cult.
benefits of a real accurate
He had
window (RAW)
at
experienced the
every pixel coupled
with FillLite, a simple ingenious software algorithm contrived by a hiree of
Dick Lyon named Paul Hubel.
RAW
unique. Given accurate
files,
ware conversion programs can do
with an array of some it
soft-
six dif-
simple and elegant
to use. Just slide the bar, in either the positive or nega-
tive direction, into
suits
not exactly
is
Photoshop and other
Hubel had made
ferent adjustments. But
and fun
it
FillLite
the sun or into the shade, until the picture
your dream. Professionals
may
cavil,
magazine, Dvorak was right to be bold and
but afterwards
in
PC
italicized.
the crest of the digital revolution, the 2004
PMA
show
was the organizations eightieth and the biggest
ever,
with
On
17,000 exhibits, mostly
digital
cameras and accessories, sprawl-
two
levels of the
Paradise
on
South End of the Vegas Convention Center
off
Avenue (occupying the North End near the Hilton were
the spruce armies of the ing
show
floor
ing over fifteen acres, or a million square feet of
many comments on
Gun and
Rifle Dealers' show,
prompt-
the comparative firepower of digital
cameras and machine guns).
As Dvorak had long predicted, death of photography but digital
its
digital
turned out not to be the
efflorescence. In 2003, worldwide
imager sales of nearly 50 million exceeded the experts'
consensus of 38 million by over one-third. Commandingly the product of the year, the
digital
camera
in all its manifestations is
giving vision to
computers and telephones, twenty-four-hour vig-
ATM
machines, taxicabs and convenience stores, and
ilance to
The
Silicon
Eye
265
high resolution to cheap, adaptable cameras. the
PMA
RAW.
convention was
And
the
theme
of
Sashimi, sushi, Canon, Sony,
Nikon, Konica, Minolta, and Olympus. But the big news, for those
who knew
the plot, was Foveon, which sold a nearly
imperceptible share of those 50 million imagers but offered the best raw
of
files
all.
New York
Dvorak, Markoff of the editor of Wired,
much
as
all
were
if it
Markoff 's piece,
knew a
the plot, so they covered the
his third
the
initial
new Foveon
show
Foveon product launch. For the Times, on Foveon, appeared on the front page
of the business section on
Today and the
Times, and Chris Anderson,
Monday
of convention week.
USA
issue of the convention tabloid also featured
product. To the industry-leading cost perform-
ance of the Sigma 10
(DSLR), Foveon
digital single-lens reflex
had added the industry-leading cost performance of shoot camera and
a point-and-
DVD quality VGA (video graphics
array)
cam-
corder for an announced $399. Still,
Foveon
in
Vegas was chiefly in demonstration mode.
The new point-and-shoot camera liant
new
partnership with
Kodak
somewhat tarnished name Polaroid either. Polaroid
did not spring from or
Sanyo or Nikon.
of Polaroid. But
it
ital
cameras
to
new
financial crisis
an
bore the
was emerging from bankruptcy with an
legacy gear that was finding its
It
bril-
was not made by
array of instant digital kiosk printers, cameras,
the stresses of
some
Italian
it
the digital era. But under
had licensed
its
name
for dig-
from Hong Kong named Giovanni
Tomaselli, head of a firm called
Surely you have heard of
life in
and Polaroid
World Wide Licenses Limited.
WWLL.
It is
an important design and
George Gilder
266
manufacturing subsidiary of an even better-known company called the Character
Group PLC. CG-PLC,
were quick
is
the
to stress,
as the
Foveon
folk
on the London Stock Exchange
listed
Media and Photography
Sector, just in case
in
you are so
benighted as never to have heard of the Character Group PLC.
Marketing the camera Uniden. Don't ask.
in the
United States, however, would be
the familiar Japanese vendor of cordless
It is
telephones and exclusive U.S. distributor of Polaroid digital
may
cameras, which as you
Wide Licenses Limited
of
recall are
manufactured by World
Hong Kong and
in
one instance
employ an X3 imager from Foveon. In the midst of this
magpie 's nest of commercial and manu-
facturing entities lies the pearl of Foveon's intellectual property
and conceptual
superiority,
which Dvorak came
to appreciate at
the wedding and which to nearly everyone in Foveon's circle
seems resplendent and undeniable. The inelegance of the competitors'
language
Bayer it
filter
evokes.
color routine
The
is
evident even in the twisted
result of using filters to focus light of
only a single primary color on each pixel sensor, the conventional
Bayer pattern sor
is
needed
is
a
to
mosaic of red, green, or blue
compute the
pixels.
A proces-
actual colors from the colors of
adjacent pixels.
Describing the process of converting a Bayer pattern into a usable
RAW file —here Foveon chief scientist
with distaste
and
—
de-alias.
are terms
such
as demosaic
Lyon's lips pucker
and de-Bayer, de-jag
Accurate from the outset, Foveon's colors do
not need to be demosaiced and their edges do not have to be "blur filtered"
and "de-jagged," and
their patterns
do not need
The
Silicon
you can see every
"de-aliasing" so
design on Carver Mead's
its rivals
surrounded on
the-line pictures
high-resolution
thirty it
curlicue in the swirly
last
it
was obvious sides.
all
high-accuracy
company
that the
You want perfect
and you get the Sigma 10
megapixels, 12x zoom,
can buy that too
267
shirt.
to foveon's partisans, had
Eye
for $1,200.
point-and-shoot,
top-of-
You want
with
4.5
RAW files, and FillLite software, and you
in the Polaroid x530.
frames a second of
full
motion
You want
a camcorder,
VGA video and you can get
from that same point-and-shoot device.
Dick Lyon could explain
to
you the
intrinsic inferiority of all
the hundreds of other cameras with slightly different features glittering across the pullulating floor.
pixels (and
more "megapixels"), but each
electric charge
ation
between
and thus could capture its
to using the
would be worse,
camera
— and
of those pixels held less
less
dynamic range
darkest and brightest level) and was
sitive to noise. Plastic color filters
compared
They may have had smaller
as
silicon
more sen-
over every pixel were a kludge
itself as
Sony discovered with
don't
(vari-
a its
filter.
new
The
picture
12-megapixel
even mention the fiasco of the new
14-megapixel Kodak.
Supporting Lyon's claims was Lawrence Matson, one of the earliest
among
Sigma enthusiasts and
a staunch defender of the faith
the professional photographers on the
tography
Web
site.
DP
Review pho-
Flying in from Switzerland to attend the
Vegas unveiling, Matson recounted the original excitement he
"
George Gilder
felt in
He
encountering Foveon imagers. seen on the
tos (even as
distortion
Web
1
)
cited early
Sigma pho-
of eyes reflecting the sky without
—
of a frog's eye mirroring an expanse of blue sky and
clouds, or the eye of a cormorant suffused with a span of heav-
enly blue.
He
described a portrait of an old
man
with a single
curl of white hair precisely distinguishable. Perfectly rendered in a
Sigma
Matson
portrait,
become
said, "that hair in a
Bayer mosaic would
a blue-and-red rope or a 'barber pole."
Because of the accurate borders between etches sharp edges that give
its
colors,
Foveon
pictures a three-dimensional
property with images seeming almost holographic. In order to avoid "jaggies," mosaic cameras use "blur
Canon images what
smooth but
it
"I
call
it
Xutella.
cameras, the Bayer burden also
per second,
It
makes the
transi-
also flattens out the picture."
Just as important for the future of
Because of the
"That gives
their devotees like to call 'buttery smooth'
textures." said Mattson.
tions
filters."
still
afflicts
difficulty of processing
and motion combo full-motion images.
huge raw
files thirty
times
systems for full-motion capture must cluster
all
adjacent pixels and process
them
together.
But non-Foveon
imagers cannot readily cluster neighboring pixels because in a
mosaic the adjacent pixels capture different colors. They would
combine
into
chrominance murk. The Bayer-based cameras
thus have to perform time-consuming and image-distorting digital
acrobatics to join together the pixels of the
same
color.
With
Foveon, the combination of neighboring pixels was natural, simple, fast,
and accurate.
In theory,
all
the Foveon claims were true and had been true
for several years.
The Foveon
design, in one form or other,
would
The
ultimately
Silicon
dominate photography.
Eye
269
Dramatizing the business
challenge faced by Foveon, however, were the exhibits
them on the
floor of
inventor Federico Faggin put ter,
but in practice
it
CEO
PMA. As
is
it,
not."
all
around
and microprocessor co-
"In principle,
Foveon
is
far bet-
Canon, Sony, Samsung, and the
others might have to push their technology harder to get a per-
formance comparable If
Sony's
ter
CCD
colors
but they had the clout
to Foveon's
were
distorted,
with emerald hue to correct them.
imagers produced jaggies, gies are invisible to the If
make
naked
to
do
add yet another Bayer If existing
Canon
it.
fil-
CMOS
the pixels so small that the jag-
eye.
people judge imagers by the
number
of megapixels, build
wafer fabs with 12-inch (300-millimeter) wafers holding more than twice as
many
chips as current 8-inch wafers and inscribe
the pixel transistors with geometries of 45 nanometers, about a third the size of
most current technologies (130 nanometers).
Sony and Samsung were already planning ries. If
Foveon cameras are
duce Samsungs and Sonys
intrinsically
in
to build
such facto-
cheaper to make, pro-
such volumes that manufacturing
scale dwarfs the imager edge.
Meanwhile, make sure that era are superior ter
how good
all
the other features of your cam-
and multiply the features year by
the Foveon imager
is,
what
is
year.
No
mat-
the chance that
WWL Limited and "Polaroid" will eclipse Canon and the rest of the Japanese in overall camera quality?
And
if
the large-volume
orders go to the Japanese camera giants, the quality of their
imagers, by brute force of scope and scale, will also likely eke
ahead.
George Gilder
270
that
ume
the message from
is
rules quality.
the semiconductor industry. Vol-
Long experience
in the
semiconductor trade
has confirmed the validity of [Nick] Tredennick's Law:
volume and you get
quality.
and eventually quality
Go for
quality
for
and you forgo volume
As the chief designer
as well.
Go
of the high-
quality Motorola "68000" microprocessor, Nick knows whereof
he speaks. Chiefly used computers,
relatively
in
low-volume Macintosh
was eventually eclipsed by once-inferior PCs
it
based on Intel 8600 family microprocessors. The x86 line sold
in
such volumes that they attracted enough investment and learning-curve improvement to
Now
same process
the
non dooms Foveon
become dominant is
in quality as well.
befalling cameras. This
phenome-
to a "spiral of decline," declares Karl Guttag,
an eminent former Texas Instruments inventor/engineer with 123 patents mostly in graphics and imaging.
Foveon on gain the
DP
of
Review, Guttag believes that Foveon can never
momentum
Yet the
A relentless critic
to
make money.
power of volume
will
be the ultimate source of
Foveon's coming dominance in imagers. All the talk of pure
camera refinements in the future,
will
dwindle into commercial irrelevance
when high-volume imager
leadership
is
set in
the market for cell phones. In 2003, the world's largest unit seller of
phone
cameras was already not Canon or Sony but
titan
cell-
Nokia. In the course of time, the highest-quality
imagers for the mass market would
come from
the cell-phone
imager market, already 150 million units in 2004 and set to
double again
in
2005.
The
All the greater simplicity
Silicon
CCDs
in cell
will ultimately
phones. The power con-
prove to be a showstopper for
CCDs,
battery-based devices. Against yields a fivefold
271
and elegance of the Foveon imager
can translate into a decisive edge
sumption of
Eye
RAW efficiency low-power CMOS
Foveon's
power advantage. Against
(complementary metal oxide semiconductor) imagers, the Foveon stacked
pixels yield at least a twofold resolution
accuracy advantage, and the
means
a decisive advantage in
ability
and
simply to cluster pixels
camcorder applications.
So emerge the lineaments of the coming world conflict photography. Foveon
first
would face
off against the
in
photography
establishment of Canon, Sony, Nikon, and Olympus backed up
by a second team of
Fuji, Konica, Minolta, Pentax,
Packard, and Kodak, and then
it
would have
Hewlett-
to confront the
imperial pretensions of Nokia, Samsung, Kyocera, Casio, and a
host of others rushing into the cell-phone space. Nearing midpoint in the essentially
first
decade of the new century the score remained
100 percent market share for the establishment
team, essentially 100 percent for Bayer mosaics. Fighting alone against this massive array of established as Clayton Christensen declared four years earlier, "zero" to
chance of success. As Faggin
says,
"Obviously
rivals,
Foveon has
we
are going
need partners." For the necessary reinforcements, he was
clearly not going to
depend
chiefly
on National Semiconductor,
Sigma, and the Character Group. At the show in Vegas, he was
busy seeking partners and finding more every ners would take the
company
into the
day. Foveon's part-
new markets
for low-
power, high-resolution imagers where the war would be won.
George Gilder
272
The argument ter
for
Foveon
is
that
it is
intrinsically simply a bet-
technology in every way. For Foveon, therefore, the chief
shadow on the show was the notion
that the best technology
often does not win. In the Times, John Markoff reported speculation that
Foveon might become the "Betamax" of the
digital
image revolution. The difference between Foveon and Betamax, though,
is
superior to rior to
that
Foveon
CCDs
CMOS
indeed decisively superior:
is
power efficiency and some
in
mosaics in efficiency of
five
times
2.5 times supe-
capture and far
light
superior in superfast full-motion and stop-motion functions,
adding up to the necessary tenfold edge. Indeed for power, high-resolution applications, there the case of Foveon, there
CMOS lacks.
is
no tradeoff
in
no
low-
alternative. In
which
CCDs
or
mosaics offer some virtue that the Foveon technology
The
superiority of the
momentum, brand financial
Canon, Sony, Nikon lineup
new world where
clout. All are
powerful assets, but they intrinsic
doubt or doldrums,
I
am
of
its reveille,
which
is
my
telecosmic
inclined to be churlish. But
following Mead's call to "listen to the technology," rising in joy at
technology in
imaged or protected.
with foveon challenged and many in
their
the camera disappears into the fabric of every
location that needs to be
champions
is
manufacturing prowess, and
recognition,
and marketing
do not eclipse the superiority of Foveon's a
is
many
all
I
still
find myself
sounding with new urgency.
Like the railroads that bankrupted a previous generation of visionary entrepreneurs
and
built the foundations of
an industrial nation,
The
fiber-optic less
Silicon
Eye
273
webs, store width breakthroughs, data centers, and wire-
systems installed over the
last five
years will enable and
endow
the next generation of entrepreneurial wealth.
Mead sums
up: "During the bubble
starting Foveon.
rience.
It
hire people.
their chests
didn't
all
middle of
was an absolutely awful expe-
struggled.
have a business plan, but were pounding
and going public
The hardest
thing
company going during
"Now," he continues,
I
months, and you could
in three
We
struggled and struggled and
ever had to do in
my
life
was
"there's
deals with people; you can
you.
It's
telling
to
the bubble.
space available; you can get fab
runs; you can get vendors to answer the phone. You can
whole time
to
being sucked off into these brain-dead
start a sensible business.
get a
It
in the
You couldn't get fab runs. You couldn't
call.
They were
companies that
not
horrible.
was
You couldn't find space. You couldn't get your vendors
answer the phone
on
was
I
sit
down and
make
they don't spend their
you how they're a hundred times smarter than
absolutely amazing. You can actually get
which means what's happening now
is
work done now,
that the entrepreneurs,
the technologists, are building the next generation of technology that isn't visible yet but
upon which
will
be built the biggest
expansion of productivity the world has ever seen." of the network,
bring radical
new semiconductor and
new
On
the edge
optical innovations will
technologies to market across the span of
microcosm and telecosm. Synaptics has already taken 70 percent of the touch-pad busi-
ness and devices.
now
supplies haptic technology for iPods and other
Foveon could ultimately take a similar share of the
George Gilder
274
imager market. National Semiconductor retains 30 percent of
Foveon and
is
manufacturing
its
revolutionary devices in a lead-
new
ing-edge fab in Portland, Maine. Foveon's are the single tronics riott at
most elegant new commercial invention
have encountered since
I
silicon imagers
Newark
meeting
New Jersey the Web as it
Airport in
dominate the next era of vessel of images
first
and videos superior
Mead
in elec-
Mar-
at the
1983. Foveon can
in
becomes
in resolution
broadband
a
and quality
to
analog films. Following Synaptics and Foveon will be several other
companies. Chaired by by
Mead
Mead and based on
student Chris Di Orio
self-adaptive semiconductors.
is
Mead
technology designed
Impinj, a radical innovator in
Located
in Seattle,
Impinj has
achieved the goal of robust analog floating-gate technology envisaged by Federico Faggin at the launch of Synaptics. Impinj chips can be trimmed or adapted after production by adjusting the charges on the floating gates.
One
be a tiny analog radio frequency
tag,
of the
first
products will
usable everywhere from
superstores to transport containers. But with the aim of drastically lowering the
applying
its
power usage of
cell
phones, Impinj
is
also
technology to an array of low-power mixed-signal
devices.
A further Mead-inspired venture is Audience, able breakthrough in speech recognition.
offering a prob-
Launched by Lloyd
Watts, formerly at Synaptics, and based in part on Mead's
cochlear researches and on Dick Lyon's cochlear model, Audi-
ence shows that the key problem of speech recognizers digital pattern
matching but analog audio reception.
A
is
not
kindred
The
Silicon
Eye
275
company, also associated with Mead's cochlear researches, called Sonic Innovations, has created directional hearing aids.
Emerging
in
1999
in Salt
Lake
City,
has
it
become
the fastest-
growing company in the global hearing-aid business. This jumble of apparently unrelated ventures embodies the singular
new
vision unleashed by
in his classes at
Mead some
Caltech and brought
twenty years ago
to diverse fruition
by an
amazingly ingenious cohort of his students and associates from
Misha
to Lyon.
There have been many setbacks on the long
toward the Foveon revolution, and there
will
trail
be further setbacks
ahead. But these and hundreds of other comparable innovations signal the first hot flare of revival
millennial crash in technology.
from the devastation of the
Epilogue
Cameras End
moments
In
Merrill,
of weakness
and Lyon
at
and enthusiasm, Mead, Faggin,
Foveon speak of transforming the cam-
era business. Perhaps, en passant, they goal.
But
as a target,
and practical goal fuse
it
is
will lead
it
to
them
astray.
may
A
achieve this
more
plausible
dismantle the camera altogether and
dif-
throughout the world.
Both
in the
etymology of the word and in the history of the
device, a "camera"
is
a
curved room, a vaulted chamber, an
arched vestibule, a cavernous black box. In camera the judge
meets privately with Perry Mason, excesses. In a camera obscura
shone on side.
a wall
—an
to
upbraid him for rhetorical
original dark
from a lensed hole that passed
Often made
in the
form
room
—
light in
pictures
from out-
of a circular building, the
camera
obscura enclosed a group of observers around a plain white table,
on which a luminous image was projected by a lens above roof.
By
rotating the lens around, the "camera" casts
in the
on the table
a periscopic view of the surrounding scene. In other words, the
camera began
as a theater, bigger even
than a mainframe computer. In the case of camera lucida
277
(a light
George Gilder
278
room), cameras project an image on paper as an aid to sketching or drawing.
Made
in the
shape of a cone with a lens and a
and
reflecting mirror or prism at the apex
a
drawing table inside,
these devices allowed painters to trace out a sketch of the scene.
At an opening on one side of the "camera" the
artist sits, partly
enveloped by a dark curtain. The camera lucida caused a small
media scandal
in
2002 when the
New
York Times revealed that
several great artists used this crutch as the first step in their
painting.
Technological advance eventually banished the
artist
from the
room, allowing considerable miniaturization and automation. Replacing him was a
silver halide
plate or film. In 1839, in Paris,
homunculus,
came
L.
J.
a light-sensitive
M. Daguerre and
the
daguerrotype, using actinic effects (the impact of radiant energy
on certain metals)
images through chemistry
to create better
7 .
Imparting a distinctive jaundice or sepia look to these pictures
were the vapors of iodine used
With the daguerrotype and
in developing
its
them.
descendents, the portrait pho-
hooded hangman
tographer
became
subjects.
Despite letting the victims remain unhooded, this
a
fixing stiff
images of posed
process inflicted a facial malady that might be called rigor photos,
hardly distinguishable that has
tis
now
befallen
much all
of the time from the rigor mor-
the subjects of these otherwise
benign attentions.
The
sepia saga of portable "boxes," vaulted chambers,
hangmen, canopied cal
baths,
cavities,
arched
little
caves of
light,
hooded chemi-
emulsions, tripods, and films dwindled over the
decades into sleek metal vessels,
full
of tiny lenses and mirrors,
The
that could easily be held in
Silicon
Eye
279
two hands, and then into miniature
Minoltas that could be held between two fingers of one hand.
Now,
in the pattern of the
computer, the camera
ther onto a sliver of silicon and
mon. Then wires.
The
it
become sand cheap and com-
will diffuse across the
final liberation will
will shrink fur-
network, often connected by
dispense with the wires, linking
the one-chip silicon imagers through the air to storage facilities
capable of holding their voluminous reports.
The
future after Foveon will confine the old paradigm of
vaulted 'cameras" chiefly to images that cannot be seen.
Still
needing a curved protective chamber rather than a mostly bare microchip
be tools for semiconductor photolithography
will
and etching, scanning electron tunneling microscopes, nuclear resonance imagers, Computerized Axial Tomography (CAT) scanners,
X
rays,
and inspection machines
for
baggage and
shipping containers. But even in these applications, the chip itself will increasingly
bear lenses and other apparatus required
for capturing high-resolution
images wherever they are needed.
Already available from National Semiconductor are chips in the form of pellets that are
which take and
a series of
consumed
orally
by the user and
photographs of the esophagus, stomach,
and then are unceremoniously extruded and
intestines,
flushed along with other
offal,
having wirelessly transmitted
their report.
Foveon can do puter:
Reduce
and disperse
it
it
for the
to a chip
camera what and make
it
Intel did for the
com-
ubiquitous. Dismantle
across the network. Render
it
it
wireless, wanton,
and waste-able. Then, as Moore's Law shrinks the actual imager
George Gilder
280
an ever-smaller portion of the chip
to
CPU tem
—
just as
relegated the
it
an ever-smaller corner of the modern single-chip
to
— the
Foveon device
longer merely a sensor,
will it
assume ever-more functions. aim toward
will
evolve into a vision system.
It
sys-
will
intelligence.
become something
It
No will
of an eye,
something of a brain.
That
is,
it
will
no longer serve merely
to reflect the visible
aspects of the world. Attaining powers of recognition and pattern
matching,
it
movements,
will identify
prints, faces, scenarios, poisons,
threats, anomalies, finger-
weapons.
manufacturing processes and descry trends will
sit
in
ambient
traffic. It
prevent automobile accidents by recognizing dangerous pat-
terns.
his
will find defects in
It
It
will
help the ornithologist find his rare bird, the hunter
dangerous beast, and the rescuer her
lost child. It will
baby-
and house-sit. It will
be described as an enemy of privacy. But
it
will
enable
us to defend privacy against the muggers, rapists, and terrorists
who would most
brutally rend
it.
It
will eliminate
most
false
charges that rip open the privacy and smear the reputation of
nonmuggers and nonrapists.
It
will allow us to take
the knowledge that our fate can be observed police will
more often be
in
—
of
human
and
reach and informed.
history people lived in small
cally difficult to escape,
risks in
that doctors
Today human beings have more privacy than
much
more
ever.
Through
towns economi-
where repute remained
at the
mercy of
rumor, and from time to time, delusionary vigilantes burned a
witch or lynched an adulterer. The increasing ubiquity of imagers will
empower
us to
document our
lives
and prove our
The
innocence against ators,
Eye
281
false charges, protect ourselves against pred-
and enable more
many long-term for Foveon.
Silicon
reliable
and
challenges
just
enforcement of laws.
ahead
lie
To save the databases
and
for the industry
of the world from diluvian
exabytes of image data, pouring in everywhere from billions of highresolution chips gushing pixels twenty-four hours a day, graphic intelligence
must be
intelligence
was distributed and
still
distributed
and
localized just as
by the PC. Capable of
localized
and motion coverage, the imagers
will
have to select their
gets, identify their subjects,
and
remitting endless raw
Foveon s chip
files,
Ironically these roles will lure the
will necessarily
selectivity
it
when Misha Mahowald when
signaling.
the
birth in Carverland,
Delbriick burst in upon the young Carver
schematic of the retina, and
Mead
created her
with first
the band of brothers began
their pursuit of brain science. Revived will intelligent
and
move
company back toward
kind of neuroscientific missions that gave
a biological challenge,
tar-
interpret the scene. Rather than
toward recognition and pattern matching,
when Max
computer
be the challenges of
imaging embodied in the neuromorphic devices and
neural-network recognizers that the company pursued under Federico Faggin in In one of his
explains
how
its
original incarnation as Synaptics.
freshman physics
lectures, Richard
the retina develops in the
embryo
as
Feynman
an extrusion
of brain tissue, with long fibers later growing back to link the
eyes to the visual cortex. Through the retina, as
Feynman quotes
an unknown observer, "the brain has found a way to look out into
George Gilder
282
the world."
Thus the
retina
is
a
window not
the realms of light but also inward into the "the whole It
is
Mead
only outward into
life
of the brain and
problem of physiology."
the prime example of the "transducer physiology" that
studied with Delbriick.
technology
—the
interface
It is
between the outside world and inside
computation, the links between thought.
It
the key challenge of analog
light
and
logic,
sensation and
addresses the ways in which analogies for the shapes
of things arise in the brain.
And
it
offers a
new way
to
under-
—the way
that
the computer can extrude photosensitive silicon and find a
way
stand the analogies between retina and camera
to look out into the world.
like the retina, the Foveon camera must fibers" to link
way
it
back
to its users
to look out into the world.
and
make
their
way
give the optical
network a
To be stored or transported,
Foveon pictures must be converted imagers are to
also find "long
to digital form. If
into the
Foveon
hands of every hobbyist
and onto every computer videoconferencing terminal and into every surveillance application, from convenience stores to ports, the pictures will require
and endow an abundance of
storewidth: voluminous storage linked to
The
natural
way
to process
air-
images
is
immense bandwidth. optical.
With much of
the information in the world interpretable as images, transmitted as images across fiber-optic lines, and stored as images in digital
video disks and image databases,
of the future world
much
of the processing
economy should move toward analog
optics.
The
The
advantage of It is
light
and image
is
283
wavelength division multiplexing.
wave network rather than
a
Eye
technology exploiting the parallel
optical
great
first
Silicon
a bit network. Exploiting the
wave network combines many
natural parallelism of light, the
different "colors" of infrared radiation, each bearing the equivalent of billions of bits per second,
width of a
human
But why
"billions of bits"
all
all-optical
—analog
convert
its
if it is
an analog system? The source
wave net
is its
indifference to content.
network can transmit any kind of information
or digital
—without
distortion;
waves into any readable form
In all-optical
down
a single fiber thread the
hair.
of the superiority of the
An
on
it
does not have
at to
until their destination.
wave networks, the different-colored streams pass
passive analog optical paths that perform in parallel
all
the
functions of active switching and multiplexing done in serial digital
form
in
mixed optical and electronic networks.
have to read the
where
traffic
digital
light,
or dropped.
are self-addressed in the very colors of the
the frequencies of the waves.
Each frequency designates
different path to a particular terminal. Like a
WDM
is
is
wave network
also
a
Foveon camera,
an inherently analog system optimized for the defining
and transmission of optic
networks
addresses on every packet at every point
must be added
Wave networks
Bit
colors. is
It is
camera on
a
a country.
The
fiber-
not merely a communications medium.
an analog processing path
at the heart of a
still
It
massively
digital Internet.
Despite the optical depression of the early 2000s, the superiority of the
wave network grows
steadily.
New
systems in prepa-
2,s
(
4
bear as
ration
many
as
-corge Gilder
one thousand wavelengths. Current 11,000-fold advance in
Corvis equipment represents an
six
years, a rate of well over four doublings every year. Parallelism
pays.
With more than 1,000
fibers
now sheathed
in a single
cable and 1,000 wavelengths per fiber and 10 gigabits equivalent
per wavelength, a single fiber installation will soon be able to carry over a petabit,
more than
worth of 2003
a full day's
Inter-
net traffic, in one second. Emerging will be a global Foveal econ-
omy, engaged in dense image
traffic,
teleconferencing in high
resolution, with full exploitation of the parallel advantage of light
and image.
As image-bearing wavelengths become asymptotically
free,
they will be wasted in unexpected ways, enabling global simulations
and experiences and transcending the
beings in time and space. Counting the
needed
to
accommodate some
width consumption
computers needed ing the
number
of
is
number
human
of wavelengths
extrapolation of current band-
tantamount
in the
isolation of
to
counting the number of
mainframe world of 1960, or
steam engines needed
to
tabulat-
run mines and fac-
tories in 1790.
The continuing prevalence
of the routine digital
camera and
the digital switch in fact represent a forced stopgap on the to
way
analog solutions that respond to the physics of the media.
image or a path, a
surveiller or a
map,
tor or a retinal recognizer, a routing
none of these
is
calculations
happen
explained to
me on
naturally
and
a spectrographic calcula-
scheme
intrinsically digital at
An
all.
or a pattern matcher,
In analog form, these
instantly.
As Misha Mahowald
our long walk in the mountains of Pasadena,
The
your ears and eyes do
it
Silicon
constantly.
Eye
285
'They do not even know
its
hard."
By the inexorable evolution of the imaging
will
become
the analog
first
industry,
Foveon s color
step in a long process of
cerebration that will end in simulating ever larger reaches of the
human
brain and extending back over fibers into a
consciousness suffused with color and
That was the
original
dream, and
new
global
light.
it is
the continuing quest.
Glossary
—A device
Amplifier
for increasing the strength of
by allowing a small force on a control element in the circuit.
force
it
The proportion between
controls
a faucet, in
called the amplifier's gain.
A
hydraulic analogy
is
flow.
a continuous electronic representation of an image,
sound, or pressure, simulating
waveform
the small force and the large
which the force on the handle controls the
— Giving
Analog
is
an electrical signal
to govern a large force
using
it
(rather than translating
it
all
points on an electrical
into digital bits
and bytes).
Rather than "crunching numbers" as digital devices do, analog devices imitate real-world inputs in electrical form. Examples are film cameras, watches with
evision sets,
and human eyes and
digital representations
tion
hands
to indicate the time, ordinary tel-
ears.
Nearly
and then convert them
all
computers employ
to analog for observa-
by mostly analog humans.
Analog- to-digital converter (ADC) inputs into digital outputs.
(DAC), which converts Aperture
—A device
Compare digital-to-analog converter
digital inputs to
—A hole admitting
for converting analog
light. It is
287
analog outputs.
measured
as the effective diam-
Glossary
288
eter of a lens, usually translated in photography as the ratio of focal
length
to aperture diameter, familiar as the
Bayer pattern filters as
—A prevalent mosaic with two times
red or blue
retinal sensitivities. is
Bit
used
—
f-number
filters, It
of a lens.
as
many green
a proportion that correlates with
was invented by Bryce Bayer
in nearly all digital imagers,
human
1976 and now
in
except Foveon's.
In information theory, the smallest unit of information, repre-
senting a zero or a one, on or
Black lipid bilayer biological
off.
— Synthetic chemicals
that function identically to
membranes, which sheathe nerves
substances
and both
in "bilayers"
protect cells and interface to other cells. Carver
Mead used
these
experiments comparing communications across
in
nerve channels with communication across transistors. Broca's Area
—A region
in the left frontal lobe of the brain that
is
active in the formation of language.
— Eight Capacitor—An Byte
bits.
sisting of
two conductors (bread
(ham) that can
briefly store
of a current.
A
memory
analogous to a
electrical device
microphones
direct current
sandwich, con-
separated by an insulator
slices)
an electrical charge or hold up the flow
ubiquitous electronic element used in dynamic
cells to store a bit, in
a finger, in
ham
(DC)
touch pads
to transmit a
reception of light intensities.
capacitors that collect
—A
It
movement
of
sound, and in converters of
to alternating current
Charge-coupled device (CCD)
to identify
silicon
(AC).
imager specialized for the
consists of a grid of light-sensitive
electric charge
and couple
capacitors in a chain that delivers an image to a
to
neighboring
memory
in a serial
stream of pixel values.
Complementary metal oxide semiconductor (CMOS)
—The
basic design of most microchip circuits, combining two comple-
mentary transistors with
a single input.
When
one transistor
is
— Glossary
289
turned on by an electrical charge, the other time at least half the transistors are
Cone
—A
light detector in the
encode trichromatic color with
Die
At any one
thus saving power.
off,
Cones come
their
wavelengths of
in three forms, to
broad and smooth but differing
The
light.
retina contains 6 mil-
—
measured
Electrical flow,
in
amperes ("amps"), analogous
water current in hydraulic systems.
—A
its
off.
cones (and 120 million rods).
lion
Current to
turned
retina that registers brightness at mod-
erate-to-high illumination levels.
sensitivities to the
is
single
unpackaged semiconductor device.
plastic or
becomes Digital
a
When
it
acquires
ceramic package and pins or other connectors,
microchip or just "chip."
Its
plural
is
it
also die.
— Numerical representation—based on bits and bytes—
that
uses only two points on a waveform or two positions on a switch (on or off) to convey numerical information.
known
as
Digital
computers are
"number crunchers."
(DSP)
Digital signal processor fast multiplication
—A
silicon processor that
and accumulation functions, which enable
handle signals, such as audio and video,
and convert them
performs
into a
DSPs perform much
form suitable
of the
work
in real
time as they
it
to
arrive,
for display or transmission.
in digital
cameras, where they
transform a digital stream of light measurements into a usable image.
Digital subscriber line
(DSL)
— Method of using an ordinary cop-
per phone line to send between 200 kilobits per second and 8
megabits per second, or in advanced systems up to 52 megabits per second.
Digital-to-analog
converter
(DAC)
See
analog-to-digital
converter.
Fan-in
— Number
connectivity.
of
inputs
that
a
device
can receive.
Inbound
Glossary
290
Fan-out
— Number
of outputs a device can transmit.
Outbound
con-,
nectivity.
—
memory The memory technology commonly used to store images in a digital camera. A nonvolatile memory chip that stores
Flash
values on an array of insulated capacitors called "floating gates"
because they represent an island of capacitance "floating"
con dioxide insulator with no active or leaky connections
memory
cuitry of the
exotic
quantum
grid
"tunneling" or "avalanche" effects.
—An imaginary device
Future that
was alleged
in
Steven Spielberg's film Back
to enable time
ever, accurately describes the
memory
travel.
cell in a
is
in constant flux in
in
which
CCD
cells also
to a
might be
flux capacitors.
—The
F-number
ratio of the
focal length of a lens to the diameter of
aperture. The smaller the f-number the larger the aperture and
its
the is
DRAM,
capacitors connected
microscopic grid of wires by transistors.
termed
The name, how-
computer memory
chip called a dynamic random access memory, or electric charge
sili-
on the chip. They are read or written by
Flux capacitor to the
in a
to the cir-
a
more
light
Focal length lens
enabled to pass.
rough index of lens
Maximum
aperture (low f-number)
quality.
—Gauges the distance between
a calculable point in a
and that point of focus where the image
is
sharpest. Focal
lengths are measured in millimeters. Small focal lengths
mean
wide-angle lenses, while large focal lengths are associated with tele-
photo lenses. In most lenses, focal lengths are
fixed,
but variable
focal-length lenses, called "zooms," are increasingly effective
and
popular. Together with the size of the imager's light-sensitive area,
the focal length determines a camera's field of view: the picture.
A
short focal length reduces or compresses the size of the image pro-
jected onto the sensor, resulting in
more
of the scene being cap-
.
Glossary
291
An
tured in a given sensor area, thereby increasing the field of view.
increased sensor area will also increase the field of view.
Thus the
benefits of reducing the size of pixels, or of imagers, are limited.
Fovea
—The
part of the retina dedicated to high-resolution color
and the inspiration
vision,
for the
company name Foveon.
an
It is
indented area composed of a dense array of cones directly behind the eye's lens, where parts of the retina.
it is
The
obstructed by less nerve tissue than other
fovea distinguishes detailed lineaments and
colors in an image, such as a crowd.
When you
words on
page or features of a face
in
focus on an image directly in front of your eyes,
you are using your fovea, but illumination, because
Fry's
a
it
it
doesn't
work
in conditions of
low
has no rods.
—A famous chain of electronics
stores in Silicon Valley, south of
San Francisco.
Graphics Interchange Format (GIF) digital images, limited to
Imager ing
—
its
256
frequency
The key component
of a
charge-coupled device
cameras the imager
—A
and measur-
(color), intensity (energy in photons),
grid.
is
for
colors.
In digital cameras, a device for detecting light
on a
Ion
—A compression scheme
camera,
in a digital
or a
CMOS
and location
takes the form
it
chip. In
most analog
essentially film.
charged atom or molecule having gained (anion) or
lost
outside band of electrons, which deter-
(cation)
one or more of
mine
chemical behavior. The nerve channels of the brain com-
its
municate
electrically
its
by ionic conduction.
JPEG — Developed by the
Joint Photographies Expert Group,
JPEG
widely used to designate a method of compression of digital
images for transfer or tion),
JPEG
is
display,
but because
it is
lossy
(it
is
still
loses informa-
unsuitable for medical images where the details must
be exactly accurate. The compression
ratio
can be as high as 40
to
1
Glossary
292
Microprocessor
—A digital computing device inscribed on
a silicon
microchip and providing the "central processing unit" (CPU) personal computer or an of
equipment such
Neumann Mosaic
(also
[CFA])
embedded processing
in a
unit in other kinds
von
as cameras. Microprocessors are mostly
machines.
known
—An
as color filter
mosaic [CFM] or color
array of color filters, typically a
imposed on the imager of
filter
Bayer pattern,
array
super-
a digital camera, in order give color to
black-and-white imagers. Mosaics restrict each pixel to a single one of the three primary colors (red, green, and blue). Although that mosaics reproduce the in
humans respond
much
performance of cones
to all three colors,
though
as the three stacked receptors in a
Nanometer
— One
billionth of a meter.
in the eye,
Nerve channel
— An
(violet)
said
cones
in different degrees,
Foveon imager do.
Ten
silicon
atoms
would be roughly a nanometer wide. The wavelengths of
between about 400 nanometers
it is
in a
row
light are
and 700 nanometers
(red).
elongated biological cell through which
sig-
nals are passed as a result of differences in electrical or ionic
charge across the
cell's
membrane.
It is
analogous to a wire in a
computer.
Nerve membrane
—A
sheath consisting of fatty molecules called
phospholipids surrounding a nerve interfaces to other cells.
It is
to the silicon dioxide layer
on
as a
—A network
cells
and
analogous but not electrically similar
both insulates the
chemically.
it
of artificial
computer system. Based on
functions of a brain,
both protects
a silicon chip that
device electrically and protects
Neural network
cell. It
neurons
a cartoon
or
neurodes used
image of the processing
some neural networks can outperform ordinary
von Neumann computers
in functions of pattern
nal processing such as imaging or
matching or
speech recognition.
sig-
Glossary
Neurode
—A simulated neuron used
293
in a
neural network.
It
sums
inputs, using a threshhold level as a nonlinear trigger for decision
making.
Neuron
—One
of the processing cells of a brain. According to various
estimates by neuroscientists, the
and
lion
have
a
hundred
much
Nonvolatile
—
the
form of memory that does not lose
power goes
out.
don't
con-
its
Popular nonvolatile memories
Compare Volatile memory.
—The timing of an electromagnetic wave
(often signified by the angle that lar
still
it
or point in
its
progress
subtends on a 360-degree circu-
Two
graph depicting the wave as a rotation around the graph).
waves
bil-
of a clue.
include flash and disk drives.
Phase
brain has between ten
neurons. In other words, they
billion
memory A
when
tents
human
their amplitudes (crests); to the degree
two
waves are out of phase, they interfere destructively (the troughs
will
in
phase
will
add
tend to nullify the crests). In chemistry, however,
phase
signifies the condition of the ele-
ment. For example, the three phases of
H2
are fluid, gaseous,
and
frozen (water, steam, and ice).
—The bottom
Physical layer
level of a seven-layer
model of electronic
and communications equipment, comprising the physical hardware, such
as the silicon
microchip or silica fiber-optic
line.
Above the
physical layer are six levels of "firmware" and software, ending with
the application, such as
Word
or Photoshop, that performs the
com-
puter function by controlling the lower layers.
Pixel
— Short
for "picture element."
The information from one
spot in
an image, or one photodetector in an imager, conceived as a dense array or matrix of spots or photodetectors. Digital
monly measure
cameras com-
their resolution in megapixels (millions of pixels),
but actual resolution
is
determined not by the number of spots but
Glossary
294
by the quality, accuracy, and arrangement of the information delivered from the spots.
Polysilicon
— Noncrystalline
or
amorphous silicon used
chips. Federico Faggin's silicon gate
Quantum
theory
—The
was
for wires
on
in fact polysilicon.
theory of the interactions
between matter on
the atomic scale and light or other electromagnetic waveforms.
Combining waves and
quantum theory coherent.
particles
make
doesn't
in
apparently paradoxical ways,
"sense," but
mathematically
is
it
the theory of the microcosm of invisible
It is
phenomena
as distinguished from the macrocosm of visible things governed by
Newton's laws or
Retina
—A
relativity theory.
densely connected three-layer array of
neurons
back of the eye that gauges the intensity and color of incident
and sends the measurements on
Rod
to the cortex
to the optic nerve,
where they are formed
— Light detector
in the
the
at
light
which passes them
into an image.
retina that registers brightness
levels of illumination, as for night vision.
The
at very low
retina contains 120
million rods but only 6 million cones, suggesting that the
human
race evolved mainly in the dark.
Semiconductor
—An element,
adapted either
to pass or to
such as silicon, that can be readily
block current by doping
its
surface with
appropriate ions and applying a voltage or current. Semiconductors thus
can be made into solid-state electronic switches.
— Silicon Silicon — Pure Silica
oxide, usually
crystalline
common element in
encountered as glass or quartz.
sand used
in microchips.
The second most
the Earth's crust behind oxygen (mixed with
sil-
icon in the microchip's silicon-dioxide insulating layer) and ahead of
aluminum Silicon gate
(bauxite) used in
—The
part of a
most wires on the microchip's surface.
CMOS transistor that controls the flow
of electricity through the circuit. Analogous to the base in a bipolar
Glossary
transistor. Previously
made
295
of metal, the gate region
was
con-
first
verted to silicon in practical devices by Federico Faggin at Fairchild
Semiconductor Corporation and neered
at Intel following a
concept pio-
at Bell Laboratories.
Single lens reflex (SLR)
—An expensive professional camera
that
enables a photographer to control the functions of the camera manually.
—The physics of materials such semiconductors. combining 24 line —A telephone company
Solid state T-l
solid
as
service
voice-oriented
phone
data connection. T-l lines are
now
often rendered by
—A mostly analog device
Transducer form
to
another without altering
sors, photodetectors,
world and
its
64-kilobit
lines into a single 1.544-megabits-per-second
its
DSL.
that changes a signal from one essential value. Eyes, ears, sen-
and other devices interfacing between the
symbolic representations or
among
real
different represen-
tations are transducers.
Transistor
—A
"transfer resistor" or device that controls the flow of
relatively large electrical
on
its
"gate" or "base."
currents by varying the voltage or current
It
can function either as an analog ampli-
fier or as a digital switch. lions of transistors,
A modern
on
a single
hundreds of thousands of analog
Mead
that led to Foveon.
—A computer operating system invented
Unix
team
led by
tran-
microchip for sensory functions was the revolu-
tionary idea of Carver
a
microchip contains mil-
but analog transistors are often discrete (one or
several in a package). Putting sistors
digital
Ken Thompson and Dennis
at Bell
Labs
in
1969 by
Richie. Designed as a
single-user version of MIT's multiuser Multics,
Unix has become
the dominant operating system for powerful workstation computers.
Visible light
—Wavelengths between 400 and 700 nanometers and
frequencies between 430 and 750 terahertz
(trillions of hertz or
Glossary
296
cycles per second). that registers
not
The
portion of the electromagnetic spectrum
on the cones of the retina
deemed by
in the
human
eye and
A
influential phobics to cause cancer.
is
one over
24
10,000,000,000,000,000,000,000,000 (10
portion of the total
)
electromagnetic spectrum.
VLSI
—Very Large Scale
Integration, a phrase used in the microchip
industry to designate devices with
transistors. Essentially
which remains
in
more than
a
few score thousand
today's digital microchips are
all
more common use
in describing chips with scores
ULSI
of millions of transistors than the technically accurate
Large Scale Integration). Carver for creating
VLSI
circuits
VLSI,
Mead
(Ultra
developed design techniques
and then developed analog VLSI devices
that can interface directly with real-world sounds, images, pressures,
Volatile
and other phenomena.
memory
— Memory
to retain its content.
common volatile memories
(DRAM) and Voltage
—
static
that requires a
power source
are the
dynamic random access memory
random access memory (SRAM).
Electrical pressure, analogous to water pressure in hydraulic
systems.
Measured
in volts.
von Neumann machine
—A
computer architecture
storage devices hold both the software program
With
a single path
characterized by the von
and data
Neumann machines
Neumann
the time of the computation
memory
and the
data.
between the processor and the memory, holding
both instructions and data, von
instructions
which the
in
same
many
in order
memory. The most
Contrast Nonvolatile
is
bottleneck.
consumed
in a "Step 'n
That
tend to be
is,
the bulk of
in the process of
Fetchit"
manner
to
moving
and from
locations according to a regular clock or counter. Although
efforts
particular
have been made
neural networks
devices in the world are
still
to
transcend this architecture
—nearly
all
essentially
—
in
the billions of computer
von
Neumann
machines.
Glossary
Wafer
—A
297
silicon disk, typically either 200 or 300 millimeters in
diameter (8 or 12 inches) potentially holding hundreds of die. The wafer
is
Wafer fab
diced into die.
—A chip
factory, often called
simply a "fab."
Acknowledgments
The path
to this
book began twenty years
writing Wealth and Poverty
devoted a chapter to
book
called
One, now
The
I
when
for that
Spirit of Enterprise that
chapter later inspired a
then broke into two books.
was on
available as Recapturing the Spirit of Enterprise,
entrepreneurship.
centered on Micron corporation, a tiny team of
It
chip designers in Boise, Idaho,
who were
establishment in the production of
memory
challenging the Japanese chips at a time
industry observers believed that the industry had Asia. Fifteen years later,
Micron reigned
The second book was
chip company. history
in the course of
encountered the microchip industry and
The researches
it.
ago,
moved
when
inexorably to
as the world's leading
called Microcosm.
It
savvy
memory
explored the
and consequences of the move of industry from the
visible
mechanical realm into the invisible domains of quantum physics.
The
combines the themes of Microcosm and The
Silicon Eye
of Enterprise.
It tells
Silicon Valley that
establishment followed
at a
much
the story of a small entrepreneurial
is
Spirit
company
in
challenging the Japanese camera and imager
time
when most
experts think that this industry has
of the microchip industry in
299
moving inexorably
to Asia.
Acknowledgments
300
Starting as part of Synaptics Corporation in the early 1990s, the imager
project
spun
off into
Foveon
1996. Employing the silicon technol-
in
ogy of the microcosm, both Foveon and Synaptics are engaged effort to forge interconnections
an
between the world of visible things and
the invisible world of the microcosm,
between sensory phenomena and
insensate computation. The Silicon Eye
and engineers have addressed sight
in
this
tells
the story of
how
scientists
sensory and biological domain of
and touch and hearing and linked
it
to the ever-expanding
world
of microchips.
The
journey
scientific leader in this
Gordon and Betty Moore Professor tech.
Hotel
I
first
at
met him twenty
the Newark,
Carver Mead, formerly the
of Science
and Engineering
at Cal-
years ago in the restaurant of the Marriott
New Jersey, Airport,
a trip devoted to teaching his
microchip division of Bell Labs.
new I
world's best teacher of the interplay
and the
is
where he was staying during
chip design techniques to the
quickly discovered that he
is
the
between the physics of microchips
larger world of engineering systems.
Without
his constant
guidance, wise and persistent teaching, and inspiring counsel,
my
would have been impossible and The
Sil-
career as a technology analyst
icon Eye could not have been written. In subsequent years,
I
became
a frequent visitor to Caltech, attend-
ing lectures in Carver's classroom, meeting his students,
with them.
Among the
students
who helped me most
were Misha Mahowald and David Gillespie, both of roles in this book.
Mead
also introduced
me
helping him teach his course on chip design.
to I
and studying
during this period
whom
play crucial
Richard Lyon, spent
who was
many hours both
with Misha and with Dick learning the intricacies of their science. In the early 1990s, at a conference on the biological origins of eco-
nomics based on Michael Rothchild's book Bionomics, Faggin for the
first
I
met Federico
time. Already a legend in the Valley as the creator of
.
Acknowledgments
the
microprocessors at
first
Intel,
301
Faggin told of the promise of the
company, Synaptics, that he had launched
in part to
pursue the ideas of
Carver Mead. Faggin led Synaptics into world leadership for
in
touch pads
mobile computers and other devices. As chairman of Synaptics,
Mead ended up gin
new
and
many
attracting
of his ex-students to the company. Fag-
his wife, Alvia, a technology journalist,
and attentive
in the course of
my
were unfailingly helpful
research for this book.
In telling the story of Synaptics,
Tim
Allen and David Gillespie
offered extensive interviews, documentary support, and valuable per-
sonal insights. Recounting the saga of the early years of Foveon as
it
emerged from Synaptics, Tobi Delbriick and Nick Mascarenhas were crucial. Delbriick also gave a critical reading to the book.
and Dick Lyon were indispensable
Dick Merrill
in explaining the intricacies of
Foveon's technology. In telling the story of the charge-coupled device, I
relied heavily
on Bob Johnstone's pathbreaking
text
We Were Burning:
Japanese Entrepreneurs and the Forging of the Electronic Age
From
W. W. Norton, Jim
Atlas Books and
proved to be book doctors in the Berkshires to give
who make house
me
be rare
in
the Discovery Institute in Seattle, where
ment
as
I
Chapman
worked
my way
me
Cohen
at
home
the kind of detailed and sophisticated edi-
torial attention that is said to
old friend Bruce
Atlas and Jesse
calls, visiting
gave
me
through
contemporary publishing. At I
serve as a senior fellow,
my
continual aid and encourage-
this project.
At the Gilder Tech-
nology Report, Nick Tredennick, Bret Swanson, Mary Collins, and
Charles
Burger
all
offered
Fleischmann became our technology forum
a
constant support and
www.gildertech.com.
Chase kept the business going while
draft with
Sandy
master of the Sigma camera and maintained
at
the author of Wall Street
insights.
I
Tom Owens and
wrote the book.
Andy
Meat and Running Money, improved
an astute critique. Former
MIT professor
Tina
Kessler,
my early
and now business
Acknowledgments
302
magnate Larry Sweet provided
a valuable early reading
orous definition of crucial terms than
I
a
could use here. The
most transformative editor of the book was self the
and
my
more
rig-
first
and
daughter Louisa, her-
author of a forthcoming work on the physics of quantum
entanglement.
My
wife, Xini,
is
at the
source and center of
all
I
do.
— George Gilder January 10, 2005
Index
abundance, scarcity and, 235-36
amplifiers,
ACM
analog, 27, 285,
(Association of
Computing
biological processes and, 181
(analog-to-digital converter), 34,
152, 155,
287
in
additive color projection, 43 airport security,
Alice's
Foveon chips and, 247
Adventures in Wonderland
167, 168-69, 179, 216,
Allen, Tim, 102, 106-8, 117-18, 130,
government
at Synaptics,
ROM
leader,
117-18
analog electronic
141-43, 147, 149-54
techniques
Allesandrini, Walter,
of,
circuitry,
140
analog photodetectors, 54
project,
analog potentiometers, 161
131-32
analog radio frequency
tag,
274
analog technology, neural networks and,
252
26-27
Allied radio catalog, 36
analog-to-digital converter
171
152, 155,
(ADC),
34,
287
analog VLSI, 97, 108, 118, 180
Altera Corporation, 108 cells,
165
analog memory, 104, 155
Alps, touch pad developed by, 164, 169,
amacrine
as,
283
Analog Devices, 201
and Synaptics touch pad 164-71 video
as,
247,248
180
analog chips, 201-2, 229
108
studied by, 139-41
as student
as,
sensory functions
WDM
131, 143, 146, 197
283-84
as advocate of, 97, 161, 166,
neurons
112
at Altera Corporation,
168-69, 179,
181, 182-83, 201, 216,
Mead
(Carroll),
flies
computers, 34-35, 64-65
digital vs., 165, 167,
253
Alcatel,
287
audio reception, 275
Machines), 102
ADC
287
neuromorphic, 175
122
"Analog VLSI and Neural Systems"
Amelio, Gil, 59
American Science Association, 93
(class),
303
108
6
1
Index
3°4
Analog VLSI and Neural Systems (Mead),
Bayer patterns, 268, 271, 288
Analog VLSI System for Stereoscopic Vision,
An (Mahowald), 181-82
Analytic Animations, 197
Betamax videotapes, 241, 272 biology, 122, 185,
Anderson, Janeen, 143
biosensors, 131
annealing processes, 63-64, 65, 144-45,
bipolar cells, 122
287-88
Bisset, Steve,
Apple Computer, 42, 59, 143, 192, 205
bits,
lis,
160-61, 163-64, 171
288 248
Bjorn, Vance,
241
Powerbookof, 164, 169
black lipid bilayer, 288 blind source separation, 148
applications, 83
Application Specific Integrated Circuits
(ASICs), 140, 169 Ardrey, Robert,
1
Blue Arc, 252 blur
filters,
268
Boeing, 195, 197
1
Boggs, David, 42
Armstrong, Neil, 77 artificial intelligence,
90-91
Bohr, Niels, 23, 24
Artificial Intelligence
Lab (MIT), 91
Bond,
ASICs
(Application Specific Integrated Circuits), 140,
169
Computing Machines (ACM), 102 astrophysics, 118-20
Association of
asynchronous
127
224
Apollo space project, 70, 75, 77
Apple
in,
bipolar transistors, 178, 188, 206-7,
258
171,
186
211-12
color and,
models of the brain
Anderson, Chris, 265
aperture,
58-60, 73
Bell Labs, 24, 33, 41, 57,
108
circuits,
104-5
Phil, 18
Boolean
logic,
90
Borel, Danier'Bobo," 160,
boulders, 53 Boyle, Willard, 58, 59 brain, 29, 42, 104, 120, 176-77, 193,
AT&T, 57
281, 285
Audience, 274-75
biological
audio reception, analog, 275
computational principles
Aurora Market Modeling, 33
electrical patterns in,
Austin High School Roundup, 31-32,
Mead's attempt
autochrome, 43 teller
models
of,
127
to simulate,
neural networks patterned
machines (ATMs), 264
Foveon chips and, 247
brightness, of color, 21
broadband, 246, 247, 274 Broadwing, 75 Future, 47
Broca's Area,
288
bandwidth, 235, 282
buried layers, 78, 80
Barrymore, Drew, 219
Bush, George
Basinger, Kim,
219
Baskett, Forrest,
Bayer
filters,
254
266, 269
127-28,
after, 63,
breadboards, 103, 128, 132, 220
axons, 177
to the
187
89
Avanex, 252-53
Back
of,
179-80
133, 140-41
39-41, 210
automatic
162-63
Born, Max, 23
W,
18-19
Busicom, 78, 79 Business Week, 143 bytes,
288
Index
Caere, 151, 154, 155, 165
economy
California,
canonical cortical microcircuit, 187
106-7
calculators, 60,
of,
70
Carroll, Lewis,
127-36, 139-41, 163, 180, 185,
126, 134
at,
281
computational neurophysics course
at,
at,
115, 116-18,
129-30
Mahowald
Mead
of,
at,
at,
analog imaging picture
14-15, 23-29, 58-59,
182,
Foveon X3 picture
CCDs,
189, 205, 275
CD
at, see
Carverland at,
93-97
at,
116, 117,
neural network research
physics
118-20
at,
imagers
power usage
CFA
277-78
Chandler,
cameras
digital vs. analog, 16,
182-83, 190
95
292
(color filter mosaics), Ariz.,
292
52
chaos, 87
Character Group
237
PLC (CG-PLC),
266,
271
imager joint ventures and, 189
charge-coupled devices (CCDs), 43—44, 57, 58-60,
mosaic, 226, 233, 236, 241, 268, 271,
272
182,226-27, 236,
242-43, 246, 254, 256, 257, 272,
obsolescence
of,
point-and-shoot,
277-80 265-67
prism, 213-22, 223, 224, 225, 227
288 as imager, 57, 59-60, 211
inadequacy
of,
203, 204, 258
power consumption
180
retina technology in, 155
Canon,
274
of,
(color filter array),
cameras, 282
solid-state,
255-56, 271
in, 16,
cellular automata, 88,
CFM
retinal,
27
Foveon chips and, 246, 247, 248, 258,
camera obscura, 111
film, 233,
discs),
270-71
camcorders, 60, 242, 265-67, 274
digital, see digital
185-86
phones, 75, 81, 96, 242, 264
129-30
Caltech Synchrotron Accelerator, 118-19
lucida,
lab,
60
(compact
digital
Ricketts coed dormitory 122,
cell
249
of,
neurology
charge-coupled devices
see
players,
CDs
174, 180,
of,
249
in Oxford's
101-8, 118, 121-22, 148, 161,
Mead's lab
camera
cats:
275
1-22, 125-36
1 1
tomography),
axial
279
29, 34, 41, 102, 204-5,
at,
(computerized
117-18
gendered culture
Lyon
Casio, 271
CAT
121-22
Dabney Hall
12
1
Carverland, 97, 101-8, 111-12, 117,
29,91, 150, 192, 197
tech),
Athenaeum
288
capacitors, 165,
California Institute of Technology (Cal-
120,
3°5
19, 163, 221, 227, 237, 240,
244, 265, 269
EOS-lDs, 233, 236-37, 239, 245-46 of,
233-34, 268, 270,
271, 272
one-chip vertical imager and, 222 pixel size at,
256
as
of,
dominant force
271 in,
234, 242,
269
178
market share
Sony
chemistry, molecular, 88
Christensen, Clayton, 232, 239-44, 247,
271 circuit boards, 161
Cirque, touch pads developed
169 Clark, Mark, 191
by,
164,
1
7
6
5
Index
306
Clauser award, 134
CMOS,
see
204
optical, 42,
computers, 96
clean rooms, 53
complementary metal oxide
semiconductors
analog, 35,
64-65
analog and
digital
aspects
of,
cocaine, 133
based on neural networks, 85
cochleas, 128
digital,
artificial,
26-27, 105, 130, 205,
1
89
(CFM), 292 42-45
color filter mosaics
"Copenhagen
234
Corning, 253
214 biology and, 21 1-12
digital
of,
of,
2
of,
212
of,
207, 209-10,
of,
current,
DAC
214
(digital-to-analog converter), 34,
289
Daguerre, L.
color separation, 2
J.
Dally, William,
Commerce Department,
U.S., 18
communications
106
theory,
Dartmouth
89
University,
Thayer School
dataflow, 177 of,
171-72
complementary' metal oxide semiconduc-
Dean, James, 221 Delbruck, Max, 22, 103, 127-28, 130,
193-94
51, 54-55, 155,
178, 179, 188, 206, 211, 222,
233, 236, 256, 257, 269, 272,
Mead
challenged
by,
23-29, 281-82
Delbruck, Tobi, 22, 26, 102, 140, 143,
288-89
155-56, 216
power consumption
271
in
Carverland, 103, 131
X3 and, 224, 227, 258 pixels, 55-56
at
Foveon, 205-6, 214
of,
complementary
at Synaptics, 187,
189-90, 191
complementary technology, 55
delta-based technology, 153
computerized
de Moulin, Ray, 225, 227
axial
tomography (CAT), 279
computer mice, 159
at,
49
discs (CDs), 27
Compaq, notebook computers
(CMOS),
M., 278
daguerrotype, 278
1
Comdex, 223
tors
289
152,
VGA, 190
compact
64
210-11
primary, 43
prism,
crystalline forms,
CTI, 154-55
39
of,
284
cross-license agreements, 188
212-18,220-22 matrix
129
Corvis, 75,
Crichton, Michael, 50
1
Lyon and science
physics
cortex, 26,
Cray, 72
211
processing
frequency
Faust, The," 24
Cornell University, 33, 89, 252
color photography, 39, colors, 189, 204,
88-89
Consumer Electronics Show, 261 Consumer Reports, 1
292
color filter array (CFA),
(Hillis),
235
connectivity,
Collimated Holes Corp., 214
in,
cones, 122, 179, 211,217, 289
Connection Machine, The
196
collective subconscious,
brightness
29
problem solving and, 90-91
Collective Electrodynamics (Mead),
accuracy
75
inability to see by,
274-75
Colley, Steve,
34-35
dendrites,
213
5
1
Index
Dennard, Robert, 49-51, 58, 59 Design Research
DOS
Dong, Milton, 205
system, 83
Doonesbury, 42
DOS, 83-84
diCaprio, Leonardo, 219 die,
289
Douglas, Rodney, 185-87, 193, 195
differential technology,
analog
165, 167, 168-69, 179,
vs.,
188,
93-94
DSL DSP
microcontrollers, 167
subscriber
(digital
Dvorak keyboards, 241
dynamic random access memory chip
182-83, 190, 204, 206, 229, 242,
(DRAM),
274
188, in
photography, 264-65
systems
43-44 43-44
and,
Edmund
237-38 27-28
247
27
Digital Photo, digital signal
Scientific,
39
Ektachrome, 44, 210
electrical patterns,
179-80
Electrochemical Lab (Tokyo), 60
electromagnetic energy, 21
248
electromagnetic waves,
245-46
1
electromagnetism, in vision, 26
processor (DSP), 41, 212,
289
electronic potentiometers, 150, 165 electronics, wires in, 71
digital subscriber line
electrons, 120, 128, 135, 178
(DSL), 289
digital-to-analog converter
152,
149
electroencephalograms, 127
high-resolution, 180 Digital Persona,
circuit,
electrical engineering, 131
Equipment Corporation, 84 VAX computers of, 107-8
Digital
digital imagers,
41
Edgerton, Harold, 36
as solving analog problem, digital electronics,
satellite,
edge recognizer
mass-market, 255
Law
dynamics, 66
Echo
237
film vs., 233,
51, 57-58, 59, 81, 155,
256
in,
colors and, 28,
Moore's
(DAC),
34,
video discs (DVDs), 27
Power Company, 34
33
energy, 128
neural flow
of,
122
energy equilibrium, 66
Diptera, see flies directional hearing aids,
El Paso
EMC,
289
Di Orio, Chris, 274
275
Enigma code, 90
Women, 134-35
erasable programmable read only
Disneyland, 111-12, 116, 119, 133, 194
(EPROM), 59,81,
Discovering
disruptive technologies,
DNA, 93
212,
Dvorak, John, 16, 261-66
cameras, 16, 27, 60, 75, 81, 96,
CCD
289
289
storage devices, 27
and boom
line),
(digital signal processor), 41,
180
photography, see digital cameras
digital
256
Drucker, Peter, 61, 159 Drucker's Law, 238, 239, 241, 243
memory, 104, 152, 154-55
digital
270
(dynamic random access memory chip), 51, 57-58, 59, 81, 155,
283-84
computers, 34-35, 75, 90-91,
optics,
Review, 267-68,
DRAM
181, 182-83,201, in
DP
153
289
132,
digital,
3°7
239-43
ETH, 186-87 Ethernet, 42
memory
152, 170, 172
9
1
Ind ex
308
eyes,
Mahowald's romance with, 132-34,
243
visible
spectrum and, 21 1-12, 216-17
Eyestorm, 2
194-95 philosophical interests
1
Feynman, Richard, Faggin, Alvia, 78, 85, 251-53, Faggin, Eric,
256
Faggin, Federico, 12, 17, 66, 68, 69-75,
77-85, 138 of,
252-53
81-84 79-81, 149-52
floating gates and,
CEO, 257-59,
Factory,
film,
microchip cameras
flash
memory,
flies,
139-41, 176
Foveon interim
as
Foveon investor and board member,
269,
274
252, 254, 256-59
on Logitech board,
1
neural network and, 96 fluid flow,
as "gentleman investor," 56,
251—53 1
59-60,
162-63 by,
69-72, 79,
81,87, 88,96,97, 141, 240, neural network ideas
95-97, 142
71-75, 78-79,
82
Foveon Corporation, 45, 66, 108, 135,
and American competitiveness, 19 board of directors of, 16, 17, 253-55 business challenge
Synaptics and, 95, 97, 141-45,
147-51, 153-56, 160-64,
project, 160,
162-63,
cell
Zilog Corporation co-founded by,
CEO at
81-82, 85, 141, 253
252
fan-ins,
at,
253-55
54-56
Consumer
Show
Electronics
digital storage and,
of
Semiconductor Corporation, 82
204
239-44
Dvorak and, 261-66 future of, 277-85 Hasselblad investment hybrid
290
feedback loops, 125
by,
282
disruptive technology of,
Sherman, 77
289
fan-outs,
at,
development of imagers
252
59, 60, 73, 74, 77-78,
search
2004, 261
framework, 240
Fairchild
233-49, 268-69
phones and, 246, 247, 248, 258, 270-71
CMOS
167-72, 175
to,
camera manufacturing and. 244
168-72,252,274,281
Fairchild,
290-9
fovea, 45, 122, 291
296 of,
silicon gate process of,
and touch pad
290
f-numbers, 290
252-53, 269
failure
87
flux capacitors, 47, 50, 58,
focal length,
microprocessor invented
Faggin, Marzia,
233, 237
floating gates, 79-81, 149-52, 161, 165,
271,277
Faggin, Marc,
vs.,
290
59. 81,
50
flip flops,
as
210
263-64, 267
FillLite,
153, 154, 156, 167, 169
72,217, 283
fiber optics, Fill
as disenchanted with analog devices,
eclipse of,
74
fiber-optic glass,
belated recognition
141,281-82
106, 107, 121,
252
105, 196, 198
of,
15, 29, 50, 102, 104,
still
in.
215
and motion camera
of,
206,
265-67 imagers
Feinstein, David, 102, 104, 105-6, 140
of, 17, 59,
274, 285; see also
specific imagers
Mahowaldand. 121-22, 125-27, 130
in
Mahowald's death and, 194-98
Kodak's interest
Japanese camera journals, 244-45 in,
218, 222, 225, 228
1
Index
market share
Mead's
of,
16-17, 273-74
3°9
and Synaptics touch-pad
213-19, 225-28, 234, 243-45,
Godel, Kurt, 193
254, 273, 274, 277
Google, 88
Merrill's analog chips at,
201-4
National Semiconductor as investor
in,
Gorman, Greg, 219-21, 223, 226-27 Canon EOS- IDs adopted by, 233, 236,
190-91, 255,271, 274 at
239, 245
Photography Marketing Association of 2004, 261, 263-65, 267-69,
Grant, Mark, 188, 190 graphics interchange format (GIF), 248,
270 at
project, 162,
164-71
15-17, 23, 204-7,
role in,
291
Photokina 2000, 223-24, 226-28
picture size of,
247
prism cameras
at,
Grateful Dead, 143 Greenblatt,
213-22, 223, 224,
225, 227
Adam,
102, 143
Gribble, Glenn, 102, 103, 142, 147-48
Grove, Andrew, 73, 78, 82
Sanyo's consideration
Sigma's adoption of
of,
Gun and
255—56
X3 chip
of, 17,
Rifle Dealers' show,
264
Guttag, Karl, 270
227-28, 236, 254-55, 258, 262-63, 265, 267-68, 271
Halla, Brian, 190,
28—29
silicon as retina at,
single-chip imagers of, 217—18,
221-22;
see also
16— megapixel camera
Hamiltonians, 66
X3 of,
Handbook of Colorimetry, 210 218—22,
226-28
at,
vertical color filter at,
frequency, of color, 2
214-18
hearing aids, directional, 275
Helmholtz, Hermann, 210 heroin, 133
Hesse, Hermann, 114
1
Frohman, Dov, 81
Hewlett-Packard, 271
HP-35
155, 291
Fuji, 19,
Hasselblad, 215, 230
hearing, computer, 96
207
X3 of, see X3 X19 of, 255-56
Fry's,
273
HDTV, 261
with Foveon chip
3-T imager
haptics, 88, 175,
Harvard University, 243
as superior technology, 272; see also pixels,
255
Hamilton, William Rowan, 66
222, 271
Hillis,
scientific calculator of,
Daniel,
106-7
88-89
Hoerni, Jean, 74 ganglia,
Hoff, Ted, 50, 84
122
Garage Technology Ventures, 191 Gates,
Bill,
holes, 178
Holland, 210
83-84, 239, 248
Hopfield, John, 93-94, 96, 102, 104,
Gere, Richard, 219
106, 121, 122, 125-26, 141
Giant, 221 Gilbert, Barrie, 201
Hopfield net, 93-95, 170
Gillespie, David, 101, 102-3, 107,
HP Chipmunk computers,
computer-aided engineering at
of,
1
30
Synaptics, 143, 147-48, 152, 154,
155
102, 107
Hubel, Paul, 263-64
J58
Hubel
FillLite slider bar,
263-64
Hughes, 73 Hunt, R. W. G., 210, 213, 214
5
Index
3 10
I
lurlburt,
Japan, 94, 164, 237, 242, 243, 266
Anya, 26
Huxley, Aldous, 194
IBM, 49-51, Intel chip
58, 72, 214, 241,
used
in,
244-45
CCD as
in,
57, 59-61, 255,
imager
257, 259
244
83
by,
camera journals
cell
phones
memory
PCs, 154
Watson research
labs of,
49
touch pads
John Henry
Ibuka, Masaru, 60
42
iconic user interface,
270-71 in,
70
technology dominated
159-60
track-stick of,
in,
chips
in,
by, 19,
269
164
Newman
Society,
134
Johnson, Craig, 191
image representation, 213
Joint Photographic Experts Group,
imagers, 96, 103, 163-64, 172, 175-83,
Joy, Bill,
262
33
JPEGs, 248, 262, 263, 291
187-92, 291 imaging chips, 187
Jung, Carl, 196
impedance, 37
Justice Department, U.S., 83
Impinj, 274
Kao, Ting, 143
information theory, 106, 182 infrared, 214,
221
Karl Zeiss,
Mrs. Willie, 32
Ingles,
Innovator's
Dilemma, The (Christensen),
239
Kennedy, John F, 75 Kikuchi, Makoto, 60
The (Christensen),
Innovator's Solution,
Kilby, Jack,
Inoue, Hiroshi, 242
Kodak, 19, 163, 164, 183, 210, 225, 227,
inspection machines, 279 Institute for Neuroinformatics,
240, 244, 265, 271
186-87
integrated circuit, 74
color films of, 40, 42-45, 1
69-70, 72, 73, 74,
Intel, 53, 57, 59,
4— megapixel camera
247, 256, 279 "crush" as goal at, 17,
of,
as advisor at,
210 267
16-megapixel
CCD
imager
202
Konica, 265, 271
70-71
4004
chip,
Intel
8088
chip, 83
Intel
8600 microprocessors, 270
Koons,
Jeff,
219
Kramlich, Richard, 252, 253-54 Kyocera, 271
International Solid State Circuits Confer-
Kyushu, Japan, 52
ence of 1971, 58-59 Internet, 27, 70, 246, 248, 262, 274,
ions,
analog
291
iPods,
273
Ipsilon,
33
Iwama, Kazuo, 60
of,
Kodak Pro-Back, 226-27 KolnMesse, 223
1
of,
Intel
digital vs.
in,
by,
218,
222, 225, 228
one-chip vertical imager and, 222
83-84
69-71, 78-82, 141, 253
Santa Clara plant
of,
Foveon technology considered
77-85, 154, 160, 164, 188, 244,
Mead
74
Kodachrome, 42, 44-45, 210, 217
239
Faggin
234
keiretsu, 61
283
285
Laos, 203, 228 laptop computers, 242 laser printers, lasers,
42
60
latch-up devices, 178-79, 188 latents,
73
226-27
1
[ndex
Lau, Jim, 224-25, 254-55, 257
3 11
photographic experiments
Lazzaro, John, 103 leadership, types of,
144-45
&
Northrup computers, 34-35
246
lenses, quality of,
of,
192
neural flow
140, 204-5, 274
Lyon, Verna Mae, 32 L., Jr., 32,
34-38, 40
Lyons, Richard, 66
122
of,
McClelland, David, 89, 96
Linux, 83
& Compass
Lion
see-hear chip at Synaptics,
Lyon, William
Legato, 33
light,
Restaurant, 201, 202,
203, 205
McCullough, 177 McNealy,
Scott, 33
Lloyd, Christopher, 47
magnetic-core memory, 256
LMS
curves, 211
Maher, Mary Anne,
Logitech Corporation:
Mahowald,Al, 113
Faggin on board
of,
156, 159, 162-63
post-Synaptics touch pad project
of,
111-22, 124, 125-36, 139, 141,
160-63
London Stock Exchange, 266 Lonergan Richards,
Long Beach
State,
17,
1 1
Mahowald, Joan, 113, 118, 134 Mahowald, Misha (Michelle), 110,
163-64, 171-72 Synaptics' joint venture with,
148, 192, 275,
astrophysics and,
254
1
284-85
18-20
background and personality
of,
111-12, 113-15
104
Lord of the Rings, The (Tolkien),
LSD,
31,
Lyon, Tom, 32-33
Leary, Timothy, 194
Leeds
of,
36-41, 42, 44-45, 210
Lauger, Peter, 25
1
14
Catholic background
133, 194
112, 114,
of,
116, 120, 134, 194
won
Lucent Technologies, 57
Clauser award
Lumiere, Auguste, 43
Douglas's relationship with, 186, 193,
Lumiere, Louis, 43 Lyon,
Bill, III,
by,
1
34
195
32-33
drug experimentation
of,
118, 133-34,
135, 143, 194
Lyon, Bob, 33
Lyon, David, 32-33
edge recognizer circuit
Lyon, Ellen, 32
emotional problems
Lyon, Jim, 32-33
Feinstein and, 121-22, 125-27, 130,
Lyon, Peggy, 40
Feinstein's
Apple handwriting recognition systems and, 42, 143
Caltech, 29, 41, 102, 204-5, 275
color science of, 207, 209-10, 212-18,
of,
31-41
Foveon, 204-5, 224, 226, 227, 256,
15,
imagers and, 175-76, 180-82, 185
impact of Disneyland ride on,
intellectual frustrations of,
Mead
1
12-13,
209-10
at by,
42,
204
193-94
mentor to, 121-22, 126-34, 143, 193-95
264-66, 267, 277
mouse invented
1
116, 119, 133, 194
education and childhood
optical
with, 132-34,
and gendered culture at Caltech, 116-18, 120, 129-30
220-22
office of,
romance
194-95
44-45, 57, 102, 105
at
149
193-96
194
Lyon, Richard, 17, 28, 29, 30, 31-42,
at
of,
of,
as
Oxford, 136, 185-86, 195,
196
Index
3" Mahowald, Misha {continued)
as
140, 143, 149, 175-76, 186, 198,
Merrill's
meeting with, 51-52, 55
249, 281
Moore's
Law
suicide in
Switzerland, 186-87, 192-97 Sheila,
1
to,
18-19 by, 24-25 96-97
nerve channels studied
13
neural networks and,
marijuana, 133 Marini, Giacomo, 160, 162 failure,
and, 15
National Medal of Technology awarded
195-97
of,
Mahowald,
market
Mahowald's mentor, 121-22, 126-34, 143, 193-95
retina project of, 120-22, 125-31,
romantic partner
Markoff, John. 16, 265, 272
Smith, Barbara
150, 152, 154-56, 191,
Martin, Kevan, 185, 187
by,
206
17-18, 257
128-29, 186, 249
"toy retinas" of, 125,
Massachusetts Institute of Technology
Lab
24
transistor built by, 14,
megapixels, 267
(MIT), 26, 89, 180, 210 Artificial Intelligence
91
at,
Megatest, 161
memory:
massively parallel processing, 89
Math works, 216
magnetic-core, 256
matrix, of colors, 39,
processing power and, 88-89
212
Matson, Lawrence, 267-68
memory
Maxwell, James Clerk, 43-44, 210
Merrill, Richard, 17, 29, 46,
Mead, Carver,
i24, i46, 202, 224, 267,
272-75
167, 168-69, 179, 216, 247, 248
140-41
of.
54—56,
201, 212-13
education and work experience
of,
at
Foveon, 201-4, 205-7, 210, 222.
277
as business consultant
companies,
and founder of
15, 19,
274-75
as Caltech professor, 14-15,
23-29,
58-59, 101-8, 118, 121-22, 148, 161, 189,205, 275
on
CCDs, 58-59
on
digital
23, 206,
of, 19,
163-64, 172, 175-83,
187-92, 198
digital
disliked by,
56
Synaptics and, 192 created
by.
vertical color filter of,
207
214-18, 222,
246 and X3, 224-30 Merrill, Sang,
203
Metal Oxide Semiconductors (MOS), 49
Carverland
metamerism, 212
41. 4S
Metcalfe, Bob. 42
of,
camera
209
of,
3-T imager
15-17, 23, 204-7,
of, see
Nikon D-l office of,
213-19,225-28,234,243-45, 254, 273, 274, 277
Lvon as student
47-48 Mead's meeting with, 51-52, 55
patents
249
role of,
as intuitive experimental scientist,
203, 228, 248
cameras, 28
Foveon camera and research
laboratory
47-49,
48-49
brain simulations of, 127-28, 133,
of,
57-58
chip design and fabrication
Analog VLSI and, 97, 180
imagers
chips, 49-50,
51-52, 54-56, 57, 66. 190,260
13-19, 57, 65, 71,
12,
as advocate of analog, 97, 161, 166,
Foveon
274
technological relics collected by, 13,
Mascarenhas, Nick, 189, 190, 191, 205 imagers created
of, see
Synaptics and, 66, 141, 142, 143-45,
241
i
Ind ex
Michigan, University
MICR
National Medal of Technology, 18-19
177
of,
3i3
(magnetic image character recog-
National Semiconductor, 51, 52, 53, 56,
148-51, 153
nition),
205,214
59, 73, 202,
Microchip, 167
Advanced Technology Group fab
microchips, 28, 52
as
202
(Gilder), 135,
investor,
of,
225
190-91, 255, 271,
274
microcircuit, canonical cortical, 187
Microcosm
Foveon
in joint
venture with Synaptics,
187-88, 190-92
MicroMonitors, 197
279
Micron, 17
photo pellets
by,
microprocessors, 60, 82-83, 84, 108,
wafer fab
206, 216, 218, 229, 256,
292
171, 244, 253, 270,
274
Faggin's invention of, 69-72, 79, 81,
87, 88, 96, 97, 141-42, 240,
Enterprise Association), 252,
254 Negroponte, Nicholas, 90
neuroscience and, 66, 177
nerve channels, 24-25, 292 nerve membranes, 292
248 Microsoft Windows, 83, 238
Windows
Miller, Bob,
NEA (New
269
Microsoft, 33, 70, 83-84, 241, 244, 247,
Microsoft
at,
nervous system, 139
Network
XP, 248
161-62, 164, 165, 168-69
Mind from Matter?
(Delbriick),
24
Minolta, 265, 271
Files
System (NFS), 33
neural flow, of light and energy, 122 neural functions, 139, 164
neural networks, 19, 42, 63, 85, 102,
miniature cameras
by,
279
139, 140, 142, 149, 150, 151,
Minsky, Marvin, 89-91
152, 170, 172, 177, 197, 281,
MIT,
292
see
Massachusetts Institute of
analog technology and, 26-27
Technology
Mohsen, Amr, 58
Caltech research on, 93-97
molecular chemistry, 88
computations
Moore, Gordon,
as
15, 69, 73, 74, 77, 82,
15, 179, 182, 188,
206,
93—94
floating gates and,
96
general purpose processors, 153-54
237, 279-80 Morris,
66,
Faggin's ideas regarding, 95-97, 142
256 Moore's Law,
in,
computer model, 85, 89-91
Hopfield on, 93-95, 102, 121, 125-26
Desmond, 116
as parallel processors, 154
Morton, Jack, 58 mosaics, 43, 44, 292
neural reflexes, 176
Motorola, 19, 83-84, 270
neural system, 29
Motorola 68000 microprocessor, 83-84,
neurobiology, 121, 131, 136
270
neurochemistry, 179
neurodes, 42, 95, 150, 293
mouse, 42 optical, 42,
neuromorphic analog VLSI, 175
204
multiprotocol storage systems, 33
neuromorphic computations, 187
Murray
neuromorphic imagers, 139, 142, 151,
Hill, N.J.,
59
185, 186, 281
Naked Ape, The
(Morris),
nanometers, 292
NASA, 59
1
16
neuron chip, 130 neurons, 93, 95, 125, 127, 140, 180,
186,213, 293
Index
3H ncuroscience, 24, 118, 120, 125
OS
neutrinos, 189
Oxford University, 136, 185-86, 195,
New
Enterprise Association (NEA), 252,
(operating systems), 83
196
254
Newton handwriting
recognition system,
New York Nikon,
Times, 16, 265, 272, 278
19, 163, 164,
227, 244, 255, 265,
Nikon Coolpix camera, 255
Nikon D-l Nobel
digital
74
cross-licensing of, 73, 188
pattern matchers, 164,
274-75
Mahowald's, 113
PC
87
notebook computers, touch pads
magazine, 16,261, 264
PDP-8, 84 Pentax, 271
"non-von" computers, 87, 88 for,
59-72
personal computers (PCs), 70, 154, 164,
217,240,241,243,270,281
Nova, 106
development
Noyce, Bob, 50, 69, 73, 74, 77, 82,
touch pads
NTSC
78
PBS, 134
nonvolatile memory, 293
1
1
patents, 190
pattern recognition, 88, 91, 95-96, 280
Nokia, 33, 270, 271 nonlinearity,
77
semiconductor industry and, 73
camera, 203, 228
conference, 148 Prize,
parasitic devices,
of,
89
on color prism, 215
271,272 one-chip vertical imager and, 222
NIPS4
Padua, University Papert, Seymour,
42
41-42, 81
of,
160-72
for,
256
Pfeiffer,
television video, 131, 132
phase, 15, 87, 293
nuclear resonance spectroscopy, 127
Michelle, 135
photodetectors, 179, 188, 217
photography:
Ochi, Shigeyuki, 242
color, 39,
OCR
digital
(Optical Character Recognition),
Olympus Pen-F camera, 37 on-chip error correction, 169-70 operating systems (OS), 83
Operation Crush, 83-84 Optical Character Recognition (OCR), 151
(OCLI), 214 optical computers, 180
mouse, 42, 204
optical reading chip, 153
252-53, 272
depression digital,
180
42,44-45, 210 see also cameras;
in,
283-84
Foveon Corporation
Photography Marketing Association 2004,
261,263-65,267-69,270 Photokina 2000, 222, 223-24, 226-28,
254-55 photolithography,
Optical Coatings Laboratory, Inc.
optics,
in,
Lyon's experiments with, 31, 36-41,
77
Olympus, 265, 271
optical
boom
264-65
151 Olivetti,
42-45
technology and
53-54
photonic devices, 180 photons, 120, 179, 189, 211 photoreceptors, 178, 212
PhotoShop, 83, 220, 264 phototaxis, 23
physical layer, 293 physics,
118-20
solid-state, 131
1
Index
"Physics of Computation, The" (course),
3*5
recursions, 193
redundant memory technology, 170
102, 106
McKenna
Public Relations, 69
Pierce, John, 41
Regis
pixels, 28, 42, 131, 182, 183, 189, 190,
register chips, 81
206, 266, 267, 268, 281, 293-94
complementary, 55-56
Reilly,
with Foveon chip, 216, 217-18, 225, 236, 254, 256-57, 264, 271 Piatt,
John, 103, 104-5, 140, 149
at Synaptics, 143, 148, 152,
214 resolution, 152, 190 retina chips, 103, 122, 130, 175-76, 177,
154
(positive/negative) junctions, in
retinas, 26, 29, 122, 164, 179, 217, 229,
281-82,294
semiconductors, 178-79, 204 Poggio, Tomaso, 26 Polaroid,
249
180,
analog, 180-81
plexiform layers, 122
P/N
Margaret Wong, 177
Reproduction of Color (Hunt), 210, 213,
artificial,
26-27, 128, 149, 204
camera from, 155
215
Foveon point-and-shoot and, 265—67,
Mahowald's project on, 120-22,
269
125-31, 140, 143, 149, 175-76,
Polaroid x530, 267 polysilicon, 73, 78,
186,
294
Mead's
198,249,281
"toy,"
125, 128-29, 186
motion detection and, 128-29
Porter, Mildred, 18
Portland, Me., 52
108
silicon,
Post Office, U.S., 151, 154, 155, 165
Rockwell, 59, 60
power, 235
rods, 179,
Company, 33 Princeton University, 32-33
ROM
prisms, 204
Rosenblatt, Frank,
Prediction
colors in,
214-15
Pritchard, Orion, 205,
Samsung, 269, 271 Sanquini, Dick, 187, 190-91, 254
213
280-8
Santa Clara, University of California
Sanyo, 255-56, 265
voltage, 58, 59
scarcity,
Qualcomm,
CDMA 2000
EV-DO
broad-
theory, 15, 24, 106,
Artificial Intelligence
Lab,
42, 102, 105, 143
294
Schrodinger equation, 135 Schwarzenegger, Arnold, 219
tunneling, 213
QWERTY keyboards,
abundance and, 235—36
Schlumberger
band technology, 248
quantum quantum
at,
252
pull voltage, 58
push
89-90
Rush, Allen, 255, 257
214
solid-state,
privacy,
read only memory),
131-32, 170
cameras, 213-22, 224, 225, 227
in
294
(digital
241
Scientific
American, 180, 181,249
see-hear chip, 130, 140 radio frequency tag, analog, Rattazzi, Gianlucca,
read only real
274
252
memory (ROM), 131-32, 170 window (RAW), 263-66,
accurate
267, 271
Seitz,
Chuck, 89
semiconductors, 70, 97, 202, 270, 294 analog, 55
camera chips and, 222 patents and, 73, 188
1
1
4
1
Ind ex
3 i6
SMOP
semiconductors (continued)
274
self-adaptive,
(simply a matter of program-
ming), 91
technological advances
7-Eleven convenience
of,
stores,
15, 1
273
sodium
ions,
179
295
solid state,
5
Shannon, Claude, 106, 182
Foveon camera
Sharp, 242
prisms,
Shima, Masatoshi, 79, 81-82, 84, 253 Sigma, Foveon chip used
by, 17,
227-28,
Sonic Innovations, 275
Sony Corporation,
CCDs
271
204
294
74, 75, 94,
192,215,217,246,294
181, 188,
pixel size at,
234, 269, 270, 271,
Foveon camera,
28-29, 44
256
solid-state division of,
60
2-megapixel camera
of,
1
silicon, 28, 52, 74, 75, 79, 88, 133, 178,
as artificial retina in
of,
272
Sigma 10 camera, 261, 265, 267
silica,
and, 234, 242, 269
market share
Sigma SD9 camera, 244-45 signal processing, 41,
19, 60, 163, 183, 188,
242, 244, 265
236, 254-55, 258, 262, 267-68,
Sigma Photo Pro software, 262-63
spectrum, 21
speech recognizers, 264, 274-75 Spielberg, Steven, 47, 50
282
spin glass, 94, 125
silicon chips, 70,
140
Steppenwolf (Hesse),
71-75, 78-79, 82, 294-95
memory,
silicon physics,
Stone, Sharon, 219
5
222 camera chips,
77,78,81,82,85,
102, 108, 136,
141, 155, 160, 187, 202, 251,
253,254 IPOs in, 252
149
of,
Support Vector Machine, 149
264-65, 280-81, 282
Foveon chips and, 246-47 sustaining technologies, 241, 243
64
silver halide, 42,
single-chip imagers, 217-18,
synapses, triad, 122
221-22
X3, see X3
synaptic lines, 177 Synaptics, 19, 66, 95, 147-56, 216, 221,
single-lens reflex (SLR), 233, 236, 245,
295
252,254, 264, 274, 281 board of directors
265
Caere
16-megapixel imaging, 218-22
SLR,
Sparc 2 servers
superconductivity, 15
surveillance,
"Silicon Visions," 134
Grace,
subconscious, collective, 196
Sun Microsystems, 33
222 Silicon Valley, 29, 33, 50, 52, 59, 63, 66,
Slick,
235
storage,
silicon semiconductors, for
digital,
1 1
stereo matching chips, 131
Stevenson, Larry, 219-20
see also floating gates silicon
267
Sony DSC-F828 Cybershot, 234
photosensitive,
silicon gates,
229
as,
214-15
1
12
see single-lens reflex
OCR
Carverland alumni
187
at,
103, 141-44,
147-50 Foveon as spinoff
Smith, Barbara, 17-18, 194
haptics
Smith, George, 58, 59
imagers
Smithsonian Institution, 90
of,
project of, 151, 154, 155
at, at,
205
of,
190-92
175, 273
163-64, 182, 187-92, 204,
Index
Logitech's partnership with,
touch pad project
at,
160-63
160-72, 175,
MICR
project
for,
148-54,
165, 189
54-55
digital,
180
physics
of,
wires
177
systolic array,
CMOS,
densities of, 188
224-25, 273 Verifone
3i7
in,
121
71-73
Trans Pecos Science
41
Fair,
"Trash 80" personal computer, 83, 142 T-l line, 295
Tredennick, Nick, 270 Tredennick's Law, 270
Taiwan, 149, 164
Tandy Radio Shack,
Tucker, Tom, 102, 118, 143, 147, 197
83, 142
Turing, Alan, 90
technologies:
239-43
disruptive,
Turner, Rich, 205, 219-21, 227,
236-37
241-42, 272
inferior,
243
sustaining, 241,
technology, crash
in,
275
ultraviolet light,
UMC,
Tektronics, 42
53-54
149-50
Telcordia, 57
Uniden, 266
telecosm, 272-73
United Parcel Service (UPS), 151, 155,
teleology,
165
185
teleputers,
242
Universal Turing Machines, 90—91
Telstar satellite, 41
The
Territorial Imperative,
(Ardrey),
1
16
Texas Instruments, 60, 73, 74, 270 Texas Junior
Academy
Unix, 33, 295
UPS
(United Parcel Service), 151, 155,
USB
(universal serial bus),
165
of Sciences, 41
Texas-Pacific Corporation, 253
THC
262
universal serial bus (USB),
262
(tetra-hydro-cannabinol), 133, 194
Vadasz, Les, 78-79
thermodynamics, 106
Thornburn, Jim, 254
Vegas Convention Center, 264
three-shot color, 43
Venture
timing dynamics, in computing,
Verifone, 189
MICR
104-5 Tobermory, 89-90, 91 Tolkien,
J.
Law Group,
191
reading project
of,
148-54,
165
Vermont, University
R.R., 114
of,
49
VLSI
Tomaselli, Giovanni, 265
very large scale integration, see
touch, computer, 96
VGA
(video graphics array), 190, 265,
VHS
videocassettes, 241
touch pads, 160-72, 175, 224-25 trackballs, 159, track-sticks,
169
IBM, 159-60
transducer physiology, 26, 127-28, 130,
282
Foveon chips and, 246, 282
VLSI
transistors, 37, 50,
bipolar, 178, 188, 14,
206-7, 224
24
295-96
(very large scale integration), 26,
57-58, 60, 70, 72, 78,
79,88,97, 111, 150, 295 by Mead,
video teleconferencing, 242
visible light,
transducers, 295
built
267
97, 121, 153, 213,
296
analog, 97, 108, 118, 175, 180 volatile
voltage,
memory, 296
296
Index
3 .8
von Neumann, John, 87-88, 90
wireless "hot spots,"
Neumann von Neumann
Word, 83
von
177,
bottleneck,
88-89
machines, 87-89, 176,
296
wafer fabs, 52-54, 80, 188, 190, 197,
248
World War II, 35, 90 World Wide Licenses Limited (WWLL), 265-66, 269 World Wide Web,
see Internet
226, 242, 253, 273, 297 at National,
206, 216, 218, 229, 256,
X3, 29, 103, 224-30, 236-37, 243-47,
255-56, 262, 266
274 wafers, 53-54, 60, 80, 226,
297
as disruptive technology,
Watts, Lloyd, 143, 274
225-26 X19, 255-56
wave dynamics, 106
Xerox, 42
Waters, John, 221
wavelength division multiplexing
(WDM),
239-43
retina in,
PARC
214, 283-84
weather modeling, 88
(Palo Alto Research Center) of,
41-42, 102
X rays, 279
wetware, 93, 211
WGBH,
Yamaki, Michihiro, 227-28
134
yoga, 134, 194
"White Rabbit," 112 Whitney, Telle, 103, 111 Widlar, Robert, 202
Z-8,
Wiesner, Jerome, 75
Z-80, 82-83, 141-42, 253
WiFi, 248
Zapacosta, Per Luigi, 160, 162
Windows,
83,
238
Zarakov, Eric, 219-20, 263
Wired, 265 wireless bandwidth,
253
Zilog Corporation, 81-82, 141, 253
247-48
Zurich, 185-86, 193
GEORGE GILDER, the best-selling author
of
several
books— including Telecosm,
Microcosm, and The
Spirit of
Enterprise— is the
publisher of the influential Gilder Technology Report.
He
lives in
Tyringham, Massachusetts.
JACKET DESIGN BY BASE ART
CO.
JACKET PHOTOGRAPH BY VARIE / ALT
/
CORBIS
AUTHOR PHOTOGRAPH BY ALAN SOLOMON PRINTED
IN
THE UNITED STATES OF AMERICA
ADVANCE PRAISE FOR
THE SILICON EYE "George Gilder lauds Carver Mead's
'intuitive
his special feel for the timing of the crests
opportunity
in
life
and
describes well Gilder's
meaning
of the
fashion, he
trends
combines
electricity,
of
phase,
and troughs
science and
own discerning eye in
sense
history.'
of
This
for the underlying
science and society.
In
usual Gilder
a compelling narrative that reads like a
brilliant detective story with
deep insights
into the
concept
of
designing our technology to emulate biology, one of the most
important scientific ideas
-RAY KURZWEIL,
of
our time."
author
of
The Age of Intelligent Machines
and The Age of Spiritual Machines
NORTON NEW YORK L
W. W.
•