RW: Jim Koford is currently CEO of Monterey Design, a new integrated circuit design automation company. Jim is a true pioneer in the semiconductor industry having originated the semiconductor industries' first logic simulator while at Fairchild in the '60's and was in charge of LSI Logics design automation for most of the '80's and '90's. I visited Jim and his wife Marcia at their home in the Monterey Highlands.
MK: As the wife of a Silicon Valley executive I think one of the keys to a truly happy marriage is for both individuals to be very strong and for there to be a mutual supporting of one another's spirit and who they are. Also to share things and I think, one of Jim and my's greatest love is music and our concert grand piano here is an example of that. This is a nine-foot Mason and Hamlin concert grand that we named Sebastian after Johann Sebastian Bach and Sebastian's counter-part is over here, Magdalena. Magdalena is a seven-foot Steinway Hamburg B that Jim picked up in Hamburg, Germany when he was on an LSI business trip and had it flown over here. So we've been able to enjoy playing piano duets and piano-organ duets on these.
JK: This, parenthetically, is Marcia's piano, The Hamburg Steinway. Well, this hobby of pianos involves working on them too. And this is a seventy five-year-old Knabe Ampico Player Piano that I rebuilt when I was in Boston a few years ago, but it still plays very well. Let me demonstrate.
MK: Technologically, this piano is at the other end of the spectrum. This is a Yamaha Disklavier. It's played by inserting a disc here and then the music is carried over here.
JK: Okay. Well, here I have the other end of my musical hobbies with perhaps even a little more current stuff. Obviously, we have some computers up here, MAC, PC, we have a slightly older Kurzweil keyboard and a couple of other keyboards here and also we have some digital editing equipment here that we use to edit some of the tapes that we've made here in the house with some of our classical music friends which four or five of them have actually ended up on commercial CD's.
RW: So, Jim, tell us about your background and where you went to school and things like that.
JK: Okay Rob. Welcome to Monterey and our home here outside of the City of Monterey. I went to school at Stanford, right in the heart of Silicon Valley, and after graduating from high school in 1955 in Cheyenne, Wyoming, I arrived on the Stanford campus in the Fall of 1955 as a wide-eyed seventeen year old determined to study a subject I even then knew I loved, engineering. And I stayed at Stanford for approximately nine years, ended up with a Ph.D. from there, after which I went to IBM.
RW: Well now, who else was on the campus at that time?
JK: Well Stanford, both in the faculty and in the student body was really an incredible place because it was emerging. Stanford had always been a fine regional university, but after the war it really emerged as essentially a world-class university and of course the other great university in the bay area, Berkeley, was certainly a strong world-class university too. So Stanford then, in a very exciting way, emerged as a probably by the end of the fifties decade it was the number one or number two engineering school in the country depending on who rated and when. So during that period, of course we had some wonderful faculty members there. The Lindvilles from Bell Labs, we had Jim Gibbons who did some interesting pioneering early work, we had Arthur Shallow was there, we had two or three Nobel prize winners in the Physics Department, we had for awhile, William Shockley, co-inventor of the transistor, taught me device theory at Stanford, great guys in control theory, information theory, Hemming was there for awhile and so it was just a wonderful place for these thirty to forty year olds who were the people who were building the foundation of the digital era, although we didn't know it at the time, that's what I thought was really happening. And then we had a terrific student body too, I mean Steve Brier was in my class, Justice of the Supreme Court, Sandra Day O'Conner, another Justice of the Supreme Court had been there a couple of years earlier, Ted Hoff, whom you've interviewed, was a classmate of mine, Paul Lowe, who came to run all of IBM's semiconductor operations was in that same small research group that we had and I could go on and on and on. Bernie Widrow, my professor, was a leader in adaptive systems and many of these adaptive equalizing algorithms that are even in things like modems. Bernie and his group developed, Gene Franklin in control theory was there. As I say, I could go on and on and on. It was really a marvelous broad spectrum technical place to be and I always felt that one of the greatest things that I got in that period of my life was just a tremendous technical education. Mathematics, Physics, Engineering, and believe it or not, Stanford was a fairly good liberal university so I picked up an interest in literature and music and art, while at Stanford too.
RW: It's incredible that here in the bay area that we do have Cal Berkeley and Stanford that seems to have just run away from the other, Europe, other academic venues.
JK: You know, you're right. Of course, I'm very biased and believe it or not, traditionally, Stanford and Cal have always been rivals, but I can almost say "we" and mean Cal as well as Stanford because in my field, electronic design automation as it is called today, they didn't used to call it that, of course, Cal is probably the premier school in the world and so those two schools are just unbelievable achievers.
RW: Head and shoulders above anybody else.
JK: So, we have a little electronic intrusion. I guess they'll give up after awhile. There, the answering machine stopped that.
RW: So, you got your Ph.D. in Electrical Engineering.
JK: Electrical Engineering. Studied with Dr. Bernard Widrow and my thesis was in pattern recognition and the thing that we did at Stanford that was a fun project and my wife Marcia who's sitting over there actually triggered it, was to build probably one of the first experimental real time speech recognition systems. And we were able to walk up to a microphone, train this system in ten or twenty words and then you could speak those words into the microphone and the computer would recognize your speech and type out what you had said, if you restricted yourself to those twenty or so words. Obviously, it was fairly crude but after all it was thirty five years ago. And today, we're now just beginning to have speech recognition systems that are actually on line and you will from time to time encounter them on the telephone system and elsewhere.
RW: It's funny though. I would have thought it would have moved forward because its been around a long time now and yet, we still have keyboards.
JK: Yes, well, it's really a very difficult problem I think because our speech recognition quality was reasonably good. But of course what, the computers of that day and perhaps the ones even of today can't do is do this marvelous deep contextual linguistic processing that the human mind does and so, you know, its taken a log more than just the raw sound recognition. It really is a very deep problem to build a computer that can sort of understand cursive speech. But I think that it is getting better and you're gonna' see more and more and more of it. There are, you know, ViaVoice and Dragon Systems have voice type writers now and if you learn how to use them, they're useful. I actually haven't used them much myself, but I have, if you will, encountered them. And so, it's like anything else in technology. It's happening. You can tell it's happening. It's just not happening, as you said, as fast as maybe we would have thought. I haven't worked in that field in thirty-five years though, so I don't know a lot about what's going on today.
RW: So, after Stanford, did you go to work at IBM?
JK: Yes, I did. And I was fortunate because one of my classmates had been an IBM sponsored student at Stanford and we knew each other quite well and he want back to IBM so I went to work in his department. Dr. Paul Lo is his name. And something else happened at Stanford that I was very fortunate. Because Stanford was, you know, a university to which many of these outstanding researchers came, one of the things that happened in the mid-50's, when I was there, mid to late-50's, as you well know was that William Shockley brought a number of very bright scientists and engineers from Bell Telephone Laboratories, which of course had been the site of the invention of the transistor and was probably the leading center of understanding of solid state theory and practice. Well anyway, Bill Shockley brought this group out to Stanford and he established Shockley Transistors in the industrial park next to Stanford. It's well known that Dean Frederic Terman, the Dean of Engineering at Stanford in the mid-40's felt that the university could have a liaison with industry and he developed a program whereby industry and the university could work together to create new companies and new products, etc., etc. And as part of that, you know, or at least as an outgrowth of that an industrial park was established on the parcels of land next to Stanford. Stanford had a lot of land because it had been Leland Stanford's farm. Well anyway, so here is the university and right next door literally are companies like Hewlett Packard and later Fairchild and the famous Xerox Park Lab was there and the Shockley Transistor was there. And so the cross-fertilization of both the university and those companies was very serious and for that reason we began, at the university, to have lectures from industrial people, Barney Oliver and Bob Noyce and people who were in companies would come over to the university and talk about what they were doing in their companies. And we saw that, the students. So I began, probably by the beginning of the 60's, to begin to notice what was happening in industry in integrated circuits because as you know the planer process had been developed and we had begun to build small integrated circuits with Bob's few gates in a package and I began to notice how that was done, you know, I wasn't really studying integrated circuit manufacturing at the time but I just began to notice how it was done, with the masks and the fabrication technique. And I realized somewhere, probably around, oh, '64, somewhere in that era, that everything you needed to make an integrated circuit, theoretically, could come out of a computer and that once the computer produced the tooling, if you will, the masks and the test tapes, then the factory could be a more or less generic operation that would make this thing in accord with the wishes of the designer who was the person using the computer. So I thought that made a very interesting opportunity to create an environment whereby you could actually design things with computers and the factories then would be able to make them and the factory people wouldn't necessarily have to understand what they were making. They would have to understand the process and how to execute the processing steps properly but think of it as sort of a processing lab for a photography operation. The people who do the film processing have to be very good at that, but they don't necessarily need to understand what the photographs are. And I realized that somehow this is how the semiconductor industry was going to work too. So I became fascinated with the notion of putting together systems that would allow you to design and make semiconductor chip tooling using computers. And that must have happened somewhere around '63 or '64 because by the time I went to IBM I had a very good idea of what I wanted to do.
RW: But did you do that at IBM?
JK: Well, actually we did. First, IBM had an established design automation department. It was in Poughkeepsie and the big idea in computers were, in fact, designed in some ways with computers, but it was a little bit different. It was more record keeping and data processing. So those computer systems would put out the schematic manual so a service man could service the machine. They would write programs for the wire-wrap machines, these automated things that would interconnect these big circuit boards they had in that day. And then finally toward the middle of, oh, the beginning of the '60's decade and certainly by the middle of it when I went there, they were actually using computer programs to make the printed circuit boards that the discrete components would be plugged in and these printed circuit boards looked very much the same as a printed circuit board does today in a personal computer. It's just that the components were much simpler and so you had to have many, many more of them so you had to have huge numbers of printed circuit boards all plugged in this big rack that, you know, went into a box called a main frame. And the thing that was really interesting was they were beginning to use these computer techniques to design part of that but still it was, in my opinion, largely a manual process. So, I very fortunately was able to perform an experiment, it really was an experiment because I was part of the Fishgell Development Laboratory, where we took a problem and actually created a system where the entire design process was done with the computer and I shutter to think of what we attempted in today's content because what we did was very simple, but that was really the key to it. IBM made little circuit modules, oh, maybe about that big out of ceramic. Maybe more like that big. They're little square circuit modules and on this little circuit module were the positive resistors, you could solder capacitors on there and you could what we call now bump transistors. So essentially it was a little circuit module where you interconnected these components and then that little thing would plug into a printed circuit board and you would build the computers that way. They called these SLT modules. So the problem that we chose was to design not the transistors but those, because they were small and there were many, many different types so it was a real problem to design all these things. So fortunately I found this old radar display that was just sitting around IBM that had been built for an air defense system called SAGE, which was deployed by the US Government in the '60's during the Cold War. Well, I had no intention of using it for radar, but this box, great big huge box had a nineteen inch round picture tube on it, CRT, and it would show vectors and numbers and letters and that was all I needed to begin this project. So, I and a couple of my collegues designed an interface box to attach it to this computer, we got a light pen from another vendor because we needed a pointing device. The mouse had actually been invented but it was a little too crude to use for this problem. We designed this light pen interface and then I wrote what you would now call a tool kit, software, so that we could write programs in the computer that could show pictures on this thing and interact with it. By today's standards, I must warn you, it was extremely crude. But it was in 1965. So what you would do, is you would enter a schematic into this graphical machine and, of course, that schematic would go into the computer, it would check certain things to make sure, you know, there were certain things that weren't shorted, it was more or less correct. Then, through the use of a series of manipulations which, again, were on the computer, you would translate that schematic into the actual patterns that were used to make these little ceramic modules. Then I rigged up a Calcomp plotter, which was a popular mechanical plotter of the day, with a rotating knife so we could cut this transparency material, Rob, which I'm sure you know very well, called Rubilift, so that we could actually make these photographic mask-like patterns which could then go into the ceramic factory and be used actually to print the circuits that were to be fabricated with this technique or with this design. So, we were able to show a complete flow from concept, data capture, and masking or tooling creation using a computer and at the time I thought, you know, it was an interesting experiment, I didn't know exactly where it was going to go, but it was what I wanted to show. So we gave a paper on that in the Fall Joint Computer Conference, out in San Francisco as it turns out, and I guess it was viewed as a somewhat seminal project because the IEEE later wrote a history of graphical displays which are so common now and they did allude to that one as the first one. The project itself, of course, didn't continue much because it was a lab project, but many subsequent projects were started at IBM and other places that, you know, ultimately led to great sophistication. So, I thought I was fortunate to be in that place at that time and do that because we had demonstrated something, we'd demonstrated that the computer as a design vehicle could virtually do, you know, most of the tooling preparation.
RW: Yeah, I was in the audience and I remember thinking we need something like that at Fairchild, little knowing that you already had been recruited to go to Fairchild. So at the time you gave the paper I just thought oh, that's fantastic. And indeed, that conceptually is really the same as we do today.
JK: Yeah. Rob, it's funny because it really is, I mean, I have to be careful because you know, I'm in the middle of one of these things now. It's a tiny fraction of the complexity of what we do today, but conceptually it was all there. We were able to start, there are two problems that you've got to solve when you do one of these things. The first problem, believe it or not, is a very human problem. Designs and system concepts do not emerge in computers. They emerge in human's minds. So you've got to solve the design capture problem. How do you, you know, communicate to a computer what it is that you need, that you want it to do and you've got to use not the language of the computer, but you've got to use the language of humans to do that. And pictures are probably the best and most obvious place to start. That's why we had to have graphics and that's why this graphical display, interactive graphical display was so important. Then you need mechanisms in the computer to translate that abstract description which is now in the computer's memory to actual masks or whatever it is that the physical world needs. You also need to verify the design because the design in the computer better be correct. But how do you make sure it's correct because it's in the computer and the computer doesn't totally know that it's correct. It can make certain checks on it, but the human has to be involved. So those three things: design capture, design verification and design translation to the physical world are still very much alive and well today. They're very much what we have to do and they're a huge problem and getting worse today. That's what is engaging the field that I'm in still very much.
RW: But wasn't that the first graphical work station?
JK: Probably it was. I have to be
a little careful because there certainly was some very large computers that
had graphics. There was the famous SketchPad Project at MIT on the Whirlwind
Computer, which was quite well known done by Ivan Sutherland and it probably
triggered a lot of the graphics work. And then there was also some very large
computers at IBM. General Motors had worked with IBM to develop a graphical
display but I think if you'll put the workstation word in there, then I think
it's probably - because this was the first time that I knew that something practical
had been done on a small computer that wasn't a giant main-frame, but you know,
it was sort of a small affordable, individual computer, that is, one person
would tend to use it. And so I think that it had the aspects of a workstation
although it was bigger than this piano by a long shot, but it wasn't as big
as this room, which was what a lot of the other computers were.
RW: How were you recruiting then to Fairchild?
JK: Well, that was partially my own doing. My mom had begun to have some serious problems with her vision. She had an athrosclorotic condition of her lenses, which was beginning to darken her eyes. As it turned out, she lived with that for many, many, many years after I came out here, but it was beginning to be somewhat of a problem and I kind of wanted to get back out to California because not only was I, you know, had lived in California for so many years, I also, I just began to sense that the action was out here, you know, there was just really exciting stuff happening in California. And Rex Rice, who had been an ex-IBMer, and had gone to Fairchild and knew me because he tried to recruit me, believe it or not, to IBM. He was at Fairchild and I had actually gone to IBM with another colleague, Dr. Hugh Mays, who was a colleague of mine at Stanford. In the context of this interview I don't probably mention Hugh as much because he changed fields, you know, in the 70's and really has not been in this field, he's in psychology. He has a very, very fine mind and a very, very smart and important part of the early story here. Anyway, Hugh also I think, wanted to get to California so the two of us actually came to Fairchild. And Hugh came a little before I did and we both had been at IBM and we both had been at Stanford, so he sort of recruited me a little bit too. So it was a combination of those factors.
RW: That's when we met.
JK: That's when we met, Rob.
RW: And started to do these crazy things.
JK: I don't know why, but from the moment that I met you anyway, you maybe were coming from a little different viewpoint, but you had the business concept and the industrial concept that what this could become. I'm sure you are the author or the coiner of the term design factory that's sort of been borrowed by other people now, but you were talking about that in the late '60's and early '70's and I honestly think that, you know, it was a combination kind of a technical vision that several of us had and then you had the business vision and the only problem is it took us maybe a little longer to get that all going than we thought at the time, but that's sort of what happened.
RW: So, you were recruited to Fairchild.
RW: And what was the task that...
JK: Well, the task was basically to continue and to extend what we had started at IBM. The graphic display project there had gotten some notoriety and there definitely was interest on the part of Gordon Moore and Bob Seeds who worked for him as you know, yourself, Rex Rice, in continuing to work on that sort of a system. We hadn't actually developed a system at IBM to layout integrated circuits; it was only to layout these circuit modules or packages as you might call them. And we wanted to really get serious about developing a system that would do integrated circuits. Clearly, we needed a graphical display that was something other than a used radar set, so we set out to design from the ground up a graphical display that would allow us to really have the complex images that would be required for even the integrated circuits of the day. So that was the first project that we really set out to do. And in fact we did that. We worked with another vendor and developed this rather complex display that we programmed to actually be used in the layout of integrated circuits. I wasn't as involved in that project, actually, as I had been in the previous project because something else happened, but I was still a member of the group and worked alongside the people who actually were doing the programming of that, there was a gentlemen named Steve Zucker who actually did most of the programming and, in fact, that work was reported in The Scientific American. There was a Scientific American paper on that project a few years later. The thing that sort of diverted me was I was still very much interested in this problem of how do you communicate a complex design to the computer. And not long after I got to Fairchild, I think Hugh Mays walked in one day and laid a paper on my desk that was written by a gentlemen named Ulrich, who was at North American Aviation on logic simulation. And I had thought years before that it might be interesting to simulate the operation of a circuit on the computer because, again, if fit with my vision of having, of interacting with the computer to put a correct description of the circuit in there. But computer simulation was a field that I was only peripherally involved with and it was really being applied in very different areas in, oh, defense work and some other areas that were quite different from what we needed. Anyway, this fellow at North American Aviation had addressed the explicit problem of logic simulation. That is the simulation of the logical operation of a computer-like circuit and I gobbled this paper up in an hour, I mean I was so interested in it. And suddenly I realized that in this simulation technique lay the key because you could have an engineer enter a circuit into one of these programs and then he would work with the computer and actually simulate the operation of that circuit. Now remember, he's not working on the actual circuit, he's working on a computer model of it. So once he's satisfied with the way this simulation behaves, two things have been achieved. The computer has now captured a description of the circuit and, at least as far as the engineer knows, it is correct because he's checked it out on the computer. So I charged off and started writing, which I think was probably Silicon Valley's first logic simulator, which you named, Rob, you know it very well, it's called Fairsim. And we worked together on that, I guess, for five or six years and we created the, what you would now call a front-end EDA environment. So you could enter a circuit, you could simulate it, and I built this language called simulation control language where you could actually put pulses on it and clocks and checkout the behavior on it and then you would be sure it was correct and you had a good description in the computer. The other thing that we did, at least I think we tried to do and this was probably some of your influence, is rather than create some exotic new circuit family or exotic new engineering technique, we tried to make this simulator as familiar as possible to people who were already designing circuits. As you know, the prevalent method for designing these circuits would be to breadboard them. You'd go build a wire-wrapped breadboard and you'd plug in discrete components that behaved in the same way that the components you wanted to put on your chip would behave and then you'd wire this thing up and then you'd operate it physically and I was going to replace all that with simulation. But what I didn't want to replace was the general methodology, the general thought process the engineer used. So we went to the data books and your group really got the standard 7400 Series Logic Functions and then we put these elements into the computer in simulation so that an engineer who had designed a system with a certain particular set of logic components that he was familiar with would find the computer equivalent in Fairsim. And I think that was a pretty key thing too because we then began to, if you will, train a body of engineers to use this methodology, but it wasn't too bad because they were already familiar with the basic elements that they were working with, it's just that they were now dealing with computer models instead of the real physical thing. And so we put that together and that sort of constituted, there were lots of other things that went on, but that kind of constituted what we called "the front end of our system". In the meantime, my colleague from Stanford, who was a year behind Hugh and me, joined Fairchild, Ed Jones, and he initially started working on test generation. Ed's an extremely bright fellow and a methodology whereby we could figure out how to test these circuits when we built them. And that was very important because as everybody knows, not every circuit that comes out of a fab is good, so you have to have an efficient way of testing each and every circuit to know which ones to ship. Well, at the same time, Fairchild was developing a series of computer control testers so that they had computers in them and they could be fed with data, which would then automate the testing of these integrated circuits. So I had another level of automation now that we could use and Ed did the original software work to feed these testers with the right pattern so that when the circuit was fabricated we could make sure that the actual circuit worked. The final thing that Ed did too was the other part of it, the key part as I said earlier, is you now have to take this abstract computer model and translate it into the actual masks, the actual tooling that goes into the factory and that's called a physical design system or a place and route system and of course we were in the infancy of that. And Ed and several of his colleagues also worked on creating that and so by, I don't know Rob, you can refresh my memory, somewhere by the early 70's we had now a complete ASIC system and the other thing that your group did of course is they developed the underlying circuit structures and the chip architectures and as I remember, we had two product lines, MicroMatrix and MicroMosaic. So we now, literally, had at Fairchild, a complete system which went all the way into a factory, really, and could allow engineers to specify circuits on a computer and end up with the actual circuits coming out of the factory being tested. And I suspect that was the first operation of that type.
RW: It was. I think it was a seminal work.
RW: And it was interesting that our reputations were severely damaged by having done all this neat stuff.
JK: Right. Yes, that's certainly true.
RW: And we had to rewrite our resumes and get rid of all of that.
JK: Yeah, there were two problems with that project. One is it really was hard and the computing equipment that we had in that era was so expensive and so limited in capability, really. I remember we had this big 366, 370, 36067 IBM computer, it would probably just barely fit in this room and it was a tiny fraction of today's PC in capability and it was very expensive, millions of dollars a month to operate and so it was a bit impractical. And then the other great development that, I think, diverted the world's attention from what we were doing for quite a while, was, of course, the invention of the microprocessor because now a semiconductor company could build one part type and, you know, apply it to all these various applications and these specialized custom chips that we were getting very efficient at building but they still, you know, you had to design a separate one for most of the applications. So I think the combination of the fact that the computing hardware was not really where it needed to be and the fact that for a while at least the semiconductor industry thought you could do everything with a microprocessor, which I believe it later turned out not to be true, but it took them a while to get there. Meant that that initial project, you know, was not a business success. I think we have to agree with that. It was a technical success in the sense that we were able to do this but there was also some problems that I won't go into in actually fabricating the parts that, you know, we had some difficulties there too. But fundamentally, I think we did demonstrate that every bit of that was practical and it was kind of the economics and the stage in the development of technology that we did it. It just wasn't quite, its time hadn't come yet but I'm not ashamed of it one little bit. I think we, it was the seminal project of what is now called the ASIC.
RW: Then you got into a totally different arena, communications.
JK: Yeah, it was really funny. I had another colleague at Stanford who had gone to the East Coast and had worked at BB&N, which was the place the ARPANET was developed. Sometime in the late '70's a fellow named Paul Barron at Rand had described a new way of communicating between computers in which data, a stream of digital data, were broken up into small chunks, if you will, called packets. And then each packet contained a header, which contained the address of where it's to go and where it came from, certain other information. And then the idea was that you could build a network of computers, which would route these packets around so you literally could communicate from one place in this network to the other through this mechanism. The Defense Department got very interested in this technology because you could have redundant routing in this network of computers so that in the event of some kind of an attack you could wipe out part of this network and you'd still, if there was a path around the fault, you know, where it had been wiped out, you could still communicate. So it was very interesting to them. So they funded a project with the Advanced Research Project Agency known as ARPA to build a network like this and it was called the ARPANET. And they put about ten or twenty universities and some companies on this network and you could send data back and forth and really, in many ways, it was the forerunner of the Internet. And all of that technology was developed in that period. I mean there was this packet switch network running around. Well, this fellow wanted to commercialize the technology because we were beginning to get, you know, lots of terminal time-sharing and lots of data being sent around, although nothing like today. And so there was a need for, you know, commercial data networks and this packet switching seemed like an ideal technology for that. So, I joined this little start-up in Boston that was going to do that. And unfortunately, '73, '74 was not a good time to be in a start-up. As you'll remember we had a major recession in the country at that time. And without going into a lot of detail, let me just say that the technical group in that company ultimately ended up working for Boeing Computer Services, which was a part of the Boeing Airplane Company really. And they needed to have a lot of this data transmitted around because when you stop and think about an airline flying these very expensive big airplanes you need a worldwide network that can manage parts and everything because it's very expensive to leave a 747 or even a 707 sitting on the ground because it needs a part and so there was a huge network all over the world that Boeing was running to manage their business. And they also had a big terminal time-sharing business. So again, this technology that we've been working on was somewhat applicable to that and they became interested in this group. We'd done a contract with them, a consulting contract, and so they actually hired the whole group and we developed for them a number of data communications devices. Probably the most interesting one was an error-correcting encrypting terminal communications device that allowed you to send secure messages over the public telephone network and also provided error control so if there was any noise on the line it would automatically correct it; a bit of a forerunner of some of our advanced modems that we have today. And so that's what I was doing when I received the fateful phone call, I believe it was in the fall of 1980 from someone I'd known previously at Fairchild named Rob Walker.
RW: It was interesting, as I was attempting to recruit you to go to LSI Logic, Boeing was really trying to keep you there and they made just a staggering counter-offer to LSI Logic and no sane person would have turned it down. It was interesting because they had all the numbers, they had it all worked out in a very business-like fashion and I said, yeah, but Jim, it's our destiny to do LSI. And he said, yeah, that's right. So all the business sense, all the expertise, out the window, it, purely emotional response.
JK: Well, I, by then, there were two things that really influenced me because that was fundamentally true although remember when you described who was doing this, Wilf, and Wilf had been the head of Fairchild at the time the previous project was shut down, probably for very good business reasons, I've kind of gotten over that, but I still wondered if he was really serious about going back into this semicustom arena. But there were several things that in the end made it virtually impossible for me to turn it down. And they were, I bet, they were partially emotional. But to start with, and remember I had been designing systems and I noted something in the systems that we were building there at Boeing. You'd have this ten or fifteen dollar microprocessor sitting on a PC board, surrounded with a hundred dollars worth of glue-logic and it just seemed to me, and by the way, some of my colleagues who were at Boeing had also been at Fairchild and we would all say gee, boy if we just had that old system, we'd get rid of all these chips and we'd build this thing with two or three chips. It would be cheaper and better and faster and blah, blah, blah. So we had actually begun to see the limitations of the classic microprocessor because any real system has a lot of specialized logic around the microprocessor, which costs real money and burns real power on the board. So there was a kind of a technical reason why maybe it was a good idea to try this again. The other thing that, of course, was you're little input here, is that you weighed heavily on my mind because you did feel that one way or another someone was going to do this. It wasn't necessarily not going to be done because the need was there, there had been some expertise developed in the interim. We weren't the only people in the world who could do EDA. So I sort of constructed in my mind this model, you know, how is it going to feel, Koford, if having done that stuff at Fairchild, taking all that heat for it, someone else does it now and gets it right and you're sitting there watching. I did not project that as a pleasant experience to go through.
RW: Well, what we had done had largely been forgotten. And people were rediscovering these things that we'd learned many years before.
JK: And I just thought that the future frustration of that would be, and I'm just predicting I would be somewhat frustrated by that. So, that's what I say, I did in that sense, make an emotional decision. But the other thing that I really must say, as you remember, it took a little coaxing on your part to get me to go out and talk to Wilf Corrigan, but I must say that influenced me a fair amount because obviously Wilf is a very shrewd business man. And you remember in the early '80's the "the U.S. semiconductor industry was done for, the Japanese had won." And so here's Wilf Corrigan whose greatest mission in life is to build semiconductor manufacturing facilities, saying he's going to start a new semiconductor company. And you know my attitude was, well Wilf, explain how this is possible, I mean, you know, it just seems like a climate is in the wrong direction. Well, as you know Rob, Wilf had a brilliant plan, absolutely brilliant plan, because he was going to use a device. He had kept track of what the computer manufacturers were doing, you know, to put the story in context, as I'm sure you know, but the audience doesn't. Wilf had been the president of Fairchild, he had sold Fairchild to Schlumberger and had signed a consent decree not to compete for a year. And so he was off being venture capitalist, I guess. I wasn't around him at that time. Well, he was coming out of that window at the time I met with him, and he had now formulated in his mind a plan for how he was going to build a great semiconductor company. And I must say I thought when he described it to me that the plan really was brilliant because, of course, what he was going to do was going to go to Japan and get the base wafers from the Japanese, buy them in bulk, if you will, and then we would develop and sell a kind of circuit called gate array, where the only thing that was special to the customer was the metal pattern, the metal interconnection pattern and not the diffusions. The advantage of that was is that he could put in just the back-end of the manufacturing facility, which would cost far less than the full diffusion factory and yet, we were a semiconductor company. We would finish the chips, we'd package them, we'd test them, you could go in, you could peer through the yellow optical filter and see people in white coats, and it was a way of bootstrapping into that business that the more I thought about it, it was just really brilliant. And, of course, that's exactly what we did.
RW: It's also interesting that in our initial business plan, we were going to license the software to a lot of people, and in fact we did. We licensed to Toshiba, to RCA to GE, Hughes, three or four others. So we were on our way to having a software business. Much as the company that you're starting right now.
JK: Yes, that's true, that's true. And we did very well at that until finally, there were a lot of other factors at work then because, of course, the success of LSI Logic, and as you know, we had an arch rival too, I mean these things, you know, there's really no totally original ideas. There was always somebody else doing basically the same thing you were doing, you know, roughly speaking. There was another company in the valley named VLSI Technology that, you know, would give us a run for our money. A little different kind of circuits and you know, some really bright guys there, though, I respected them a lot. But those two companies probably primarily in that first half of the '80's decade, really showed the world that an ASIC business was viable as a big business. And I, again, this is where your design factory concept, Rob, came into strong reality and, of course, you ran engineering and I did the EDA in that period and you put together the applications engineering staff and the design centers all over the world. It was a practical business now. People could go to LSI Logic, they could design a chip, they could do it with confidence, when the chip was completed and came out of the factory, you know, ninety some percent of the time it would work and we got that much higher later. So we were able to establish ASIC as a commercial reality that I think was really a great achievement and we'd never done that at Fairchild. I mean it was clearly too early, but at LSI there was no doubt what we were able to do.
RW: But it turns out we could have been in the software business, which is more profitable than the silicon business.
JK: Yes, that's true, but you know, Wilf loved his fabs. And I felt I was in the software business, you know, we had, oh gosh at our peak I think we must have had hundreds of customers, but you're right, people don't view us that way now, but we really were. I mean, we had a complete EDA system that does everything that the current software companies do and then some. Because we had packaging, you know, we had a number of things you would find in a semiconductor company but you don't typically find in a, so we had a. I think for a few years there we were one of the primo places where smart graduate students went to work on EDA. LSI had a good reputation in that period I thought. I know we got a lot of smart people in there.
RW: Well, I have a thesis, and it is that the silicon business, since its inception, Shockley's invention, has been a loser and only Intel, uniquely Intel, makes money at this. Now it's difficult to prove that, because so many of the semiconductor companies have their, they're like IBM and they submerge their semiconductor results, so it's not clear what's happing there. But I did go back with LSI Logic and I looked at all their profits since we started it in '81, total sales of a little over a billion dollars and a return on those sales from '81 to now of 3.4 percent.
JK: It's not been a great business,
you're right, Rob. Of course, I agree with you because what makes Intel great
is probably not the silicon business and more and more people are finding that,
it's what we call IP, intellectual property that is the key and software is
almost pure intellectual property.
RW: Jerry Sanders really said it maybe first and best, that it's the canvas of the, silicon is the canvas and what the artist paints is...
JK: I think it's probably like the publishing business or the movie business. I mean, you know, the core technology film and that sort of stuff is, you know, part of the movie business, just sort of there, but you know, it's the great movies, the great writers, the great stories, the great performers.
RW: Okay, so you were at LSI Logic for a long, long time.
JK: Long time. Fifteen and a half years.
RW: Wow. So what caused that final blowup?
JK: Well, you know, the problem was is that I, fundamentally, was there to do the EDA stuff, although as you know, Rob, we got into a lot of other things, microprocessor cores and we started the lab up at Stanford under Dr. Abasil Gamal and they did the Machen, but fundamentally I was into the IP end of it. And LSI did, of course, begin to exhibit its fundamental roots as a semiconductor company, as you well know, LSI started building manufacturing facilities in several locations and perhaps it began to look a little bit more like a conventional IBM integrated device manufacturing semiconductor company. But the real problem was is that as the years wore on, of course, a very large and somewhat large and fairly competent external EDA industry grew up and companies like first, oh, Daisy, Mentor, Valid, and later companies like Cadence, Synopsis, Avanti, you know, established themselves as major corporations in the valley and it began to become more and more difficult to provide the kind of EDA that we wanted to provide, electronic design automation in a company like LSI Logic. Furthermore, there was ultimately a very powerful argument that began to be used and that is if you buy the tools from one of these generic CAD companies, you can use them in a number of different factories, not just those of LSI, and of course LSI wasn't interested in having their EDA tools used in any factory other than LSI's. So there began to be a movement, certainly was one externally and it even began to happen internally away from the proprietary, we'vw-developed-them-ourselves EDA tools into using the more generic tools and that, of course, led to a, you know, somewhat of a lack of support for that kind of activity. And so finally it became pretty evident that the internal tools effort was slowing down. Now when you've been involved in something as long as I had been involved in that, I have to admit, it's a little hard to give that up, and so there was a somewhat painful period while I, more or less tried to maintain our own internal tools group, but finally it really wasn't working out and so Wilf suggested, I had mentioned to Wilf, as a matter of fact I think in this very room, earlier, that when I retired I would maybe like to go do a small research lab somewhere and work on some ideas that I had. Well, he remembered that. Wilf has a long memory and so when the tensions became to the point where, you know, probably we had to do something, he suggested well, why don't you go do that research lab, Jim, that you always wanted to do and we'll just continue up here on the course we're on. I at the time felt that I wasn't really super happy with that but I was interested in looking at some new things. Because what did interest me was that there's this rule, you know, that Gordon Moore first articulated, that the number of devices that you can put on a semiconductor chip doubles every eighteen months and that rule, strangely enough, has continued in effect to this day. That's why we can put ten, twenty, thirty million transistors on one chip. So, I knew, because I'd also been associated with the SIA and I knew what this semiconductor roadmap was that the technical problems in designing advanced chips were only getting worse and I knew by then I knew enough about the existing companies, at least I suspected, and later it became more or less known, I think, that they were going to get trapped in this problem space and so maybe it would be a good thing for me and some of my colleagues to go out and do some advanced R and D work. Wilf said he would fund this thing. It would be much smaller. I mean I had at the peak of my group at LSI, I think we had three hundred fifty people in it; this would be a ten-man lab and kind of off campus, so it would really, really somewhat like the Stanford Lab, you know, we put together. So, I decided to do that. And so we went off, that lab started in June of '93 and in the meantime, LSI was still running on the existing design tools that my groups had put together. We did four complete systems from 1982 to 1993 and I think by then we had designed twelve, thirteen thousand chips, different chips. Now I think that number is up to sixteen thousand LSI reported that in their most recent annual report. So that's a lot of designs and probably as much as anybody, say save IBM has done. I certainly felt we had achieved pretty great success in what we'd set out to do. And clearly it wasn't just me, I mean, I had the privilege of working with some just splendid engineers for many, many years. Some of whom are still at LSI, some of whom are with me and my new company now. But it was probably time to go work on the future again. So that's what we did. I also was joined in that effort by a Yugoslavian mathematician that had recently joined the company and even though he didn't know anything about, really about EDA and semiconductors, he knew a lot about graph theory and some of the underlying mathematical techniques. And Ed Jones, who was at LSI joined us, who had done all this work that I mentioned earlier and over the years had done the physical design part for all of these EDA systems and he'd run a big group doing that. Doug Boyle who had gone to Sun from LSI rejoined us and Doug had done a lot of work in multiprocessing, what we call shared memory multiprocessing computing and I wanted to apply that computer technology to some of this new EDA work. And so we had a mathematician, a couple of engineers, maybe a little on the older side and a computer expert and, you know, just an interesting collection of expertise. So we started this lab called The Advanced Development Laboratory and within a couple of years we had produced a multiprocessing, what's called a placement program that sped up the process of placing an integrated circuit and was actually getting very good results. So, I thought that was enough of a technical development, we'd filed some patents on it, quite a number as a matter of fact, that maybe we should look at doing another much more complete EDA system. Well, LSI as a company just wasn't ready to get back into that and so finally it became clear that we could keep kind of playing in this lab. Well, you remember the Fairchild days that we mentioned earlier, but you know, we loved being in the research lab, you and I were both there, but you know, we got to the point somewhere around 1970 where we said hey, we got this stuff running, let's go down to Mountain View and put it on line and see if we can sell this stuff. And I don't know, I guess I have that problem, I like to go into the lab and work in the lab for awhile, but then I like to go see if you can use some of that in the real business world. So, we decided, a couple of us, Doug Boyle and I decided in the Fall of 1996 to, well Doug left actually earlier and I decided to join him later to found Monterey Design Systems, which is the current company that I'm, well at the moment, CEO of. And we essentially spent the usual first several months getting funding for it, it's funded primarily by Seven Rosen of Dallas. They funded Compaq and Lotus and they've been a very successful venture firm. And we were fortunate again, I guess, in the sense that we were able to take a completely clean sheet of paper. We did not bring anyone else from LSI Logic. Doug and I were the only two who came directly even to this day. And there's a lot of new technology that's been out there and we've been fortunate in having some world class academics join us to help us work this out and one of them is now going to be joining us full time for a year at the company. And we've got about, oh I guess, fifty five people in the company now, probably half of 'em are Ph.D.'s and we got this thing running and we're working with our, called beta customers now to, you know, finish the debugging process.
RW: The existing tools that Cadence and Avanti have really had their roots way back in our work in the '60's.
JK: Yeah, they do, they do and they have roots in some of the work that Berkeley did in the '80's. There was fellow that we haven't mentioned named Carver Meade who certainly, you know, achieved some notoriety suggesting certain types of new approaches to design. I don't think it's appropriate to make this interview a, you know, any kind of a commercial for my new company. We'll let the next interview of someone else do that, but I do feel that it is time for a new approach. We've taken one, we've got, you know, we can put oh probably as many as ten or fifteen computers all on the same chip at the same time doing the design work which is an interesting development. And, of course, the problems are daunting, I mean the ASIC's of today people talk about three hundred to five hundred megahertz clock speeds. There are four and five million gate designs on the drawing boards, and these aren't microprocessors. This is what the LSI type of product is, standard cell product anyway has grown into. Power level's ten's of watts. We have to worry about, actually worry about, there's a function of EDA now, which is based on Maxwell's equations so you actually have to extract, you have to do field solutions to find out, you know, the capacitives and the resistance in these complex interconnect structures. It's an amazingly complex business now and we, of course, in our current company, will only work on part of it, you know, we're not, believe it or not, this has noting to do with logic simulation or test generation or any of those other problems. We'll leave that up to others. So...
RW: But you've gotten sixteen million dollars?
JK: We have raised sixteen million and we'll probably close some more. Oh yeah. This is an area that, fortunately we've encountered some pretty technically literate venture capitalists who realize that there are some problems that the industry badly needs to have solved. This physical design problem is one of them. There are some other ones, like, you know, masks, just making a set of masks, Rob, is now a quarter of a million dollar deal.
JK: Somebody's got to do something about that because that, you know, that is probably, you know, going to have a serious impact on what gets designed in the future. I kind of look at it, you know, Gordon Moore published a paper about two years ago in which he suggested that the limitation that finally causes Moore's Law to fall off may not be technical, it may be economic. And it may well be that Moore's Law is sort of an upper esentote, an upper ceiling that Moore's Law is a natural phenomenon because it is you know, kind of the best that physics, the speed at which physics and engineering can move in the semiconductor. But what's interesting is that Moore's Law has probably been achieved because of tremendous amount of investment in achieving it and if that investment, and Moore's Law gets more and more expensive to achieve as time goes on. Now if that investment begins to trail off well maybe the fundamental physical possibility is still on that, but you know, the actuality begins to trail off.
RW: So if I'm right, certainly the Europeans have lost money consistently in semiconductors..
JK: Right, right.
RW: ...so we know that. We know that the big American companies, the RCA's, the GE's and so on, they lost a ton of money.
RW: We know that the Japanese lost money from time to time on RAM's.
RW: So if you take Intel out and you say, well Intel is a computer company...
RW: ...and now you just want to talk about the semiconductor industry, indeed, its probably lost money...
RW: And now we're asking for, oh, I need another two billion dollars for a plant.
JK: For a wafer fab, yeah, yeah.
RW: And there comes a time when people are just going to say, sorry, I'm not going to give you two billion dollars.
JK: Right. Well, you know, we've
got this other phenomenon, I mean, I'm sure you were well aware because you
were involved in this in companies like TSMC and Charter, UMC and we work with
UMC and, you know, the giant mega factory that builds everything for everybody,
you know, sort of has emerged. It's not the old LSI, TI,
Fairchild where, I mean, these people are just, they just run this processing plant. It's like in the newspaper business I guess, you know, two newspapers will share a printing plant and I think the publishing industry is an interesting guide because there's not nearly the investment, but you know, I rememeber in reading the history of the publishing industry, you know, in the era of Merganthaler and all of the technology that's used to print pictures, that was the big deal, but now nobody even thinks about that, you just have these big plants that crank out, you know, documents. And the publishing industry is all a matter of intellectual property and writers and, you know, various types of, you know, products that are not unique because of how they're printed, but they're unique because of their intellectual content and I guess that's where people make real money associated with the semiconductor industry.
RW: It's a crazy business, but it's worked for our houses.
JK: It certainly has worked! And you know, I think that LSI is actually doing reasonably well now but LSI's got a lot of intellectual property. It has a lot of unique things that aren't just its semiconductor part of it. But I think you're right, Rob, the raw semiconductor business is a tough business. It's really tough because you have a huge capital investment required. Sometimes you've got to make that investment in a down market so you've got it on line when the market turns around. I mean it just sort of has all these, you know, problems associated with it as a business and yet it's fundamental. I mean none of this would happen if we didn't have semiconductor process...
RW: We revolutionized the world. Unfortunately, we lost money doing it.
JK: Right. At least somebody who understands this will give some of us a medal.
RW: Okay Jim. Thank you very much.