SEMI Oral History Interview
John Robinson & Mike Grimes
Texas Instruments (retired)
Interviewed by Craig Addison, SEMI
John Robinson joined Texas Instruments in 1973 as a development engineer after receiving a PhD in materials science. He later became operations manager for MOS production and from 1981 to 1995 served as manager of TI’s silicon operation. Mike Grimes joined TI in 1969 in the materials group after receiving a Master’s degree in physical chemistry. He progressed to engineering manager in the silicon operation in the mid 70s and became manufacturing engineering manager during the 1980s. Robinson and Grimes both retired from TI in 2000.
Select a question to view the response.
JR: TI was my first job out of academia. I received a PhD in materials science at Argon National Lab as well as Cornell University. After a post doctorate I decided I wanted to go into the industrial world and joined TI in 1973 in a wafer fab operation and I worked in the wafer fab for about three years as an engineer and had an opportunity to move into applied research in the materials organization of TI and did so in 1976. That’s how I got into the semiconductor industry.
MG: Mine was similar. Got a Masters Degree in physical chemistry and as soon as I started looking for work I was very interested in electronics and of course had heard of TI. This was in 1969 and it was an interesting area and I also knew that TI had such a good reputation in the area I decided I would interview with TI and consequently I went to work in the materials group. First of all I was working with crystal growth as an engineer and we were doing some development work and later went into the silicon wafering part of the operation and then in subsequent years worked in all the aspects of silicon materials.
MG: TI when they acquired the transistor patent rights from Bell Labs started playing with growing crystals to fabricate the transistors. My understanding is that that was in 1952. That was germanium crystals. Later they saw the need to go to silicon and they had a very aggressive program to produce the transistor market as well as the technology. This was all done on Lemmon Avenue here in Dallas. They started out with germanium and did grown junction and developed their own crystal growers and were really ahead of the market and ahead of the technology industry-wide at that point. Then they decided to try to develop a market so they designed a transistor radio with IDEA [Industrial Development Engineering Associates] and the Regency radio was born. But it used germanium transistors. That was about 1955. It wasn’t a raging success but it certainly got everybody’s attention and started the ball rolling.
JR: In the early days TI had to develop their own infrastructure for materials, everything from silicon to photomasks simply because there was no industry infrastructure available external to the companies who were starting in the business. So TI had a major effort in research and development and applied engineering on the technologies associated with producing materials, starting with polysilicon and ending up in wafer processing. Simply because it was not available anywhere and if you wanted business to grow you had to do it yourself rather than wait for some other company entrepreneur to come in and take hold of those technologies. And that was all pretty much in place by the time I entered the materials operation.
MG: Likewise it was when I went to work as well. From what I understand of the history, they started developing the polysilicon technology to make purified silicon. You could purify germanium fairly easily but silicon was another animal and they wanted to go in the direction of silicon because of the fact that it worked at higher temperatures and higher frequencies. So they thought that was the way to go from the start and they developed their own methods of producing polysilicon to single crystal and to growing crystals and wafering…manufacturing the transistors. This was all in the late 50s, early 60s.
JR: By the time I entered TI was growing single crystals by the Czochralski method and was also producing float zone material for different applications. I don’t even know if anybody produces float zone material anymore. Those were the two materials, for ultra high purity applications, and maybe higher voltage applications for power transistors. Then the float zone material was applied to that for those products.
MG: Well, Mark Shepherd [who later became TI’s CEO] actually grew crystals from my understanding.
JR: In fact, that was where a significant number of leaders of TI came through -- from the materials organization. One of the founders of a lot of technology in TI was a guy named Fred Oshner who later became a manager in the research department.
MG: They hired a guy by the name of Gordon Teal. He came from Bell Labs as an engineer, and he had experience in crystal growth so he helped develop it at TI as well. A lot of the early technology came from Bell Labs, from the patent rights that they bought.
MG: Siemens developed a method of generating high purity polysilicon. They had a technique for doing that. They developed it and licensed it to several people and I think they did in fact license it to Monsanto.
JR: In the later years, in the 70s, Jack Kilby was involved in some of the materials, not directly in the wafer section of silicon materials but in some of the more exotic applications of silicon materials. TI developed a very significant technology in polycrystalline production after licensing the Siemens method for generating high purity polycrystal silicon. And the technology was quite novel and produced granular silicon. That technology was applied for some number of years internally and later was licensed to Hemlock Semiconductor. Also a similar product, using a different process, was developed by Ethyl Corporation in the 80s for generating polysilicon. But TI was integrated from polysilicon all the way through to polished wafers for wafer fab applications, including epi. Epitaxial layers were put in single crystals in many cases because the science in understanding of the substrates may have been there but the ability to execute a commercial process to generate some of the purity required wasn’t always there. So epitaxial layers were used in the very beginning, especially in the pre-integrated circuit days for the use and application of npn [and] pnp transistors and diodes.
MG: There were a number of challenges (laughs). One was just getting crystal yields to a point that it was profitable. And trying to maintain the integrity of the crystals in such a way that they did not effect the devices that were being built at that time. This was in about 1969. Of course the purity of the polysilicon wasn’t tremendously understood because the tools necessary to measure the purity were not in place. So you never really knew exactly what you were dealing with but you knew at times that there were impurity influences going on that were affecting the device yields. But you may not know exactly what they were.
There were things like carbon and oxygen which were just beginning to be measured. Then there were a lot of mechanical issues too. How to make a wafer flat so you could print on it was not well understood. There was a lot [that was known] from the glass industry but it didn’t really seem to apply well to silicon, it was a different animal. And how you etch material…understanding all the properties of silicon in terms of its structure…what was the best structure to build on, the orientation…the cleanliness wasn’t well understood. The rooms that we worked in were dirty compared to today. I mean they were filthy compared to today (laughs). I remember we worked in a basement for a long time to try to polish wafers and a lot of times it would rain cement out of the ceiling onto the polishing equipment. We didn’t worry too much about it, as long as it was shiny (laughs). But as the competition built the need for higher yields in both the materials and devices became a real pressure point to improve quality. For a while it was just get the product out, but after a while it had to have a certain amount of quality and the quality became the major issue and major push.
John mentioned float zone which was growing on a pedestal. It had certain purities that couldn’t be achieved in Czochralski type crystal growing. But always Czochralski was the major method of growing crystals. We started out in ’69 [when] the largest diameter was an inch and a half. Although we did grow crystals down to three quarters of an inch, half inch crystals for some applications. In about 1972 or so two inch became quite prevalent, and the next largest diameter was 100 mm in about 1975 or so.
JR: When I joined the materials department from the wafer fab I had a fairly unique perspective because we in the wafer fab would always blame all of our device failures on the materials people (laughs). I would go over to see the engineers in the materials department knowing that I had yield problems because of materials and walk away convinced that there were never any materials problems whatsoever. So when I joined materials, I joined it in an unusual project which was to grow single crystal silicon continuously. By that time TI had developed granular polysilicon which then afforded the potential opportunity, I would say, to feed polysilicon into a melt and pull a crystal simultaneously and if one was successful you could model a condition where that would be a lower cost silicon material. And we were successful in developing to some level a crystal puller that could grow two inch ingots in a semi-continuous manner but what became apparent in that project was that it really was not a cost effective and practical solution to low cost silicon. Simply because the dynamics of crystal growth failure accumulated over time and consequently you really weren’t able to grow continuously because you’d have interruptions due to crystal failure, associated with a myriad of different things, one of which was that the process of concentrating carbon in the melt would subsequently generate failures in the crystal at some point.
And that project was also integrated with a JPL [Jet Propulsion Lab] project where TI had to grow low cost Czochralski crystals because at that time, which was 1976, we’d just experienced the oil embargo from the Middle East. Oil prices went through the ceiling and the government was really pushing a big initiative to have alternative energy sources. We all know what happened to that. That project really lasted for about a year and then it became obvious that the funding for that was going to have to far exceed what the company thought it would take to really be successful on a commercial scale, so I moved into an applied engineering role in understanding the dynamics of oxygen and carbon and crystals and subsequently into the devices. The focal point for all this work was to understand how to make high yields in dynamic RAM products which then became the driver for technology for silicon for the next several decades. As Mike said, the challenges were getting the attributes necessary to meet the demands of the wafer fab in a cost effective manner.
MG: I might add a couple of things. From the beginning TI built their own crystal growing equipment up until about 1972. Most of it was RF heated, using radio frequency to heat the graphite. And then they built their own float zone equipment, all of which was designed and built internally. Then in 1972 they got away from that and went to commercial suppliers of equipment because there was such a pressure on equipment for building devices. As it turned out we helped develop a lot of…worked with a lot of these suppliers and helped develop them, like Kayex, Hampco and Varian back in those days.
I might add another thing. When I first went to work the prevalent thought about single crystal silicon, was that if it was single crystal, it was perfect, because it wouldn’t be single if it wasn’t perfect. And through the 1970s single crystal silicon was studied so well that that was not the case and there where all kinds of factors that determined the quality of the crystal and there was actually, I’d say, not an evolution, but a step function improvement in crystal through the 70s both in purity and properties that were able to be produced.
JR: The defect structure was the major revelation and the effect of the defect structure on devices. But there was just an overwhelming amount of knowledge generated, as Mike said, all centered from carbon, oxygen and intrinsic defects.
MG: And it was just not understood in the late 60s and early 70s. But there were a lot of measurement tools developed in that period that were really helpful to the materials technologist to then understand how to make better silicon crystals. I think this had a big impact on the yield of the devices, especially as devices became more complicated.
JR: By the 70s there was an external infrastructure for crystal growing, Varian was one, Siltec designed their own crystal pullers for their own application and would sell those. Kayex Hampco, which I think came from a hamburger processing-equipment company, produced crystal pullers. So the evolution of the diameters required fully redesigned equipment.
MG: Leybold Heraeus was also building some equipment in Europe at that time. They were filling the European need basically. All of these companies were building equipment that was good, but I think they worked with their customers in such a way that it made them build better equipment and a lot of development was done after they sold a lot of the equipment on the floor.
JR: It’s one thing…to run a crystal in the lab. It will work in the lab, but you take it out and try to replicate it and have very high utilization rates and low failures, that’s a different matter.
CA: Let’s talk about the wafer diameter changes. What challenges did that bring?
MG: There was always pressure on the next diameter. The perception was the larger the diameter the more devices you could build on a wafer and that would cut down on cost and increase yield as well, because a lot of devices around the edge of the wafer were defective. So there’s always pressure to develop bigger and bigger crystals and wafers from the fabs. But at the same time they wanted better and better quality in the crystal in the wafers themselves.
JR: It was like a relay race. One problem is maybe defects in crystals. So you’d solve that on, say 4 inch, or 100 mm diameter crystals… then new geometries come, so the flatness requirements change, so you have to have a flatter wafer. Then a different printing process comes along where you print bigger, larger areas. And the steppers, as they change, drive flatness requirements for the silicon wafer producers. Generally the initial problem was related to the bulk properties, oxygen and defects but then…Mike and his engineering people were able to get really ahead of those kinds of problems. But the flatness problems…they would come in and almost obsolete a whole set of equipment for polishing, grinding, lapping because it was extremely difficult to meet those demands for larger diameter flatness control from the existing equipment.
MG: When we had 2 inch, most of the photomasks were like contact printing and it was pretty forgiving. If you had a wafer that was not particularly flat you could actually bend the wafer and make it conform to what you were looking for. Later with photolithography, that was a different ball game. And of course with vacuum chucks and this sort of thing they were able to use wafers that maybe were a little more forgiving. But when you are talking about smaller diameters it wasn’t such a big issue. But as you got into larger diameters it became a big issue …ways to measure, how do you measure because everyone had a little different take on how to measure flatness.
JR: Initial flatness measurements were absolutely archaic where you’d use glass plates and use interference fringes to measure flatness and actually have to visually count these things. It was unbelievable. The first diameters that really accentuated the flatness problem was the later stages of the 100 mm and it was driven by DRAMs and the geometries at that time, 4K and 16K DRAMs were the initial type products produced.
MG: John mentioned the equipment. What was interesting was that when you went to larger diameters you were in a whole new development phase…and especially Czochralski crystal, you would have to have larger geometries for everything. Your heaters, your graphite, your crucible or quartz, the equipment itself, the power requirement, the amount of polysilicon you were utilizing. All of these were requirements to get a reasonable yield and all had to be developed…a lot of time with people who were suppliers who weren’t particularly interested in your problem. They made products for something else. I know in the early days the graphite sources mostly made graphite for the steel industry. And so it was not particularly pure and they didn’t particularly care if you liked what you were getting from them or not. And so they’d say this is what it is, and if you like it fine, if you don’t want it, that’s tough (laughs). So you’d have to go in and develop with some of these suppliers and it was a continuous battle to develop with them your requirements for the silicon industry. Eventually they came around and there were some specialized groups that supplied the silicon industry and did a great job eventually. But the pressure had to really come from the people who were consuming these products, not from the supplier. He was not very interested at first.
But you were fighting that group…you were working with larger diameters. Then you had these other requirements imposed on you and all at the same time the wafer fab, the people manufacturing devices, were asking for larger and larger diameters. They were always looking forward to the next generation of diameter of equipment. So then you had to have a team as well going out and looking at potential equipment…and the capital that you would [need] to supply their needs. So it was really a chess game back in those days, especially in the late 70s and of course throughout the 80s.
MG: They really didn’t want to build a new fab on the current diameter. When they built a fab they would want the next generation.
JR: It had to do with what the engineering capability internally was in terms of building fab equipment. At one time TI built all their own semi-automated wafer fab equipment and once there was a capability both internally and externally for a larger diameter. Then as the demand -- either driven by capacity or demand for existing product or development of a new product -- came about then the diameter would shift upward and that process dribbled down through the silicon wafer manufacturers and all the suppliers in the supply chain.
JR: Very little. Well, really, almost no sway other than we could discuss with them the material characteristics that you might expect from a diameter increase but by the time that was all going on the industry had developed enough that there were multiple producers of silicon wafers by now so if you weren’t going to produce the next diameter for them they would go to someone else.
MG: We didn’t operate in a vacuum either since TI was so vertically integrated. We knew what they were up to. They would call us in very early. Also we would supply them with prototypes so they could do some testing on the equipment. So it was a hand in glove relationship. We knew what was coming but a lot of times it was a matter of resources and what fire you put out next.
JR: The device producers had an economic model. They were all generated by marketing, for what they could sell the chips for and the diameter drove the cost. And they rolled that into cost calculations using dollars per square inch of silicon to determine the economics. In every case though, as you raised the diameter of silicon wafers, the dollars per square inch went up instead of down because there was no opportunity to engineer-out [the cost] through shrink geometries. Instead of building a 4 inch wafer on equipment and then being able to build 6-inch wafers on the same equipment you had to replace all the equipment because it’s not compatible, whereas a wafer fab could drop their cost by shrink geometries on a given diameter.
MG: What was going up exponentially during that period was not only diameter, but the requirements. They would not only want 100 mm, go from a 2 inch to a 100 mm, to 150 mm, they also wanted the wafer to be flatter, cleaner, have better crystal characteristics and all this. And they put pressure on us again back to our suppliers. A lot of the time the chemicals we used were very impure to say the least. And so then we would have to put pressure on them to develop ways to get a lot of the impurities out of their product and a lot of times they were very resistant to that. Some of this was bought outside of TI, some was bought inside TI. And of course the more pure the consumables usually translates into cost. And a lot of times they did not want to pass any of these costs along to the device manufacturer. It was so much per square inch at 100 mm then it should be the same for 150 mm. Never mind the fact that it was flatter, cleaner, had better crystal characteristics and was more difficult to produce.
JR: When you asked earlier about our role with the customers in driving those decisions for diameter expansion, the information we gave them hopefully was good information about the cost structure they would have on silicon materials whenever they scaled the diameters up. So rather than using linear scaling or just using flat scaling models from one diameter to another the scaling went up as well. That helped them, but it didn’t deter them because the payoff in the end was in the chip side of the business.
MG: There was always various takes on, for instance, diameter…accepting what was a standard. There were very few standards in the early days. For instance, IBM, who at that time was a customer of ours -- we sold a lot of material outside of TI -- was always a quarter inch over the standard. They were one and a quarter, two and a quarter, three and a quarter diameter. Well, you had to tool up especially for them if you made any product for them. And then just subtle differences in diameter would sometimes cause you to scrap wafers that otherwise were perfectly good wafers but maybe they wouldn’t fit some printing equipment or some of the other tooling that was used in the wafer fab.
JR: My perception is that there was not industry collaboration on diameter. It was “who can do what when” to get the lowest cost chip, rather than a collaboration.
JR: I think it happened in the 200 mm to 12 inch transition. I don’t think it was before that.
MG: SEMI was a big player in trying to help establish [standards], but for some reason there was still a lot of resistance across the industry. It was like, “this is what I want and you give it to me” kind of thing. But we did see some standardization attempts on 100 mm, but it seemed like all of our customers had a different take on it. If you just take the diameter, was it plus or minus 10 mils or was it plus or minus 5 mils, plus or minus 15 mils (laughs). Usually the tighter the better but there were several attempts across the industry to try to get it standardized but it seemed like wafer fabs were just never really real interested in doing that. Eventually it started coming about though, in the late 70s, probably with the 200 mm. Because the equipment to measure flatness had to have certain tolerances as well. You couldn’t just take any diameter into (the equipment).
JR: And also because the end user of silicon wafers had a SEMI standard to go by as well that was developed by the standard committee. Since that started occurring there was a standardization but there were also always exceptions to those standardizations. If one of the producers didn’t like this specification, it didn’t matter whether it was there or not, they would write [their own]. The 200 mm was probably the stage where the wafer had a standard. It had a standard thickness-uniformity, there was a standard measurement, it had a standard for radius of the edge, some standards for flatness.
MG: I remember the 100 mm had “standards” (in quotes) but if you looked at it closely it was like 10 different standards. It was not like a standard.
JR: It was what the engineer in those different users preferred so that was what generated those specifications.
JR: I started out as I said as a development engineer and had a small development group in the silicon operation. After about two years of doing that I became operations manager with the silicon department…the division manager divided the silicon producing entity up into two separate entities, one to produce wafers for MOS customers and one for bipolar customers. And I was operations manager for MOS production for about three years until they were recombined and from 1981 through ‘95 I was the manager of the silicon operation.
JR: Just about everything. Diameter control was one. Flatness was another. The defect characteristics, oxygen characteristics were probably the other major factor at that time. Later it evolved to particulates and surface metals, as Mike has mentioned…[but] in the very beginning it was flatness and crystal defect uniformity.
John: No, it was recombined for economic reasons because the other way was just completely wasteful management. It was a noble experiment but it required two engineering managers, two planning managers, two finance managers and all this so putting them back together made more sense than having dual costs.
MG: I started out as a line engineer and kind of worked my way through various departments, all of silicon crystal growth. We call it wafer shaping, sawing and lapping and polishing…we did some epitaxial work, we called it advanced products, and some radiation hardened devices at that time. We did some preliminary steps, it wasn’t the finished product. So I kind of worked my way through all of those entities and became engineering manager in the mid 70s and stayed in engineering most of my career. Did some development work…although a lot of times we did development on the fly, we did break out some of that but a lot of times it ended up being part of production. Later I ended up in both manufacturing and engineering. I was manufacturing engineering manager for a number of years in the late 80s, then we divided up, before MEMC came on the scene, we had just the TI facility and then later I became manager of that facility.
CA: Before we move on to the 80s, anything about the 70s that we missed?
JR: There was one thing, and this had a big influence on TI internal silicon and that was the periodic shortages of polysilicon. And those periodic shortages further drove the wheel of upper management at TI to become totally self-sufficient in polysilicon because we’d had bad experiences in the 70s from polysilicon shortages. The shortages also created a huge growth in polysilicon capacity around the world…in the 80s. So it was feast or famine sometimes in this business and the famine always drives excess in capacity. So in the early 80s there was a huge excess of polysilicon capacity that took some time before it was utilized.
MG: John’s right. There was a big scare of supply of polysilicon and that’s one reason why TI got involved in granular polysilicon and in fact built a plant during the 70s to produce the material as an alternative to the Siemens type process of producing polysilicon. It turned out to be a lot more difficult to grow crystal from it that we anticipated although I think the technology has come a long way since then and it’s pretty well accepted as a source of polysilicon material. But also trying to get the cost down was a big factor because [in terms of] a material going in to the devices, silicon is one of the largest cost factors in producing a device. And still is. As John said, there was a lot of emphasis on energy too and you could produce granular polysilicon with less energy so there was a real push to make that happen and they were willing to invest a lot of money in it.
JR: There were some changes. There were some products [that] we internally were not as good [as manufacturing] as other producers and we would have to play catch up…part of the products were purchased externally. A lot of the Japanese wafer fabs [owned by] TI initially started buying all of their silicon from Japanese suppliers so that was a factor as well because a lot of the growth in silicon consumption at TI during the 80s was through the expansion of fabs in Japan. And so that drove a shift in strategy from our internal silicon operation and led to a focusing of our energies internally to focus on expanding into the Japanese market. Always during the 80s there was a battle with other suppliers because here was the silicon business for TI [which] was a rich plum for anybody to have.
MG: One thing, we started getting a lot of competition, as John said, from Asia for supplying polysilicon as opposed to producing it internally. So it was an economic issue that had to be addressed. Eventually, we had a Siemens-type polysilicon plant and we eventually closed it and then we were using our granulated polysilicon material but we eventually shut it down as well in the 80s, for a lot of reasons. And we eventually got completely out of the polysilicon business because we could actually get higher quality pure polysilicon at a lower cost by buying in externally. Of course that’s the kiss of death if you are an internal supplier.
JR: And by the way there was a huge glut in supply. You could name your own price for polysilicon.
MG: The same companies that built polysilicon plants also started making wafers so it became a huge economic issue around producing silicon wafers because they needed to sell their product. If they couldn’t sell the polysilicon they weren’t [going to] sell the wafers so they put tremendous pressure on internal suppliers, people like Motorola who had an internal supply of silicon. A great deal of pressure was put on them to get out of the business, Delco was another one. I’m sure Western Electric had an internal supply as well. They eventually went out of the [silicon] business. They could get higher quality, lower cost wafers from competitors so there was a huge swing during the 80s.
JR: There also was a transition…the capacity for polysilicon at TI, let’s talk the Siemens process, was old capacity. I mean in the mid 80s it was probably 20 years old. By the time a company like Hemlock who’d use new technology expands their plant, then the old technology can’t keep up economically, or the quality can’t keep up either, because all the knowledge is integrated into that new facility. So that was a big factor in the Siemens side of polysilicon. The granular poly was a similar situation although the economics of scale came into play since basically a polysilicon refinery is a big, huge chemical plant. And economics of scale are extremely important. So internally…I remember 50 metric tons a year was our capacity, or 55 metric tons. And our consumption was maybe 60 metric tons. But Hemlock’s capacity was 1,500 metric tons and so the economies of scale, in effect, priced our internal capability out of the market and at the same time the demand for wafer prices were going down then the only way we could do it was to take advantage of the cost reduction through polysilicon.
JR: I’ll give you an example. The TI chambers were maybe three feet in diameter for making Siemens poly. The chambers in some of these new plants are like 30 and 40 feet in diameter so you get huge scaling factors that save both power and labor.
MG: What we are finding today is there is a marriage between the two types of processes to make Czochralski crystal because it’s a perfect marriage in that you can put the chunks of the polysilicon from the Siemens process… then you can fill in with the granular material and that way you don’t have to have such a large crucible to melt down and to grow the crystal.
JR: There was a transition where…in the 80s we had a major initiative to penetrate the Japanese market. Part of it was economics and part of it was pride as well. So we were very successful in doing that. We at one time supplied well over 50 percent of the silicon to TI’s Japanese operations. The problem was the economies of scale. We were not necessarily large enough to have the back up R&D that other producers had simply because it was their major business. They were silicon suppliers, they weren’t chip makers. TI was a chip maker that had its own internal supply of silicon. It was a huge difference. The primary business for TI was chips, not silicon. As we were able to keep up with technology internally with the resources we had it was a huge advantage for TI to have an internal producer because we were extremely cost effective. We had the best customer service of anybody, we had the best cycle times. We did a great job in serving the customers cost wise as well as all other service measures. So through the 80s that was not really a problem. In fact our business grew because we invested in an expitaxial operation and that made a great deal of difference to us economically and it still does for silicon producers. And so through the 80s was really an expansion. I’m trying to think of our revenue numbers, but they tripled. From the time when I first started managing that business they went up by 3X in the 80s; wafer sales and epi. We expanded significantly during that time. And we in fact expanded to the limit of the capacity of the buildings we had.
JR: There were some MOS devices that started up on epitaxial wafers but that soon died simply because DRAM economics wouldn’t allow you to use epitaxial wafers. But later generations of MOS devices were made using epitaxial wafers and I think the way TI designed the chip, epitaxial wafers made the chip surface more economical. In other words you got more chips per square inch so that drove the epitaxial wafer business.
MG: But we were always in the epitaxial business as a materials supplier. Even in the early days back in the 60s we had what we called pancake reactors. We’d put expitaxial layers on silicon mainly for our silicon small signal devices but that technology was not good enough for later generations. In the late 80s we thought DRAMs were going to have to be built on epi. I think everybody thought that, so in the early 90s there was a huge ramp up in epitaxial capacity across the industry. Everyone thought the latest generation of DRAMs with the small geometry they had and the quality they were going to require was going to be requiring epitaxial wafers. And as it turned out that was not the case. And they also thought the yields were going to be low…so both were going to drive huge demand for epitaxial wafers. But it never really came about. The wafer fabs found ways around all that.
JR: They might have started out with epitaxial wafers but after the fab got started they started designing out the epi because of the price difference between epi and polished wafers.
JR: The expansion occurred really up through the late 90s in one form or another. The wafer expansion stopped because the building was full. We could not add more crystal pullers or anything or wafer processing. But I think the big transition was whenever the 200 mm product lines came along. We needed more space because all of 125 and 150 mm wafers filled up our existing operation and we needed significant change in space to put in 200 mm capacity and…at that time the company probably felt that…there is only so much capital we have and if we expand wafering then that consumes capital we might spend on chip manufacturing.
MG: And it was going to be significant capital for a building and equipment that was going to be required and then that was also being pushed by the 200 mm and the larger diameters because all the equipment we had was not going to accommodate [larger diameters].
JR: We could make almost trivial amounts of 200 mm wafers, enough for samples, but not anything for commercial use. So 200 mm expansion now required a whole new building, a whole new factory and that was probably the driver of it. TI decided that rather than expand that way to go out of the silicon wafer business through a joint venture.
MG: TI was really reluctant to get totally out of the silicon materials business but it was a real capital issue.
JR: A capital issue more than anything else because internally we were the low cost producer and we priced our products against the open market and we were the low cost producer as well as the lowest price to our customers and so had a very nice profit margin and reduced prices along the way each year as well.
JR: IBM and a few were our last external customers. That stopped in the late 70s because the internal demand required our full capacity. So there was a phase out of external sales of polished wafers and epitaxial wafers.
MG: When I first came to TI we probably had 50 or 60 external customers all over the industry.
JR: There was competition…here are TI’s competitors in chips and they are buying wafers internally and competing for capacity which was scarce for wafers. After a while it doesn’t look right, it doesn’t make sense to do that.
MG: Good management (laughs).
JR: As it turns out we had some of the best negotiating cost structure anybody could imagine for polysilicon supply, for crucibles, graphite. We had outstanding cost structure on our raw materials. We had extremely good yield performance and we also managed with less people. So you take all those together and we had cost structure that was outstanding.
MG: By any comparison we had fewer engineers per billing dollar than any other manufacturer that was standalone. Of course we had the TI overhead to deal with…we were not given any advantage by being part of TI, but we really had a good team and we really had some good cost structures as John mentioned.
JR: We did have advantages but they were advantages that at the end were hard to see because we were treated basically as any supplier in bidding for products.
JR: I don’t know who stimulated the first discussions of it, but the final decision was made at the top level, the CEO of the company.
CA: Then you started looking for a buyer or a partner?
JR: What actually happened, we designed a big expansion project and we needed several hundred million dollars of capital and I think once the CEO of the company saw me present that two or three times to him he finally said, “We don’t have that money but we need 200 mm silicon wafers,” so the logical way to get that is to form a joint venture with a company that has expertise in 200 mm.
CA: And that company was called?
JR: That company was called MEMC Southwest. It was started up in 1997. The majority owner was MEMC.
MG: And during that period was the ramp up to 200 mm so as wafer fabs were brought on for 200 mm their requirements for the smaller diameters started dropping off so that it either had to be supplied by the joint venture or from an outside source. It turned out it went both ways. It had to be brought on quickly to keep up with TI’s requirements.
JR: In 1995 there was chaos in the silicon semiconductor industry in terms of volume demands. ‘94 was the planning year for this joint venture and things were just going through the ceiling in terms of demand and projected demand and the same in ‘95 and ‘96 and then about the time we got the plant completed, in ‘97, ‘98, the market fell out of the semiconductor business.
MG: They were projecting an almost exponential growth.
JR: The exact numbers were the industry had doubled in the three years prior to 1995 and the forecast was to double again by the year 2000. Somewhere in there that didn’t happen and about the time we completed the 200 mm factory and we are starting it up the demand had evaporated for 200 mm and was evaporating for smaller diameters as well.
CA: What impact did that have on the operation?
JR: The decision to keep the plant open was there and that remained a decision that was driven by the long term and you suffer through.
JR: That was part of the joint venture. That operation grew nicely through ‘97, ‘98. When the downturn hit in the late 90s that operation suffered as well.
MG: And slowly TI disengaged from the joint venture in that they allowed it to be operated as a separate entity and we ended up having to compete very succinctly with the outside market, which is understandable.
John Robinson and Mike Grimes were interviewed November 18, 2004 by Craig Addison of SEMI.