Interview with Harold Vitale
September 16, 2009
Atherton, California

RW: Harold Vitale was responsible for the first high pin count computer-controlled IC tester, the Fairchild 8000A in the late sixties. This was followed by the 8000B and then to the Sentry Series of larger VLSI production testers. After the Schlumberger acquisition of Fairchild, he moved on to a series of larger test design challenges at GenRad, Trillium and Credence Corporation. Largely ignored in semiconductor histories, ever faster VLSI testers were key to the success of microprocessors and ASICS. And Harold was there from the very beginning. Harold Vitale

HV: Good morning, Rob.

RW: So why don’t you tell us about when you were growing up and your family and so on.

HV: Okay. I grew up in Salt Lake City, Utah. And, at that time, the city was, you know, a couple hundred thousand people. Seemed like a big city to me. I went to school there, started University of Utah after high school. And then I went in the Navy for a couple years. And when I - while I was in the Navy, I started out as a deck hand, terrible job. I mean, and I met one of the electronics technicians on board the ship. And he got me interested in taking a correspondence course in electronics. And so I - that’s how I got introduced to electronics. And that gave me a strong interest so that I had a desire to go back to college and major in EE. Turns out that - in my family was just a younger sister, three years younger than me. I introduced her to my Navy friend and they ended up getting married. Prior to that, I was home on leave to Salt Lake City with this friend and my sister introduced me to her girlfriend from high school and we ended up getting married. So we had a - we had a good combination of four there and - and always seemed to get together in one part of the United States or another. Those are my early years. I had lots of - I worked a little bit, odd jobs, driving a truck. This was right after high school. Just something to earn money. Even worked in the Salt Lake City Hogle Zoo and they had a little train track that the kids would ride on. I was the engineer on that train. And when that business was slow, I would go and cook hot dogs and hamburgers in the food stand. So I had a good - I enjoyed my childhood, - I wanted a car, of course, like all young men and my dad had this old ’31 Chevrolet four-door sedan sitting in the garage. And he said, I’ll let you have it if you’ll take auto mechanics and I’ll give you a hand and we’ll rebuild the engine. So that was my first car. I know you like automobiles and I got this thing in pretty good shape and enjoyed it a lot. Well that was all before I went in the Navy. Then I – do you want me to just keep rambling on like this?

RW: Sure.

HV: It’s me, huh? After I got out of the Navy, I went to the University of Utah and majored in EE. And I did - I really enjoyed school. I got pretty good grades and I learned - I thought I learned a lot. Math was a favorite subject. What else can I say? Semiconductors were just coming along at that time. So we had a couple of courses in transistor theory. I learned a lot more about vacuum tubes when I was in college than I did transistors.

RW: So when did you graduate?

HV: 1959. At that time, I’d interviewed several companies on campus and GE was the one I really, really liked because they offered a three-year engineering program. They called it the ABC course, pretty complicated name. But it - it dealt in - first year in electronics was kind of fundamentals, got into statistical communications and some fluid flow and mechanical dynamics. We’d get a lecture from one of the senior engineering people on a topic. Then we’d be given a real problem that had come up in one of GE’s many business departments. And we would do the analysis or whatever was required and write a report and then give a presentation on that. We worked in groups of three. We did all that stuff in the evenings and had worked full-time during the day. That training took about thirty hours a week. Well I met some great friends there. One of them I still have contact with. Another - another couple were - turned out to be the parents of Tom Cruise. And just about the time we left Syracuse, Tom Cruise was born. His dad’s name was Thomas Cruise Mapother II and the actor’s name was Thomas Cruise Mapother III. But anyway, that was just one of those funny things that happen. Interesting things. Well I enjoyed that training. And while in Syracuse - it was in Syracuse, New York, I worked on some missile tracking projects where they used two ground stations to transmit a signal to the missile or the capsule and then it would transmit back and through some math, it’d work out where it was exactly. Well, they wanted to add some command and capability to that carrier. So they had me simulate a phase modulation scheme and make sure that it left enough purity in the main carrier. Anyway, that was fun. And I went to Electronics Park, nice campus that GE had. Their television receiver department was there. In those days, everything was cathode ray tubes. But they were trying to develop a movie, TV movie projector. So it needed to, you know, project onto a real large screen. And the scheme they had required that a thin film plastic film to be coated with a thin layer of oil. And all this had to be done in a vacuum. And then a regular CRT electronics would do a raster scan with video on it, put depressions in the oil, then shine bright lights through it, recover all the signals and the colors and project it. Well, they wanted me to work on how to get this thin film of oil on the tape. And you had to refresh it because there’s enough energy in a beam to kind of make the oil get sticky. Well, in - so in fooling around trying to figure out, get some ideas in the lab and I took a thin metal rod and dipped it in this oil and went over to flick a drop onto this thin film and I saw that the oil drop - the drop of oil snapped down to the film. I said ah, Millikan’s oil drop experiment. Well, remember that? We’re looking at scientists determined the charge on an electron by charging up an oil drop and then calculating or measuring how fast it was attracted to another charge. And he, knowing the weights and all that, he could do it. So I said well I’ll put some oil in a syringe with a needle, you know, like what we get when we go to the doctor. And then I’ll apply a large potential between the two and see if I can move oil. Well, sure enough, I got this little stream of oil coming out of there. Well, I won’t say more about that except later on, I realized that that might have been the precursor to ink jet printing, without even realizing it back then. You know, printers were big, huge mechanical things. And so that’s what I did back East. And I came out here and - and worked in Palo Alto for GE and designed - I was kind of a junior engineer but I did some design on medium power microwave tubes. This one was called a ring bar structure and it was the topic of an engineer’s thesis at Stanford. So we had a pretty good relationship with Stanford at that time. We finally got a tube that we had targeted for the Stanford Linear Accelerator. Varian beat us out on that one. I don’t know whether I was the cause of that loss or not. But I had a lot of fun doing that. I kind of realized that I would enjoy working with computers more than traveling wave tubes. I could see Solid State was probably going to displace all those traveling wave type tube devices at some date. So I went to a General Electric computer lab in Sunnyvale. And that’s where I met Jim Downey and Bob Shriner and some other fellows. And they put me on a project that was involved with the development of a printed character reader, automatic, you know, document reader. They had previously designed the magnetic print reader that is still used on checks. Anyway, that was kind of fun. And then GE decided to close that lab down and move all their operations to Phoenix, Arizona. I went there. That didn’t work out. They wanted to move me to Oklahoma City. So I had interviewed at Fairchild and I called Dave Masters or Bob Schriner, one of those two guys, I called them up and I said hey, is that job still open? Yep. Would you consider me taking the job and they said yeah. Come on up. So they moved me back here and that’s - so that’s how I got to Fairchild.

RW: And that’s how GE eventually got out of the computer business.

HV: Right.

RW: It’s just – just totally incompetent manager.

HV: Yeah, they just couldn’t do it. Boy they put a lot - they must have put a lot of money in that; I don’t know how much money went into that. They finally sold that to Bull, I think, French company. But anyway, I met Bill Davidow. He was the head of the computer lab and Bob Sahakian was there. Sahakian built a little printer that required special paper and a set of - it was a dot matrix printer with eight electrodes that went across the paper, we’d give them a pulse so that we’d print a little character and I worked on that. That was kind of fun. Anyway, at Fairchild, the project had already been thought through pretty with architected concepts to provide a functional tester for the bipolar and MOS, I think it was called MicroMosaic, the very earliest - we call that a gate array.

RW: MicroMatrix was the gate array and MicroMosaic was the cell based.

HV: Oh, okay, yeah. So -

RW: Standard cell.

HV: Yeah, yeah. That’s – that was what was successful, right?

RW: Mm-hmm. Very.

HV: So the idea there was to get something that was better - well, first of all, Fairchild Instrumentation was in the automatic test equipment business but they didn’t have anything available that could handle large numbers of digital input and output pins. So that led us to say, well we need something better than a big array of toggle switches. And so we used a PDP8 from Digital Equipment Corporation, 12 bit computer, rack mounted and that was mainly the control and its memory provided the storage for the test vectors. Jim Downey helped on pin electronics. I designed some logic interface and did the software and a couple of techs helped. We put that together in a short time. It allowed you to program the test vectors, apply them at a fairly slow rate, maybe maximum of 10,000 vectors a second. Didn’t have any programmable timing, so it was pretty much a tool to learn more about what we - we needed to do to be - to provide effective testing. We got all the test vectors from your CAD group. And that worked very well. The IO for that - that system was just a teletype, with paper tape reader and printer. And it was pretty difficult to handle large sets of data. So we put together a high speed paper tape reader. DEC had a proprietary mag tape but it - it wasn’t compatible with IBM’s standard tape. And so anyway, we got that going. We only built a couple of those. Instrumentation group fabricated it and a technician, and I de-bugged it - we used two together. They were used for some of the early products. It didn’t have programmable reference levels for the drivers and the comparators. You had to manually set those and set up the power supplies for the device under test. So we learned a lot about requirements we either overlooked or felt like we had to address later. And that evolved to a - well from that 12 bit controller, we looked around. We were going to - to buy a computer which in hindsight was good and bad. But Dave Masters who was working at R&D said hey, what do you need? If you’ve got 12 bits, I can give you 24 bits. And that’d be memory and instruction. And man, that sounded pretty good to me and the speed sounded good. And so he proceeded - him and with some help to des - des - design that. And it had a kind of a - pretty good architecture. It had two memory buses so you could interleave data transfers between the tester and some peripheral, concurrently. It had an interrupt - priority interrupt system. You probably remember that when you were working on it. We had direct memory access interfaces for mag tape, printer, hard disk, card reader, and terminal. So that gave us a good controller. The next version of the tester we called the 8000B. Went from A to B. More computing power. We sped up to where we could transfer data from memory for vectors. We moved that up by factor of three or four, I think. Still not fast enough but an improvement. And that was a direct memory access with some smarts built into the control logic to process the different fields that we were transmitting to the pin electronics and formatters and pin electronics. I didn’t realize it at the time but I think somebody, maybe it was you, gave me this idea that said, you know, when you’re moving data, setting it up into where you got a bit per pin or several bits per pin, there may be fields that don’t change. So you don’t have to store those or transmit them if you provide a - some kind of a flag to tell you what to do. So we did that. And when you put the pattern together, the software’d figure that out and then when we sent - transmitted that to the hardware, it would only move the data from the master to the slave side of a holding register if a certain trigger came along. Well, you know, that - that was really useful because we only had 4 - 4 or 8K of memory to hold the patterns. And we needed some compaction schemes. Several years ago, someone said to me hey, you put together about the simplest data compaction scheme I’ve ever seen. And I said, well, what did I do that on? Oh the 8000. Oh I - it just seemed like the right thing to do at the time. But anyway, that - that was the 8000B. We packaged it with TTl. The 8000A, I used RTL for the logic interfaces. And, in those days, we in the instrumentation group had cards with OR gates or flip-flops on them and you specialized the interconnect on the back plane of all things. Then on the next version, we used T 2L, MSI. That was a step forward. And we added a few things to that version. It finally got all new packaging which is pretty convenient, had a middle part with nineteen inch dimension standards. And then it had two wings that folded out for servicing. That’s where most of the electronics - tester electronics were. And the center part of the package housed the computer and - well no, the computer we packaged in the side doors but some of the power supplies went in the center. They were big then. So we built a few of those and some of them ended up at Fairchild, MOS division. And we could get patterns in them from the CAD system and apply them to the device under test. It became pretty apparent though that we had to move the pattern speed up, had to have programmable timing edges for drivers and comparators or strobes to the comparators. So we had a marketing engineer from national hired by Bob Schriner. And he had some good ideas for, of all things, building a separate memory buffer with solid state RAM. And that’s - that was what held the test vectors. And then behind that we built sixteen timing generators to make edge placements and then the test vector data and the timing went into a formatter board where we could format RZ, NRZ, different format wave shapes. And so now we had a real tester. You could - you could say well, given that the timing edge is specified to be here, when should the outputs change and at what time. So we’re getting closer to being able to test the device like it was designed to run. And gave us some clock capability, which we hadn’t had before and, as you know, lots of the devices need clocks that are much faster than the data rates. So we were feeling pretty good about that. We had programmable voltage references for the pin electronics and the power supplies. And that tester caught on. The test head was compact. I wish I had a picture here. It’s about - we called it a carousel. It was about that square, about that deep and the pin electronics, that’s with - the card with the drivers and the comparators on it, were on small cards that fit in there just like on a Kodak carousel projector. In other words, where the slide went in, we plugged the card in. The output/output of the cards were in the center. So we could have a DUT board, DUT device interface card, from the pin electronics to the DUT that mounted right there. And - and that allowed us to keep the wire distance from the pin electronics to the DUT fairly short. And it’s good to keep it short because you want the round trip to be within the - the rise time of the drivers - of the tester and the device under test. If you don’t, you get some reflections. You helped me a lot with guiding me to some papers on time domain. And so that was really useful that - well I said I was designing traveling wave tubes so I had been working at pretty high frequencies.

RW: Now is this all at Fairchild in Palo Alto?

HV: Well we started it in Palo Alto. Fairchild bought a facility in Los Altos Hills and that’s where all of the instrumentation group moved to. No, no, who went there? I know that’s where we went. We were building the testers right there so I guess it was just the test group. And then finally, we moved that operation down by the San Jose airport on Technology Drive. And we changed the name from Fairchild Instrumentation to Fairchild Systems Technology. I had an interim time when I had an office in the Rust Bucket. And I got so engrossed in working - oh we were in Sunnyvale too. That’s - that’s where it was. I got so engrossed in working on this tester that I neglected my office. You finally had to clean it out for me and send me my personal belongings. Oh man, I guess I was too focused on getting the tester work done. Yeah, so I left out an important part. We had good facilities. We always did.

RW: Now who - when did the tester get the name Sentry?

HV: That was right after the 8000B. So that would have been in couple years after we finished the A and the B, two or three years after that. And then we had a series - let me go into that for a minute. We had a series of Sentry testers. The first one did not have this high speed buffer. The second one did and we called it Sentry II. And we continued to evolve that tester through, golly let’s see, through the late sixties up into the seventies. And there were a lot of interesting features added like the buffer memory needed to be large - we needed a lot of vectors, you know. This as devices grew, the number of vectors got large. And so it was too much to put in a 4 or 8K memory. And we didn’t want to have time insertion while that buffer was being reloaded. In other words, we didn’t want to run a - have to run a bunch of vectors and then have a long time delay while we put new ones in be - and so Bob Houston - do you remember Bob?

RW: Absolutely.

HV: Probably the most brilliant applications engineer and he would have been brilliant in whatever he did, just brilliant. He could take those testers and make them do things we hadn’t even thought of. And that computer we called the F24, Fairchild 24, 24 bits, Bob used all the capability of that and it turned out to be a wonderful tool because with that tester, well anyway, let me finish. Bob was instrumental in defining an instruction field that we added onto the buffer memory. And that gave the tester the capability to repeat vectors. So you could say on this vector, repeat it N times. It gave the ability to call a vector subroutine. If you had a set of vectors that you used periodically, frequently, you could make a call to that subroutine. It would execute them and return to the statement following the call. You could change IO control on the fly. That really allowed us to do a lot but I remember there were still situations where we’d say gee, we just need a little more memory. And probably always has been. Always will. And Bob would find a way. He’d look at the vectors and he’d find a way to use those tools we had and compact things down.

RW: Did the 24 bit computer, was this sold to anyone else?

HV: You know, no. At one time, Bob Schriner had an idea that it would be a good thing for Fairchild to get into the minicomputer market. And he hired Kay Magleby from Hewlett Packard who had been instrumental in developing their minicomputer - 16 bit computer. So Kay Magleby came into the company and set up a design group and he took over responsibility for that computer. It was never brought to market as a standalone system. But the work that Dave Masters and Chuck Runge did - Chuck was software architecture. Dave was - did the hardware architecture and - and they did a lot of the design. But Chuck did a - really a nice system. He had a good operating system. You know, for file transfers and movements and he used a higher level language so that the user could program it. It was a COBOL-like language which that may not mean a lot to modern people but it gave you English-like text in a statement. So you could say set power supply 2 to some value and the value could be a variable. So you might want to calculate that and then apply it. So we - we all - we were all very proud of that computer. And we couldn’t figure out why it wasn’t marketable but 24 bits was not - everything was 16 bits then going to 32. That may have had something to do with it. But it did our job quite well. We eventually went from ferrite core memory to solid state memory. Fairchild dynamic MOS 4K’s helped them solve a few problems that they didn’t notice. But anyway -

RW: Well the one advantage the computer had was that it made it very difficult to clone Sentries..

HV: Yeah.

RW: - which would have been an obvious thing to do because it was - it was the leader.

HV: Yeah.

RW: But it was burdened with all the stupidity at Fairchild and all the overhead and - and all that. Had - had you used a DEC computer, mini computer, or one of those that was commercially available like Calma did, it would have been very difficult. Then they could have cloned the Sentries.

HV: Yeah.

RW: And undersold Fairchild.

HV: Yeah.

RW: But because it was this one-off computer, it was very, very difficult to do. So it turned out to be a competitive advantage I think.

HV: Yeah. Yeah, you know, Teradyne did a 16 bit, their own 16 bit controller. Do you remember that?

RW: No.

HV: It came out on the J259. Nick DeWolf on one of the history tapes mentions that. And I was aware of it. You know, yeah, think about - you brought up competition. When we first started introducing a computer controlled tester, there was a set of users who - who just, you know, they liked - they were used to digi-switches and POTS to adjust voltages and all kinds of things. And so they kept asking for more - for more manual control. Well we were driving towards more computer control. You know, why? Why not? I mean, lots of flexibility, lower costs, easier to calibrate, you know, that’s the way everything went, of course. But, I know there was - somebody who came out with a competitive tester and it was just loaded with POTS and switches and dials. And customers would say look at that. That’s what I want. Why can’t you get that in your head?

RW: And it’s also very expensive.

HV: Yeah.

RW: Those are more expensive than bits in a memory.

HV: Oh yeah. Look at the orders of magnitude difference. And anyway, yeah, it was fun going through those things. That particular computer allowed us to - and the language - allowed us to do a lot of things like if you were collecting test measurements, let’s say leakage currents and you wanted to monitor how the leakage current was doing across a wafer, we could do a histogram easily. We’d just say - the programming statements were just a couple - take this measurement and then you’d - you’d define it to be an element of an an array. And then we’d say print the array, you know, software array. And there you’d see the histogram. So we had a lot of fun doing that. I remember one time when I was given strong directions that I needed to keep closer track of what every engineer worked on. So we had all these codes. Say for two hours this week, I worked on this code. Three hours I worked on that. Ten hours I worked on that. So I said I’m embarrassed to assign this bookkeeping problem to somebody. So I took and - and had somebody punch all this data in cards and then I wrote a program in the tester language, FACTOR, it was called. Fairchild Algorithmic Compiler Tester Orient, I think it was FACTOR. So I wrote a little program that would take all this data and - and print out a nice report that I could turn in and then I could go back and work on the real stuff. So -

RW: Well the Sentry went on to become a cash cow.

HV: It did.

RW: And very, very successful even when they sold the semiconductor portion, when that was sold to National, Fairchild continued to make testers.

HV: They did. Yeah, and they continued until just several years ago. When that change took place, Schlumberger ran the tester division. And just before that happened, marketing was telling us we were getting hit by a smaller, lower cost, lower performance tester. So we took some pieces of a Sentry and packaged them up into something we called Sentinel. Well it was program compatible with Sentry. It used the same test head as Sentry. So it was a great, its appeal was that you could do your initial development on Sentry’s and then if you had a long production run, you could move it over to this smaller footprint, lower cost system. But you’re right; a lot of those were sold. IBM encouraged us to expand the pin count from 60 to 120 pins, which we did in a brute force way but we did it. We just slaved two Sentry’s together, made 120 pin tester. Later on in my career, we did it differently but IBM bought a bunch of those and we sold quite a few of them in Japan as well as the smaller pin count testers.

RW: Now, at some point in time, though, you left Fairchild.

HV: I did. In Well it was right after Schlumberger came in. They reconsolidated Sentinel with Sentry and I said well what do you want me to do now? Well, we’ll find something for you to do. So anyway, a few of us knew that GenRad wanted to make an entry into the digital test market. So Brian Sears who had run Zincom - a memory test company that was bought by Fairchild, left Fairchild and started this project at GenRad. So we went over there and worked on that. It was similar to a Sentry but not compatible. You know, it had to be a little bit better. So it was a little faster. The timing system was a little bit better. We started using ECL gate arrays for the formatting and - and got those initially from LSI Logic. Your company offered those when you first started and those worked - they worked good. I liked them. But after that tester was - the GenRad tester, one version went up to 120 pins and they wanted another version up to 240. It didn’t work out. I mean, the market just didn’t accept it quite well. But we did it in a hurry. Maybe that was part of the problem, I don’t know. Anyway a few of us heard that LTX wanted to get into the digital test market. They were a spin-off from Teradyne. And their - their main product line was analog. Analog with just enough digital so you could test D to A’s and A to D’s. They had been very successful. They put a lot of those testers out in the field. They used a Data General 16 bit computer, which later was dropped and they found a way to replace it. But their main contribution was I think in applying sampling theory to testing analog devices so that required a pretty good digitizer, a good digitizer. So you’d digitize it, the signals and then, you know, then transform into the frequency domain and you can learn all you need to know. And it was fast. You didn’t have to have a big array of filters to get a Fourier transform at low frequency; they take a long time. I think they hadn’t been started at Teradyne but a guy named Sol Max was really brilliant MIT graduate. He used to tell us that he had two first names, Sol and Max. But he put together a lot of the early work for testing CODEC’s in production, using their testers. So that was when, you know, sampling was introduced in the telephone system. Brilliant guy. Anyway, they - they wanted to expand their product line into digital - well I called it - we used to call LTX big A, little d. So lots of analog, and a little bit of digital. They wanted to go into a lot of digital, big D, little a. So the CEO of LTX, Graham Miller, made a proposal and we worked out a deal to start that kind of an operation up in San Jose mainly because here’s where the experience is - the earlier designers were either here or back in Massachusetts with Teradyne and a few sprinklings other places. So we made an agreement where we would design a tester and we were going to aim it at CMOS testing. The financial agreements were that they would loan us most of the money. And we’d be a corporation with stock issued. But, and then later on, they’d buy the stock from us if we met certain milestones which were, you know, two working prototypes, had to show that we could do it more than one time. At least two customers who paid for the tester to show that it did the job they wanted and then we had to break even for about a year or in the time frame of a year. And we did those things. It wasn’t easy. I think we spent about - I think about twenty-five or thirty million dollars for the development. We did it in little over a year. The key contributions we made - that were made in that system were that it had timing per pin. So every tester pin had its own timing generator. So you could program edges - each timing location for each pin independent of all others. And what’s the advantage - what’s the advantage of that? Well there were a couple key things. Once we could do that, the job of calibration was easier because we didn’t have to worry about accuracy - we could achieve higher accuracy because we only had to worry about what was going on on that channel, not this timing generator that supplies three other pins and where do you locate the calibration points. And we developed a TDR technique to measure a distance from the formatter out to the pin card and then a method for calibrating. So that was one big thing. Timing accuracy and flexibility. That meant that we could do things like if you had a hundred and twenty pin tester but you were only testing twenty pin devices for some number of runs and they were in volume, it was fairly easy to partition the memory, with its own timing, up into multiple sites and then essentially run testing concurrently. So it was a throughput enhancement. So that was good. We built in a picoammeter on each pin because we believed that leakage currents were going to be critical. That was not quite as successful as we’d hoped for. You - you had to - well it was fairly fast. It used a - a ramp technique. But we had similar - no, we had a pattern memory with an instruction field again. That we just made deeper. And we packaged - it was an all ECL machine. We didn’t have enough confidence to go to CMOS like we probably should have. So it was ECL. You guys were no longer doing ECL. So we went to a company in San Diego, I forget their name. Anyway, they finally got us some - some parts. It wasn’t easy. Well there were a lot of problems they had to solve. But - so that was that tester. Then it evolved to - up in speed. Went from twenty megahertz up to thirty or forty. We could produce narrow pulses. That timing generator per pin helped us there so five nanosecond pulse widths or less, meant we had to have some pretty fast edges. One thing I was kind of proud of was we had a carousel test head again but our edge speeds were pretty fast, and the speed of CMOS edges is fast, as you well know. And so CMOS rise time was much faster than a round trip even though it was short. The round trip is from pin electronics and back so you’d get reflections, you know, the device would send an edge, to high - high impedance comparators, and that edge would be reflected back. And now the device is either in the high or low state which is low impedance. So edges get reflected - inverted and reflected back. So you had to be careful where you strobed or your - you couldn’t really check the timing of the device properly. So I remember some old notes that you put out - it was in the Fairchild ECL manual - using diodes to clamp. Well, we had a - a diode bridge on each pin and had - a lot of the testers had that for - so that you could use the diode bridge to determine - if the device - device output was low in the steady state, then it had to sink so much current through the bridge or if it was high, it – device - had to source some current through the bridge and you could program what those currents were - what two currents were. So I said there must be a way to use these - the diode bridge to do some clamping here. And - and on top of that, we had 50 ohm drivers but you couldn’t always use it as a terminator because the device, early devices didn’t want to drive 50 ohms. Anyway, came up with a technique to use this diode bridge to do some clamping and get rid of all that ringing. And I shared that with some other guys. We got a patent for that, my one and only patent, Rob, but I was really proud of that. I don’t know if the camera will show this. Can you see that? Anyway, there’s my name while I was at LTX and the front page of the patent’s always got a little bit of the circuitry. It was simply for using digital to analog converters to control the currents and we’d set the voltage to bias this bridge so to act like clamps. And I was pretty proud of that. We could get a lot of publicity out of that because prior to that, someone at Bell Labs, their pitch was “keep the distances short enough.” Always keep it less than a round trip delay of the device. Well, you guys are building such high speed CMOS that - that wasn’t practical. So we - we had to come up with a way to allow full speed testing and not have the quality of the signals corrupted. Well we sold a lot of those testers. Actually Intel bought a lot. We had a big contract with them and they used it for some of their early 32 bit microcomputers. That’s when their advertisement came out, you know, comp - what was it, computer in the - in the box or some - I don’t know. Anyway, I can’t even remember the name of that first version but a tough customer. The high accuracy paid off and they wanted it because that meant money. When you’re binning - when they were binning their devices, they didn’t want to throw away a good one and they didn’t want to ship a bad one. So -

RW: Well binning relates to testing and then having various specs usually for speed.

HV: Right. And so, yeah, so that was the binning. And the software, we had a Pascal based system then, I think. And so, yeah, we could sort out, you know, the speed categories and then the customer could sell them at different prices. The highest speed ones obviously at the highest price. So that was - that was a big driving influence on them wanting that tester. There was only one other company doing a tester per pin at that time, I mean, timing generators per pin. They were eventually purchased by Teradyne. They got their tester introduced a little before ours but we passed them up, especially with that Intel contract. We beat them out on that and so that was was a good feeling. And things were rolling along really smoothly. We were meeting all of our goals for the financial part and getting ready to transfer - the name of the company was Trillium - to LTX when we had one of these market slumps. And so we didn’t - we held on for a longer time. The name Trillium was interesting. I don’t know if - did you ever hear of it? You may have.

RW: Well I don’t know how it - how it came about.

HV: Yeah, here’s - do you want me to tell you how it came about?

RW: Yeah.

HV: Okay. Here’s the founders of Trillium. Let’s see. That’s me if you can see it. That’s Jim Healy. He was marketing guy. Michael Chockley came to - he was - came to Fairchild after you were out of Fairchild. Tony Taylor was the software expert. Gary was - Gary Ure was a logic expert. Carlos was a hardware designer/manager, could do almost anything. Had - had the drive to work day and night and then Ed Chang’s, logic and then Bob Houston who was - I mentioned earlier. So this was our rogue’s gallery picture when we founded. So all those were founders. So I was a co-founder. Tony had never worn a suit in his life and he showed up for this picture and Michael said we’re - Michael said we’re going to delay the - the photo shoot for a while. He took him down and bought him a suit, got it - the pants hemmed up and look how handsome he came out. Bob went to Credence. There came a point in one of these turndowns where LTX just decided to take off their top level, reduce cost by removing the top level managers. So that’s when Bob had just left there before me. Jim had already gone to Credence. Tony went to Credence. Carlos started his own assembly company. Ed Chang decided to try a startup in Taiwan in testers. And Harold went over to Credence. And there I worked on what they called the - well it was a small tester. It was - it was almost a - almost a tester on a board. And they - all the boards were in the - in the test head. It was very small compact tester. I know you told me LSI - I know LSI used those and it was a great tester. It had all the features that I talked about, timing and - and formatting was done on a CMOS chip. And they did - they temperature compensated that by - I don’t know if you - if you knew about that but that - that was what kind of held us back from CMOS at Trillium. We - earlier we all - we went to CMOS for everything except the timing generators. And -

RW: Because as the temperature changes, so does the delay.

HV: So does the delay and - and we’re trying to hold edges to 250 picoseconds which - and the temperature the temperature variation caused shifts larger than that. So but if you get the data there and then clock it, you’re okay. And all of our formatters took the format and positioned the - the clock so that in the final flip-flop, it - everything that had a rising edge got clocked, had the same clock, same logic structure for the output flip-flop. And anyway, it was those kinds of things that really helped us. And I wanted that - well greferring back to Trillium company. I wanted that tester to be really rugged because it’s got to be rugged. And, I mean, the test environment in like LSI Logic is very good, clean rooms but, you know, you got to get DUT boards in place. If something goes wrong, you’ve got to change boards in a hurry. You’ve got to diagnose it in a hurry. So we put a lot of emphasis on good diagnostics down to the board level, of course, maybe a little bit deeper sometimes. I didn’t want a bunch of cables coming out the front-end where you had to get in there and disconnect cables to get a board out. So we brought all the cables off the back plane side of the the card cages. And we had pretty large boards. We had metal runners on them so that they had good alignment in the card cage. That meant the card cage had to be rugged. So the designers came out with their first cut at the card cage and I said ah, it isn’t rugged enough. How do you know Harold? I said I have to be able to stand on it and it doesn’t take a permanent set in its deflection. They thought I was crazy but that worked well. I mean, that was the tool that I used. It was 150 or 60 pounds. Okay, well anyway, I was over at Credence and with the SC, that was the name of the tester. At LSI Logic, I’m pretty sure you guys bought a lot of those testers.

RW: Mm-hmm. We actually put one for diagnostic work - we actually put it in a vacuum chamber. It was small enough you could fit it in a vacuum chamber.

HV: Ah.

RW: And then we used electron beam so that we could look down and we weren’t limited to the pins around the integrated circuit.

HV: Oh yeah.

RW: We could actually go down to the little fine lines which are very tiny.

HV: Yeah.

RW: And we could see the ones and zeros as they went up and down.

HV: Yeah, great.

RW: And so it was great for diagnosis.

HV: Oh yeah.

RW: We could really. And -

HV: Compare that with -

RW: Some IBM guys came by and we showed them that and they said God, we can’t do that. And it was the combination - well it was having such a small tester, you could put it all in a vacuum chamber.

HV: Yeah.

RW: And the price was right. And we got involved very, very early with them when they just had a cardboard mock-up. And -

HV: When I got there, they still had that mock-up.

RW: Well -

HV: Anyway, they were shipping the tester.

RW: Going back to LTX, did you ever make any money on that, on the -

HV: LTX, yeah.

RW: Was -

HV: Yeah, I made - I didn’t make as - we didn’t make as much as we thought.

RW: But they did buy you out?

HV: They finally bought us out and probably made about half what I anticipated. Each of the founders had their own personal investment in the enterprise as well. But it was good to have LTX to back up our bank loans and the milestones we negotiated that, and talked about that for quite a while. And I thought they were good milestones, you know. First, can you design something that works? Does it work well enough that your customers’ll use it and pay you for it? And can you run a business and be break even?

RW: Yeah. Well typical of venture capital type -

HV: Well that’s probably where they where those originated then.

RW: Okay, so did you eventually get out of the tester business having been in it?

HV: No, I stayed in the tester business and I just got out two years ago. The last several years were at Credence and I worked - I - they needed kind of like a - I was kind of tired of running big projects. I mean, you know, they take a lot of time and I had neglected my family enough. So I just didn’t - and now my kids were grown but I wanted to have more time with my grandchildren. So opportunities came along to run a whole new project but I said, you know, I just don’t have the energy to do it. And they said well how about working on this SC and we need somebody who’ll just make quick - good, quick decisions on things that need to be done and I said you want a mercenary. Yeah. We need a mercenary. So I did that. Bob was there. He did some initial work that that showed that the tester was capable of going from an accuracy of 500 picoseconds down to 350, although we couldn’t get it. They hadn’t been able to get it there. So that was my first assignment to figure out what was causing some signal distortion and was hard to find but it was a simple fix, a few resistors for a transistor on each channel. But it was worth it because before I got there, they had sold a bunch of testers with 350 picosecond accuracy to be done. And so it it really needed a solution. And I found one. So that was great. And then the - the custom IC they had in the pin electronics went obsolete. The company doing it just said we can’t manufacture it anymore. And so I got the project to find a part, keep the same specs, don’t make them any better, don’t make them any worse and to reduce and if you can reduce the cost, do it. So I accomplished that and that was good. Saved a lot of money for the company and we sold a lot of those testers to Test Houses in Taiwan. They were a huge customer. So then it was time to work on a new tester and the next version was really a small tester. The whole tester, you know, a floor style PC, it’s about that high, that wide, and we put the whole tester in that box, in a box that size. It had a server, a small cabinet with a PC in it to run it and an AC to DC converter. So there was only a few cables. The power cable, we sent 48 volts out to the tester box and an Ethernet cable. And then the internal controller tester was on PCI standard, Peripheral Control Interface standard with nice bus - had a lot of features that were good. And then every card took 48 volts in, converted that down to what it needed, 5 volts, 2 ½ volts, whatever. And those converters and were really efficient. The switching devices that were developed had really low losses so we were able to put a lot of electronics in a small box. The - they had a memory group there who had developed an IC. I think - I’m not sure who built this but it had 64 channels of format and timing and a lot of buffer - memory buffer, not all the memory on it. And so we used that in this logic tester. And we got 128 pins on a card about that size. And you’d plug the card in and you could have three or four of those cards in the tester. Because it was on this PCI bus, we could buy analog instruments like digitizers. They just plugged in. We had to provide the software to drive them. But it gave us a pretty good big D, medium sized a, product and the DUT board had a big mechanical assist that plugged it right into the top where all back edges of the cards were. And that’s where the pin electronics were located right along there. So we still had to use techniques to damp the ringing out. But we knew how to do that so all that worked good. That - that tester was called the Diamond 10. Credence a few years back purchased the Schlumberger test operation but, by then, the managers had done a leveraged buy-out. And then after a few years, Credence bought that operation. I believe that tester had every feature you could think of. You know, sometimes you get too many features in a tester. And it makes it difficult like I remember when I was at GE, I was in Phoenix and I had a demo of their new tape drive. Well it did everything, you just put the reel on there and it threaded the tape and, I mean, it did everything. But things kept going wrong with it. And you know, just needed more engineering I think. But anyway, this Schlumber - ex Schlumberger tester was liquid cooled. Some of the testers I worked on were liquid cooled too but Schlumberger used this liquid where you immerse the boards in it. And it’s great as a coolant. You got to make sure you don’t have leaks, stuff like that. But that tester, they sold only a few. I don’t think it’s done much. They still build it in low production. Anyway, LTX during this economic crunch, LTX and Credence merged and now the company’s called LTXC. So they have some pretty good technology for the near term here, I think.

RW: Didn’t they buy Tektronix too?

HV: Pardon me?

RW: Didn’t Credence get Tektronix?

HV: Yes. Should have mentioned that. Their early main product was a Tektronix tester. And they bought that operation. They kept the engineering and the manufacturing in Beaverton, Oregon. And some of the engineering was done here in Fremont. But that was quite a successful tester. It was around when we were introducing Sentry. Sold mainly to Department of Defense oriented facilities. But it had a lot of good features. It was a pretty large footprint but had lots of cables. I wish I’d have had some pictures to show the evolution of size and that. But Rob, it was, you know, we had to follow the trends in semiconductors.

RW: Right, Moore’s Law.

HV: Yeah, I mean, -

RW: Smaller, cheaper.

HV: Yeah, smaller, cheaper.

RW: Faster, faster.

HV: Faster, more accurate. So that’s when I was working on that Diamond 10. Well I architected and designed the timing calibration scheme for that tester and I’d heard earlier about - in a paper at a conference, a Japanese company gave a paper on what they call a - a virtual ground technique. So what that meant was your calibration fixture is just an adapter board and you have a trace from every pin of equal length, equal physical length, that goes and join together at the center of the card. Well, if you think about just one of those lines, all of its neighbors, hundreds of neighbors, all in parallel look like a real low impedance so it’s a virtual ground but it can get down to an ohm or two. So what that allows you to do is establish a reference pulse. You can drive all the pins concurrently except one. And you program it to be a receiver. Well all these other pins or signals come into this node and they just essentially add up. Some are at different places in time but you get a composite pulse and then you align your comparators on each pin up to that composite pulse. So now they’ve all been aligned to one edge. You don’t know exactly where it is in time but they’re de-skewed. And then you turn around and you align each driver to its comparator by program control. And you can iterate on that. Well I found that we only had to iterate twice, usually the first - well second iteration. Then go through the process again. And that allowed us to get plus or minus 350 picoseconds of accuracy, and a simple piece of hardware. Well I had a phase detector on there which consisted of a couple of MSI devices, so I’d have a CMOS NOR gate and a MECL comparator. And anyway, I could do a comparison. I had to finally get down to time alignment to our clock, right. So I had to do a phase comparison of the system clock and get everything lined up. Anyway, I enjoyed that kind of work and it was good for an old man, you know because I didn’t have to multitask too many things at one time.

RW: Well, you’ve always worked in Silicon Valley.

HV: I have, yeah.

RW: Yeah, well other than your GE days. Now what makes Silicon Valley successful? What are the attributes?

HV: The people, I think.

RW: But where do the people come from?

HV: Well, I think our old buddies at semiconductor attracted them from other places. And so that’s one element of it. I mean, these guys in this photograph are all, you know, top notch people. And some of them came from outside the area. I think it’s a nice place to live, great education, you know, Stanford, Santa Clara and then over to the East Bay. And I think it drew the people that it took to grow this valley. I mean, entrepreneurs, capital venturists. I don’t know, what do you think?

RW: Well it’s a magnet for smart people.

HV: Yeah.

RW: And ambitious people.

HV: Yeah.

RW: You can - engineers can actually make a lot of money.

HV: Well that’s key.

RW: Well, I’m going to come over there and talk a little bit about the importance of testing.

HV: Okay.

RW: I wanted to say a few words about testing. If you read any history book of Silicon Valley, or of semiconductors, you won’t find anything about testers in there. And yet it was, for me, it was a central part of our success at building custom integrated circuits, so-called ASICS. We started out at Fairchild and we made the automatic generation, well semiautomatic, based on the customer’s simulation, we made that a requirement on each prototype. So we didn’t wait to get a production order. We wouldn’t even go to fab until we had a production test tape in our hands. So when they came out of fab, we were able to see if they passed the test. And that would tell us that it was a good design, at least from our standpoint. It might not work in the guy’s system because he had a design flaw in it but it worked as advertised as the simulation showed it to be. And at Fairchild, we turned out several hundred prototypes that were all production ready. And some of them are still in use, a little frightening. They’re used in arming atomic weapons. Guy from Sandia visited me a while back because they were going through an analysis to see if these things would still work, whether the bomb would go off or not on these thirty year old parts. But we looked at it and, sure enough, the work is fine. Let’s hope they don’t have to. Any rate, so later on, we started LSI Logic and we brought that same requirement that the design engineer, who was the customer designer engineer who’s doing the simulation, use that simulation to make a test tape which was done automatically and then the agreement would be if this part passes this test, pay us. So this was the completion of the prototype. And if it didn’t work as simulated, then it became our problem at LSI Logic. We had all the equipment and all the knowledge to go through and find out what was wrong. And then we would fix it on our dime. Our competitors were not testing them fully. They were doing a few tests, packaging up some, send it to the customer, see if it works. And if it didn’t work, how is a customer going to figure out what’s wrong. I mean, it’s rather specialized and takes rather specialized knowledge to do this. The nice thing about our approach that one had to have this test tape ready to go - was that when we got the go-ahead, we were all ready for production. And we didn’t have to wait around and go to a different group and try to find the engineer that had worked on it and he was on a new project and get him work on a production test. So we bypassed all that by forcing that up front. Now the nice thing about it was that our competitors did not require this. They went and they did some half assed leakage test, sent it off to the customer and then if it went to go into production, there was this huge delay. And we were ready to go get the order. So we’d done a number where we got the prototypes out and we got an immediate production order and we were immediately in production which a lot of people want. And fortunately, our competitors in this business didn’t and don’t do this. So it gave us a huge competitive edge. Any rate, so at Fairchild and at LSI Logic, testers were the lynchpin of our design process. And it was the way to go. And yet, there’s no history there. So I wanted to change that. And Harold is the guy who’s been behind the thing from the very beginning. And so thank you Harold!

HV: Well thank you Rob. It’s an honor to - it’s great to see you again.