The
Birth of the Microprocessor - My Recollections
Federico Faggin
Invited paper presented at the Microprocessor Forum San Jose,
CA, October 15, 2001
It is a pleasure to be here
tonight to share with you my microprocessor adventure. Looking back, it seems
to me that all my prior experience prepared me for the design of the first
microprocessor. When I was 19 I co-designed and built a small digital computer
at Olivetti in Italy. I say small, but actually it was the size of a cabinet. It
was 1961. Then, after I studied solid-state physics at the University of Padua,
I became the MOS group leader of SGS-Fairchild in Italy - what is now ST
Microelectronics - where I developed their first MOS process technology and
designed a couple of integrated circuits. I then came to the US in February of
1968, to work for Fairchild Semiconductor in their R&D Laboratory in Palo
Alto. At Fairchild, following some preliminary work done by Tom Klein, I
developed the silicon gate technology, the world's first manufacturing process
for the fabrication of P-channel, self-aligned-gate MOS integrated circuits. I
also designed the first commercial integrated circuit with self-aligned gates
(the Fairchild 3708). With silicon gate it became possible to make integrated
circuits 3 to 5 times faster, 50% to 100% denser, more reliable, and with much
lower leakage current than with aluminum gate technology, the only other MOS
technology available. Intel, founded in 1968 by mostly Fairchild employees --
among them was my boss, Les Vadasz ad Andy Grove, both familiar with my work --
adopted the new silicon gate technology.
In April 1969, Busicom, a
Japanese calculator manufacturer, having heard that Intel had the best
manufacturing technology in the industry, visited them with a very demanding
custom job. The Busicom design consisted of a programmable decimal computer
with ROM-based macro-instructions and shift-register data memory, partitioned
into seven different chips. The chips were to be used for a family of products,
starting with a desktop printing calculator, at the low end, and ending with
cash registers and billing machines at the high-end. Dr. Ted Hoff, who at that
time was the manager of the application research department of Intel, was
called in to evaluate the opportunity. Ted was familiar with computers, having
worked with IBM 1620 and 1130 at Stanford University. Looking at the Busicom
design he was surprised by its complexity. He was also concerned with the
magnitude of the task: Intel didn't have enough engineers to do all of that
work. With the encouragement of Bob Noyce, Ted started considering alternative,
simpler designs that might use the three-transistor RAM cells instead of the
six-transistors shift register cells. He was naturally led to consider the design
of a general-purpose computer that might be programmed to perform calculator
functions.
This
initial work occurred in July and August of 1969. In September Stan Mazor,
coming from Fairchild, joined Hoff's group. With one more person on the team the
work started accelerating. Working with the Busicom engineers, sometimes in a
mildly adversarial climate, the architecture of the CPU and three other
companion chips took shape to the point that Busicom people started feeling
comfortable with it. In October 1969 the Busicom management chose the Intel
architecture over theirs. Masatoshi Shima, the principal designer of Busicom,
remained at Intel until December 1969 to program some of the critical routines
with the newly minted instruction set, to verify that the job could be indeed
done. Intel was to start immediately the logic and circuit design of the four
chips, slated for sampling seven months later, according to a very optimistic
development schedule prepared by Les Vadasz.
In April
1970 I joined Intel as the project leader of the Busicom chip set. Les Vadasz,
my new boss, had been rather secretive with me in the job interview. I only
knew that I had to design a set of four chips with random logic. The first day
of work, I was introduced to Stan Mazor who gave me the general specification
of the chip set - essentially a few block diagrams and the instruction set. He
also told me that the next day Shima would arrive from Japan. We both went to
greet Shima at the San Francisco airport and as soon as Shima settled in my
office he asked me to show him the progress made. He was expecting to check the
logic and circuit design of the chips, just as scheduled, but no work had been
done since his departure from Intel in December. I gave him the material I had just
received from Stan, and he immediately said that he already had that. "This
is just idea" he said, "where is design"! Shima was furious
because his product schedule was irreparably compromised.
It took me
the best of one week to calm him down and make him understand that I had just
joined Intel and that I could not possibly have done the design in one day. I
finally succeeded in convincing him that we were both in the same boat, that if
he stopped bitching and started helping, we might actually get the job done
sooner. From that point on I was on my own. I was the leader of the project,
but I had no staff, Hoff and Mazor were no longer involved with the project,
and my boss Vadasz, the manager of the MOS design department, was 100%
preoccupied with the 1kbit DRAM - the 1103, the future of Intel.
A few days
after Shima's arrival, I prepared a new project schedule, more realistic than
the one he had been given. I promised that it would take nine months to get to
samples, knowing that I was still very optimistic. So now I was saying that the
project would take nine months instead of seven, and we were already five
months late. There were a number of agitated calls in Japanese across the
Pacific. The project actually got close to be cancelled, but eventually Busicom
management decided to continue with it and asked Shima to stay on to help me
for six months. This was definitely not a good start.
My first
task was to develop the design methodology for random logic circuits using
silicon gate technology. This work had never been done before. Intel had used
silicon gate technology only for memory chips and had no experience in random
logic. There were two critical innovations of mine that made the CPU possible:
the use of buried contacts and the use of bootstrap loads. The first innovation
allowed to make direct contact between poly-silicon and junctions without
having to use metal. Used properly, it provided a second layer of
interconnections, reaching the gate density necessary for a cost-effective
chip. The other innovation was the bootstrap load -- a circuit trick widely
used with metal gate MOS circuits - but considered unfeasible with silicon gate
technology without using an additional masking layer. Bootstrap loads allowed
the use of dynamic nodes without sacrificing power and speed or having to use a
lot more transistors for the job. There was no other viable technique with a
good speed-power product, except for four-phase logic, which had its own major
drawbacks. I figured out how to make bootstrap loads without an additional
masking step, paving the way for an efficient design. Unquestionably, the speed
and the circuit density enabled by the silicon gate technology were decisive
factors in the birth of the microprocessor in 1971 instead of a few years
later. I also devised a method of random logic design that combined logic and
circuit building blocks -- including some novel circuits -- with layout
information, to make the design faster and less error prone. I then proceeded
with the logic and circuit design of the 4001 first. The 4001 was a 2 kbit
metal-mask programmable ROM with a 4-bit mask-programmable input/output port. Next
came the 4003 design, the simplest of the chips, a 10-bit serial-in,
parallel-out and serial-out static shift register. Then the 4002, a 320-bit
dynamic RAM with a 4-bit output port and finally the 4004, the 4-bit CPU. During
this time Shima, under my direction, gave me great help in a variety of
technical chores. The rest of the team consisted of three layout draftsmen and
one electronics technician.
The
project was like fast-tracking four buildings, with design and construction
going on simultaneously. All was done by hand: the circuit schematics used
large blue-line sheets, the composite layout was done with colored pencils over
mylar at 500X. Then came the rubylith cutting, done by hand for each layer with
the help of a precision cutting table. We also had to design and build the
tester and the test programs for both chip characterization and production
testing. I worked 70 to 80-hour weeks until the project was completed. I was
lucky that Marzia - our first baby, just born in March 1970 - was channeling
all of my wife Elvia's energy. She also spent the three summer months with her
family in Italy, leaving me guiltless to work around the clock.
The 4001
was the first chip to come out of manufacturing, in October 1970. I was nervous
because I had never designed a chip of that complexity before. Fortunately the
chip worked perfectly giving me confidence that the rest of the project was
sound. Shima, a bit more relaxed by the good news, went back to Japan at that
time. In November, the 4003 and then the 4002 were also manufactured and they
both worked. And finally the big day came. It was the end of December 1970. If
the 4004 worked I would have met the impossible nine-months schedule. Fortunately,
by the time I got the 4004 wafers, everybody had left the laboratory so no one
could witness my trembling hands. I loaded the first 2'' wafer into the prober
and lowered the probes onto the first die -- an oscilloscope was also connected
to some of the critical signals. I pushed the start-test button and… nothing. I
mean nothing, not even a wiggle anywhere in the scope. I told myself, it's a
bad die, let me try another one: same story, there was nothing moving. I
changed wafer and I repeated the process. Same result. By now I was profusely
sweating, thinking: How could I have screwed up so bad? I continued to test
more dies and more wafers. No life. After 20 minutes of agony I finally decided
to take a look under the microscope; something must be wrong! It didn't take
long to find out that the buried-layer masking step was left out of the
process. No wonder there was no life! Most of the gates were floating! I made
peace with myself, simultaneously with getting mad at the manufacturing guys
for this screwup that cost me a three-weeks delay. Now the project was
definitely going to be late.
I finally
received new 4004's in January 1971. I spent most of the night alone, testing
it. Everything that I tested worked perfectly and by 4:00 am I went home in a
state of elation, sharing my excitement with Elvia, who was waiting for me. That
was the night the 4004 was born. In the following days I found a few minor
problems that I soon corrected, but the calculator project could now continue
with confidence at Busicom. I sent Busicom fully working samples in late March
- the calculator kit included one 4004, two 4002's, three 4003's and four
4001's containing the calculator firmware -- and by April, word came that the
calculator worked perfectly. We could now start production.
Soon after joining Intel I
found out that the 4000 family was an exclusive design for Busicom. I was
disappointed because I wanted to design a standard product. I also found out
that there was another single-chip CPU design going on, the 1201. The 1201 was
casting in silicon the microcomputer architecture designed by Datapoint for
their intelligent terminal. It was another exclusive custom design and Hal
Feeney had responsibility for the project. I actually thought that the 1201
would become the world's first microprocessor, since the project had already
started and Hal had only one chip to design while I had four -- the 4004 being
the last to be designed. I resigned to that fact and soon I was too busy to
worry about it. But then the 1201 project languished and Hal was asked to
design a new memory chip. I inherited the mothballed project in January 1971,
with Hal completing the project, working under my direction.
By that
time I also started urging management to market the microprocessors broadly. I
was getting the message by everybody, including Hoff and Mazor, that the 1201
may be OK but that the 4004 was good mainly for calculators, and that it would
be difficult to get Busicom to relinquish its exclusivity. But I wanted to see
the 4004 in the market! I argued that the 4004 could be used effectively in
many control applications and to prove the point I used it as the controller of
a production tester I was designing. I used the success of that project to
convincingly show management that anybody else could use it as well. Some time
later I found out from Shima that Busicom was in financial difficulty. I then
suggested to Bob Noyce to renegotiate the contract trading the exclusivity
clause for a price concession. The strategy worked and by November 1971 the
4000 family was announced to the world with a new name: MCS-4 - microcomputer
system 4-bit. By April 1972, the 1201 with the market name 8008 was also
announced. It was the world's first 8-bit microprocessor, the core member of
the MCS-8 family.
Hoff,
Feeney, Mazor and I were all involved helping the early marketing efforts of
the newly formed microcomputer marketing department under the direction of Hank
Smith. I vividly remember a trip to Europe to visit customers with Hank. This
trip occurred in the late summer of 1971, a few months before the MCS-4
introduction. I found out that the more computer literate the customer, the
more resistant he was to consider using the microprocessor. The worst meeting
was at Nixdorf Computer where they nearly ridiculed us for the poor
architecture of our machines. Some of their criticism was valid, to be sure,
but the level of hostility was only justified by the more or less conscious awareness
that a turf war was beginning with the semiconductor guys. Well, we know who is
left standing 30 years later!
The first
applications were primarily driven by non-computer customers that had a problem
to solve - not a philosophical debate. For many low-volume applications, the
task of designing the hardware using conventional components was too daunting,
costly and inflexible. With the microprocessor, the solution was a much more
flexible software program. Gas pump controllers, traffic controllers, automatic
blood analyzers, in addition to all forms of calculating machines were the
first applications of the MCS-4. Intelligent terminals and programmable
microcomputers were the first applications of the 8008. It turned out that the
MCS-4 became the archetype of the microcontroller business, while the 8008
became the archetype of the microprocessor business.
During my
1971 trip to Europe, listening to customer inputs, I realized that we could do
better. In the next few months I developed the basic architecture of what was
to become the 8080. I described it in a memo to Vadasz in April 1972, urging
him to let me start the project immediately. But Intel was in no hurry, they
wanted to see how the newly introduced products were doing in the market before
committing more resources. The project only started in November 1972 and our
market lead shrank to six months to the Motorola's 6800. Nonetheless, the 8080
was highly successful.
Intel in
those days was a memory company. Microprocessors were important only insofar as
they helped sell memory chips. I had always felt a second-class citizen at
Intel. I believed in microprocessors so I decided to start my own company,
completely dedicated to the new business. I made that decision in the summer of
1974, a few months after the market introduction of the 8080. I finally left
Intel in October 1974 and with Ralph Ungermann, one of the managers reporting
to me, I started Zilog.
I
conceived the Z80 microprocessor family in December 1974. It was a Saturday. I had
been struggling for some time to figure out our first product. Then it came in
a flash: I said aloud, Super-80! And with that exclamation, the basic
architecture of the Z80 family came vividly into focus. I don't have enough
time left to tell that story, except that the Z80 is one of the most successful
microprocessors ever produced: it is still in high-volume production today, 25
years after its market debut.
With Zilog a new chapter of
my life started: I changed from an R&D guy into an entrepreneur and CEO,
activity that has continued until a couple of years ago. Over the last 25 years
I started and was CEO of three start-up companies. Two years ago I decided that
it was time to slow down a bit. I am now the Chairman of the Board of my last
start-up, Synaptics. We produce human-to-computer interface products using
neural networks and mixed-signal technology. I am also Chairman of the Board of
IDT; board member of five other companies and an angel investor. For the first
time in my life I have plenty of time to travel for pleasure, read about a
variety of stimulating subjects and generally enjoy life. However, I am not
ready to retire yet: if the right project were to come along, I know my
creative juices would be flowing again.