Interview with Stan Mazor
June 9, 2000
Los Altos, California

RW: Stan Mazor was the third designer on the Intel 4004, considered the first commercially available microprocessor. Ted Hoff designed the architecture, Federico Faggin the logic and physical design, while Stan wrote the application programs and defined some instructions After contributing to the 8080 and 8086 designs, he went on to become a computer-aided design specialist at Silicon Compilers, Synopsis, and others. Well, Stan, tell me about your early days.

SM: Well, I was raised in Oakland California, went to public schools, Oakland High School; met my wife when we were fifteen, and went from Oakland schools to San Francisco State across the bay. In high school, I was active in the club called Junior Statesmen and they have a position called the governor which sort of runs that organization from a youthful perspective and I was fortunate enough to be the governor in '58 and had a chance to meet some of the legislators and became a legislative advocate--or lobbyist for the young people of California when I was sixteen.

RW: Didn't Bill Clinton do something like that?

SM: I'm not sure. I don't know. There's also an organization called Boy's State, which is a similar organization, so perhaps it was Boy's State. I have a picture from those days. I'm don't know if you can catch that at all, but that's a photograph of me standing next to Cranston and, of course, Harry Truman and, at that time, the Mayor of Chinatown. This was when I was sixteen and going out for photo opportunities.

RW: So then, did you go off to college?

SM: Yes. I went to San Francisco State, across the Bay, a commuter school, and we had a very small computer center, a small computer called the IBM 1620 and after a year or two I learned how to program that machine and worked as a student assistant, lab assistant, in the computer room. And I have another picture, which is of me holding what it is now difficult to understand what it is, but it is a deck of cards, Holerith cards, in front of a card reader of that machine. That was 1963. So I started working on an IBM personal computer in '63.

RW: So what was your first job like?

SM: Well I went from there to the Survey Research Center at Cal Berkeley as a programmer and that was in around 1960 - late '63, '64 and I worked programming the 1620 on campus at Berkeley. And not too long after that, I got recruited down to Fairchild Semiconductor in Silicon Valley and joined them in 1964 as a programmer.

RW: Now was this Fairchild R&D?

SM: This was Fairchild Camera and Instrument, Fairchild Semiconductor down in Mountain View and I was a business programmer and scientific programmer. And then in 1966 I went up to join Rex Rice up at Fairchild R&D on a computer design project.
RW: Yes, Rex did a lot of the very early work on - the computer itself was hardly larger - that much larger than today's minicomputer.

SM: A little bit larger.

RW: Yes, two or three times.

SM: So in 1968, I was working at Fairchild R&D as a computer designer or logic designer and we worked on a computer called the Symbol computer and I have a picture of that team and a picture of the computer, which was a little bit larger than today's personal computers, but not that much bigger than a regular computer.

RW: Well, it was much smaller than mainframes of the day.

SM: Yes, it was smaller than the mainframes of the day. And that computer is quite interesting. It was to be a high level language computer that has had the operating system and the compiler directly in the hardware and unlike what we had studied in school, which was to minimize logic, our charter was to use as much logic as possible to maximize logic so we put absolutely everything in the hardware and its hard to believe it today, but even the text editor was done in the hardware. So it was a completely hardware system and it was a misguided approach really in the way the industry evolved which was read only memory and more programmable systems. But I was working on the decimal sorting point unit, the calculator that subsequently worked on - shared the same characteristics of the Symbol computer that we did decimal floating points. So I became pretty good at doing decimal floating point arithmetic. And there were people at Fairchild that left the group saying we were working on the wrong thing, we should work on something smaller and so I left in '69 to join Intel and I ended up working on small computers instead of large computers.

RW: And that was the 4004?

SM: That's right. I joined Ted in September of '69, Ted Hoff, at Intel and we were working at that time on the 4004 project for Visacom. And I brought a couple of artifacts with me on that this morning. This is an early user manual of the MCS4 as we called it, the Microcomputer 4, and I worked on all aspects of that including writing software and working on development systems for it.

RW: Now that wasn't what we would call a desktop computer today. It was an embedded controller, isn't that right?

SM: That's right. The problem we had was to develop computer chips for a calculator that looked like an ordinary desktop calculator and our approach was to create a general purpose machine and to program it to act like a calculator. And so that was our solution. But we had to simultaneously do the architecture for the computer, the packaging, the logic design, and then write the programs to make it act like a calculator. And of course since it was a new design, we were not being compatible at emulating some other machines so it was original work. But we had stolen a little bit from the PDP8 and a little bit from the IBM 1620 so the computer looked a lot like some machines that we had used before. I had programmed about ten machines at that time and had been a computer architect for six years. So we had quite a bit of computer background that went into that machine.

RW: Well, but at Fairchild there was work going on there on a calculator, chip set too, that was somewhat similar, was it not?

SM: Yes. If you look at the history in 1969, there are several interesting things to comment on. First of all, the magazine article featured a computer on a chip, which was actually an arithmatic unit. So the idea of computers on a chip had been mentioned in "Fortune" Magazine or, I can't remember the cover exactly, but it had been, the idea had been around. And there were a number of people who were working on multiple chip CPUs, usually four bit slice arithmatic units, or, in the case of Fairchild, a bit serial machine. But none of them actually put together what we did, which was a complete CPU on a single little chip and in this case, big enough to handle a four bit data word. So I think we really had the essence of it, which was it had to be packaged very inexpensively and have a low manufacturing cost to be embedded in something like a desktop calculator. And on the other hand, it was more general purpose than these bit serial machines that were really focused just on decimal arithmetic.

RW: Well, Ted Hoff, Federico Faggin and yourself. Who did what?

SM: Well, Ted had the idea of a processor instead of doing custom chips or specialized chips for the calculator before I joined in September. And then he and I refined the architecture and I wrote a lot of test programs and sample code. I put in maybe one or two extra instructions to the instruction set, but the idea was really Ted's idea before I had gotten to Intel. Then Federico joined after we had the architecture just about done, and Federico did most of the logic design and all the circuit design and all of the layout. So Faggin really gets the major credit for making the thing happen by actually doing the engineering design and Ted really gets the credit for doing the original concept and architecture. And I was sort of the middleman helping out around the project as I could and making contributions as I could.

RW: Well, initially, all the credit went to Ted. Why was that?

SM: Well, Ted was my boss and it was appropriate that the manager was the person who drove the project, but the patent that was issued has the name - myself, Ted, and Federico. So I felt that we were all sharing in it as appropriate and certainly in Federico's case, when he left to found Zylog he got temporarily written outside of the Intel history. So we were not there promoting the fact that he was a great contributor when he was competing with a very good processor, the Z80 from Zylog. But after that blew away, then I think he certainly got his share of the credit as well. I don't think there are any hard feelings.
RW: Well, that's good. In recent times, though, you have gotten the recognition, like the Kyoto prize.

SM: Yes, after twenty-five years we've gotten some attention and awards and it's nice to have those things happen before you die. Let me tell you a little bit about those days though, very interesting things we did. We built a video game out of the 4004, playing a spaceship game on a TV set and we demonstrated it as an application example as you typically do within an engineering group. And I remember several management people suggesting that who would ever want to play a game on a TV? And another thing we worked on was a scientific calculator, the idea of having a calculator that could do scientific calculations. And that was also rejected because the slide rule was quite popular and effective. And of course subsequent to that, both video games and scientific calculators were quite popular. So there were a number of - when you're in an applications engineering role, you often have a mania or ideas to do things, and you'd like to see them become products and oftentimes they do become products but it's somebody else's company. Those were past times. But in 1969, Victor Pora of Computer Terminals came looking for some memory chips and he was not aware of the work that we were doing on the 4004 and I had a chance to propose to them a single chip CPU, which became the 8008. And that has an interesting history in that I had proposed the chip although it was almost a hundred percent their architecture because they had come with a particular computer in mind. So in the case of the 4004 that we did, it was all our own, and in the case of the 8008, it was really by and large the Data Point architecture. We did take out a few things and add a couple of things to make it a little bit more practical.

RW: Well that original 8008 was really the beginning of the Intel line of desktop machines.

SM: That's right. If you look at the two Intel product lines, they came out in 1971 and 1972. And the 4004 was really an embedded controller for use with Read Only Memories and the 8008 was used as a general purpose computer, 8-bit data processing, usually running out of RAM and was a precursor to business machines. I later worked on the 8080 and I share a patent on that chip and I did most of the architecture of the 8080, the instruction set in particular, and that was very much compatible with the 8008. And then the 8080 was what really put Intel on the map in 8-bit computers. And out of that came the 8086 which was compatible with the 8080 and then, of course, the lineup to the Pentium as we know it today. And there are many of the originals traces back to the features of the 8008. Also an interesting sideline is that DataPoint, the guys that we did the custom chip for, never ever did use the 8008 and probably were ultimately put out of business by people like IBM with the 8080. And so it's quite an ironic situation that they had proposed an architecture which probably ultimately buried their company and they never took advantage of the LSI that got generated for them.

RW: Well, on this compatibility issue, as you know, the 8086 would run 8080 code. It would run 8-bit code and it had some special instructions in it to allow that to happen, that compatibility to happen. And then in the famous 16-bit battle, Operation Crush, where Intel, Motorola and Zylog were all having it out, it was decided at Intel, and I was there then, that we would not - we would not stress this compatibility because it made it look like an old architecture, which was the charge made by both Motorola and Zylog. So we actually removed from the datasheet two instructions that allowed compatibility with the 8080. And that became another Intel non-history event as we repositioned it as really a great new machine. But today, the Pentium III has some of those very same instructions in there.

SM: Well, in the early days, you're quite right. We stressed the 8-bit features of the 8086 and we were countering TI in the marketplace, saying that, you know, the world didn't need 16-bit so then as the world was ready for it, then we emphasized, as you say, the 16-bit features. So that's quite correct.

RW: Well, show us some of your goodies you got from these recognitions.

SM: Well, a few things here, we had a ten year celebration at Intel and this was a wafer and this is the size of wafers that we had in those days, which is a lot smaller than the pizza size wafers of today. It says Ten Years History at Intel. And then we had a period of time when Intel's business wasn't too good and so we were working at what we were calling the hundred and twenty-five percent. Everyone was asked to work overtime. So this glass represented, you know, the hundred percent mark and then the hundred and twenty-five percent mark. So it says "I Survived the 125." So this is one of the darker days in Intel when we were all working especially hard because business conditions seemed down a little bit. Then we have some souvenirs from the Intel awards that I have here. This is the Kyoto prize that Faggin and Hoff and myself received from the Kyocera Company and the Kyoto Foundation, Inamori Foundation.

RW: That's like the Nobel Prize, isn't it, in Japan?

SM: Well, in Japan, that's the technical prize and there's really no technical prize in the U.S. that you can compare it to.

RW: No, they don't have one for engineers.

SM: And then we got a very nice award from the Federal government and from the Ron Brown, American Innovator Award, if I can lift this thing up. It's pretty heavy here. And that's a nice piece of glass and this is what Ted Hoff, Faggin and I received also. So after about twenty-five years the microprocessor was recognized and we were give some very nice honors.

RW: Well, it's too bad Ron Brown wasn't the recipient of a good airborne radar unit as the plane slammed into a mountain because the air controllers had it wrong.

SM: Yes.

RW: So, when did you leave Intel?

SM: I left Intel in 1983. If you recall that Intel was primarily a memory company and the memory business was kind of down in the '83 period as the Japanese at that time were making memories bigger, cheaper and so business was kind of down and I went to start at another company which was called Silicon Compilers, a CAD company. I had started in CAD in around '64 and worked on computer-aided design of chips and board level things throughout my career and this was a chance to join a computer-aided design company.

RW: Now who were the other principals in that?

SM: Actually, an Intel guy and Ed Chang was Intel guy and Phil Kaufman is probably well-known from Intel, started that company. He was president. John Doer was on the Board of Directors, the Chairman of the Board at that time. And we were very ambitious and tried to do a lot of good things and we were probably a little too ambitious and so the product didn't work very well and after a while we were bought out by Mentor Corporation, another CAD company.

RW: Yeah, you were a competitor of mine as I was doing simple Gate Arrays that worked and you guys had all this wonderful technology, but the circuits didn't really work and I think Kaufman was one of the big problems there because I don't think he thought it was worthwhile to actually go to silicon, you know, that that was sort of a given.

SM: Well, I think that everyone assumed that an ASIC chip with custom blocks would be faster than a Gate Array but, of course, it turned out that the technology involved in doing the CAD tools put us quite a bit behind the learning curve of the silicon, whereas Gate Arrays were able to move quite rapidly along to the new technology. So though theoretically the ASICs would be faster, it turned out in practice that Gate Arrays were better, so counterintuitive.

RW: Yeah, well, we also had Toshiba technology which helped a lot. Carver Meade was involved with that too, wasn't he?

SM: Yes, Carver had a student named Jake Johansson who had developed this idea in college called Bristle Blocks and the main claim to fame of Silicon Compilers was the data paths. And so we had a very good data path compiler. We didn't do anything else very well. And so that was Carver's early work also.

RW: Well, when you left Intel, did you leave on good relations with people like Andy Grove or were they…

SM: Yes, I had done a data processing project for Andy, I had known him socially and, of course, I had worked for Fairchild R&D for both under Andy and Gordon Moore. So I was in fine, fine straits with them. I was the first application engineer in Intel to go the Europe, so I went to Europe in '75 and '76 and supported customers there. I came back and started up the training organization in '76. So I had done a number of things in applications engineering and then training. But by '83 it was getting to be a pretty big company and, you know, I was looking for something, again, closer to ground floor opportunity and that's why I went off to Silicon Compilers. So I left on fine terms and certainly we were not competing with Intel in any way.

RW: Well, then you went to Synopsis though.

SM: That's right. The Silicon Compilers was a CAD company that had a very broad product line and wasn't too successful and when I went to Synopsis we had one very silly little product which took in a net list and put out an optimized net list. And I was employee twelve there and built up the applications engineering group and customer support group. And that was a very nice company and we had a very successful product line. And I stayed there for five or six years and we had a good business and made a good contribution in the industry.

RW: Yeah. Yeah. And today, it's one of the survivors. So then what did you do?

SM: Well, I had tried several startups that went well and I thought I'd try something completely different so I joined a company building financial models called Katz Computer Aided Training Systems and our idea was to apply computer aided design techniques to financial models and that's what we did. And that company didn't do very well and it got bought out. And then after that, I went to a company called BEA Systems, which is in the data processing arena, and I also worked on that product in training people in application support. And that company turned out to be very successful and that's used in on-line transaction processing And now I'm at Cadaver, which is another computer aided design company and similar to Silicon Compilers where we're working on automatic layout of artwork for Standard Cells.

RW: Well, you've been in computer-aided design of integrated circuits almost longer than anybody who's active today.

SM: Well, I started out in computer-aided design in '64, but at that time we were doing work on printed circuit boards. So a lot of the problems are the same, but is hasn't been integrated circuits, only about half of that time.

RW: Well, still, you're like the senior statesman now right?

SM: Well I've got the gray hair, anyway.

RW: Jim Koford I think, is, who's one of the early pioneers has scaled back his involvement with Monterey.

SM: I just saw Jim this weekend at the Design Automation Conference and, of course, he was with you at Fairchild R&D and I worked with him at Fairchild R&D and, simulator work that he did and, of course, at LSI Logic where you guys had a fabulous software team and I know he made a lot of contributions there and now he's in at least his third tryout at Monterey Design Systems, that's right.
RW: Yeah. Well, so what are the changes that you've seen over this thirty-year period of - I mean what have been the fundamental changes driving this stuff?

SM: Actually, a lot of the problems are the same and some of the solutions get reinvented and part of the problem with being around for a while is it's a little aggravating to see people make some of the same mistakes. But generally, it's a scaling issue, you know, I think the problems are about the same. Either we were trying to route a board and now we're trying to route a chip or now we're trying to route a cell. So there's not that many changes taking place in a way and I think that there's new opportunities around, abound in the Valley for different kinds of companies and, I think, the real question is if Moore's Law will continue, then I think we'll continue with the same kind of growth we've had. We're at the point right now as we cross through .25 micron, as you know, where people said that we would be at the limit of what you could do with light and they didn't think that we could do things in semiconductors that we're doing. And now we're working with shadows and infringing effects and so we're pushing the light a little bit further. As we get down to a tenth micron, which is the geometry feature for metal, let's say, metal spacing and metal width, a tenth of a micron, then it seems like we're going to hit some walls that right now it's not very obvious how we're going to get past them, but maybe we will and find some new technologies.

RW: It's been an incredible ride, I must say. You remember our first custom circuits at Fairchild were thirty-two gates.

SM: Yes.

RW: And at LSI Logic, our initial product in '81, our initial products were maybe a thousand gates.

SM: Yes.

RW: And now people are at fifteen million transistors.

SM: That's right.

RW: It's just incredible. Okay, so you've been in the computer business for a long, long time. What are some of the stories that come out of that?

SM: Well, one that's kind of interesting is this Little Indian, Big Indian thing. It's turned out that when I was at Intel, I defined the 8008 instruction set and the 8008 addressing modes and I did it in a strange way which we ended up calling Little Indian, which means the low address comes first. And I did that because the DataPoint machine was bit serial and I was trying to do something compatible for them, although I knew it was backwards at the time. Well, as it turned out, they never used any so that by making the Little Indian first was kind of a mistake. And the Intel 8080 and 8086 also were then Little Indian. So we have something in the industry where the Intel way has been more or less the wrong way and we did it for the wrong reason. And that's one of the…
RW: And you're responsible.

SM: And I'm the guy that did it and it was my mistake.

RW: They didn't give you an award for that.

SM: No, they don't actually. Then, just a couple of other things you often hear about the products that succeed but you don't often know about some of the products that don't make it and usually for every one or two products that come out, there's one or two that don't make it. And when we were working on the 4004, Ted and I thought it was a little too aggressive and we weren't sure it could be done, so we started with another chip called the 4005 and it was a joint project with MIL, which was an affiliate of Intel in Canada. And so we defined the architecture to be much, much simpler than the 4004 and the idea was the Canadian company MIL would actually design the chip and we'd provide the memories. Well, wouldn't you know it, that in a competitive situation, we were against MIL someplace and the guy, the customer was arguing with me that he thought maybe the 4005 was better than our 4004. So we ended up in a competition mode and it turned out they could never make the 4005 so it went away. But it's almost like you work really hard for your son and your daughter and in the end, in some sense, you have to choose or you have to defend one against the other. Then another story of some interest is that Intel was very aggressive in defining a chip which later became known as the 432 and this was a radically new architecture and multiple processor and very exotic and somewhat object oriented, and Intel spent more than 60 million dollars, I would say, on the development of that chip and it never really did work or meet all of its objectives and so they had to abandon it. So the 8086 turned out to be a really stop-gap measure which was thrown together sort of, not exactly at the last minute, but, you know, to compensate for something that had not gone the way that they had anticipated. And a lesson for me out of that was that if you give people a completely blank sheet of paper and say, you know, you're not constrained, go out and do the right thing, they may not succeed, whereas in the case of the 8086, the people were constrained to do something in a very shorter amount of time and very compatible with the 8080 and the consequence of that is they ended up with a very successful chip. So an irony again in life is that you often look at when you're doing it as the constraints as something that's holding you down and limiting you and we don't realize that those constraints are in fact sometimes precisely the enablers that allow you to succeed.

RW: That's fabulous. Well, so what are your hobbies now?

SM: Well I started in computers in '64 and after about 30 years of computers they're not quite as interesting to me as they once were. And you can always imagine someone getting into computers for the first time and they're very excited about it, but for the last ten years, my hobby has been studying and working on the architecture and design of French chateaus, which sounds kind of strange. And so, I've been designing a French chateau the last two years and trying to take what we call a modular approach, build something in stages. So right now we're building what my son dubs a micro-chateau in Ashland, Oregon. And we went to France, to Normandy, and I don't know if you can catch this one, but this is the model that we're adapting - it's 182 feet long, it was built in 1630. And one of the characteristics of this is it's only sixteen feet deep so it's very long like a Hollywood front, but it's not very deep at all.

RW: Sixteen feet.

SM: Sixteen feet, that's it. I have a design; this is going to be a one-room shack, if you will, or a one-room cabin, which we didn't build. This is the current design we're working on which is the foyer and subsequently we intend to add the left and the right wing. And we might do that someday, but for the moment we're just building the foyer section. This is a little bit bigger picture of the stage two, which is the foyer in the middle and the wing on the side and it's more or less about eighty percent to scale of the chateau in France. We went to Normandy, my wife and I, and we took measurements of it, so we're copying it to some extent. And we're building the chateau out of Styrofoam, there's a little piece of Styrofoam block, and this is out of recycled drinking cups, if you will. And so the idea is to use an ecologically friendly material. And the idea of building chateau out of blocks is like they were built originally out of sandstone or limestone, only we're using Styrofoam. We glue the blocks together.

RW: Well, is this earthquake safe?

SM: Well, we've got steel reinforcement rebar and also concrete, so we think it's going to be quite earthquake friendly and also very stable, but it's a new building material, it's commercially available. We'll have to find out what it's like. That's one of the reasons to do things in modular stages is to also get some familiarity with the material.

RW: Thanks Stan, from the first microprocessor to the first plastic chateau, always the innovator.

SM: Thank you Rob Walker