First-Hand:Beginning of the Silicon Age
Contributed by: Morris Tanenbaum, IEEE Life Fellow
For more on the life of Morris Tanenbaum see IEEE Oral History with Morris Tanenbaum
Discovery of "transistor effect"
The “transistor effect” was discovered in germanium by Bardeen, Brattain and Shockley in December 1947. The discovery ignited a rush to develop practical transistors and incorporate them into electronic circuits. Unfortunately, there were two limiting factors in the use of germanium transistors. The band gap of germanium (the energy gap between electrons that are bound to the Ge atoms and those that are free to travel throughout the crystal and carry electrical current) was only 0.7 electron volts (ev) and that limited the use of germanium transistors to environments only somewhat above room temperature and, therefore, also to relatively low power applications. In addition, the surface of germanium was chemically active and required hermetic enclosures of metal, ceramic or glass for stable operation which significantly increases the cost of germanium devices.
It was well known that silicon, which is found in the same column of the periodic table (Group IV) and has the same crystal structure as germanium, was also a semiconductor but had an energy gap of 1.1 ev. Since the temperature dependence of the electrical properties of semiconductors is an exponential function of the energy gap, silicon transistors could operate at significantly higher temperatures and power than germanium. However, silicon is much more chemically reactive than germanium and also melts at a much higher temperature. While silicon crystals could be grown by the same techniques as germanium crystals, it was much more difficult to purify. In addition, its chemical reactivity caused it to be coated with a very thin but impervious layer of silicon dioxide whenever it was exposed to the atmosphere. That made it more difficult to make good electrical contacts. Later, the presence of that thin oxide layer was found to be very useful in the manufacture and stability of silicon devices, especially integrated circuitry.
During the first few years after the discovery of the transistor effect, efforts were made at Bell Labs by Gerald Pearson to construct a silicon transistor. Although transistor-like effects were observed in these early structures, they were too weak to construct a device with positive power gain.
In 1953, Bill Shockley decided that it was time to make a more serious attempt to produce useful silicon transistors. I had arrived at Bell Labs in 1952 and had initiated a program to search for other semiconducting materials that might have useful properties. Welker at Siemens in Germany had discovered high values of the electron mobility in the Group III-V compound, polycrystalline indium antimonide (InSb). I grew the first single crystals of these Group III-V semiconductors, studied in detail and published the electronic properties of InSb and GaSb. It turned out that the crystal and electronic structure of these Group III-IV semiconductors were very similar to those of germanium and silicon. However the energy gaps of the new materials I studied were similar to or below that of germanium.
Initial efforts to make silicon transistors
After a year of III-V study, I was invited by Shockley to lead a small group to study silicon thoroughly and determine if it could be a useful transistor material. I was joined by Ernie Buehler, a crystal grower par excellance, and Leo Valdes, an electron device engineer, who would characterize any devices that we might make. We started by repeating Pearson’s work but with silicon produced by Dupont which was significantly purer than that which was available to Pearson. Buehler grew crystals by the Czocralski pulling technique that had been so successful with germanium. By adding a small amounts of N and P type impurities during crystal growth, Buehler produced NPN and PNP structures similar to those that worked so well with germanium. The results were similar to those of Pearson. While we could obtain transistor-like effects, the “alpha” of the structures were too low for useful power gain. (The alpha factor measures the fraction of the carriers (electrons or holes) that are injected into the middle layer and reach the collector region of a transistor.) That, in turn, is a function of the lifetime of injected carriers and the distance that they must travel. The lifetime of the carriers depends upon the chemical purity and crystal perfection of the semiconductor. Since we were using the purest silicon available and knew that the crystal perfection was high, the only remaining variable which we had was the distance of travel, i.e. the width of the center region of the transistor. However, with the best techniques that he could manage, Buehler was unable to grow center layers reproducibly thinner than one mil, i.e. one thousandth of an inch. While that was this more than adequate in state of the art germanium, it was not good enough with the best available silicon.
Using the rate growing technique
Buehler then suggested that we try to produce our structures by rate growing. That technique depends on the well known fact that the amount of a given impurity that is incorporated into a growing crystal depends upon the rate of crystal growth. Thus, with the proper amount of N and P type impurities in the melt from which the crystal is pulled, either N or P type regions would be produced. Since the rate of crystal growth is more easily controlled than the addition of impurities into the melt, it should be possible to produce structures with much thinner center layers. In addition, during the pulling of a given crystal, one could produce many transistor-like structures in a single crystal.
Using the rate growing technique, Buehler produced crystals that contained NPN structures with a P layer significantly thinner than one mil. The next challenge was making electrical contact to each of the three layers. That was easy to do with the much larger N type outer layers. I figured that I could make good contact to the very thin P layer by heating the structure and alloying a very thin wire of aluminum, a P-type impurity, to the thin P-type layer. The alloying was easy, but, when I examined the resulting structure, I discovered that the P-type layer had disappeared. Earlier, it had been observed that very lightly doped N-type silicon would turn P-type if heated to several hundred degrees Centigrade and rapidly cooled. It had also been observed that if the silicon was later annealed at a few hundred degrees Centigrade, it would return to N-type. (Later, Wolfgang Kaiser at Bell Labs discovered that this effect was caused by oxygen (probably introduced from the silica crucible used to contain the molten silicon during crystal growth) which acted as a donor impurity when dissolved in silicon. Upon heating at a few hundred degrees centigrade, the silicon precipitated into small particles within the crystal and was electrically neutral.) When I learned of this, I annealed my samples, the thin P-type layer reappeared and my NPN transistor was made. With the center P-type layer significantly thinner than one mil, we obtained high alpha and a very usable transistor. To the best of my knowledge, this was the worlds first silicon transistor.
Presenting our work and Texas Instruments
Shortly after the invention of the transistor, a group of transistor scientists and engineers had established an “off-the-record” semiconductor device conference held under the agency of the IRE (later to become part of the IEEE). The conference was held off-the-record to permit investigators to discuss their latest results without prejudicing their ability to obtain patents. I presented our results at the 1954 conference. After I had presented my paper, Gordon Teal rose and said that they had made a silicon transistor at Texas Instruments. I asked Gordon to tell us more about it. He said he was not authorized to do that. For a long time I wondered whose transistor was the first silicon transistor. Finally, in the book Crystal Fire, Riordan and Hoddeson reported that the Texas Instruments silicon transistor was first made in April 1954. We had obtained our results in January of that year. (In their book, Riordan and Hoddeson described the TI work as the first silicon transistor. They were unaware of our silicon transistor and had missed the Applied Physics article. I sent them copies of my lab notebook and, in a later article, Riordan acknowledged the precedence of our device.) While ours was the first, the excellent and independent work of Teal et al. also merits credit. Neither results were patented. Our Bell Labs patent lawyers pointed out that we had used well known techniques and the combination of those with the new material, silicon, was not patentable. Our results were published in the Journal of Applied Physics in early 1955. To my knowledge, the TI results were never published. Texas instruments did start production of their silicon transistor primarily for use by the military. We informed our device development people of our work but did not encourage manufacturing development because of the complexity required and the unknown processes that required annealing to regenerate the central P type region. In addition, I had begun work on an altogether different and more promising process for making a silicon transistor.
The Diffused Base Silicon Transistor
During this period, Calvin Fuller, a colleague in the Chemical Physics Department at Bell Labs, had been studying the solid state diffusion of donor and acceptor impurities in silicon at high temperatures. He discovered that acceptor impurities from Group III of the periodic table diffused more rapidly than donor impurities in Group V. By simultaneous diffusion of appropriate concentrations of both into an N-type silicon crystal, NPN structures were produced. Since the rate of the solid state diffusion process is exponentially dependent on temperature, the process could be easily controlled and very thin P-type layers could be made reproducibly. However, the resulting p-type middle layer was completely covered by the relatively highly doped diffused n-type layer and making electrical contact to it without electrically shorting it to the diffused N-type layer was a challenge.
My first approach was to polish a very small angle on the diffused specimen, thereby exposing the P-type layer while broadening it geometrically. I could then alloy an aluminum wire without contacting the heavily doped, diffused N-type layer. (At this time, Cal Fuller was preparing his samples using the floating zone technique without the need of a containing crucible and which did not show the N-type oxygen conversion effect that complicated the rate grown silicon material.) However, I could not obtain a high enough alpha to give power gain. I assumed that the polishing had disturbed the surface introducing crystal imperfections and decreasing carrier lifetime. I tried careful etching of the polished surface without success. After several weeks and many samples, various annealing experiments, etc., I was becoming very frustrated.
One evening, it was my wife’s turn to host her regular bridge game at our house. With her encouragement, I decided to get out of the way and drove back to my lab to do a little house cleaning in my lab. I had left an aborted sample in my micro-manipulator. I looked at it and said to myself that since I hadn’t been able to make any other method work, why not try the obvious way of simply melting the aluminum wire directly through the heavily doped N-type layer that covered it. I did not expect that to work because the N-type layer was so heavily doped that it was more metallic than semiconducting and would probably short circuit to the melted aluminum wire. Having nothing better to do, I brought the aluminum wire into contact with the outer N-type layer passed a pulse of electrical current through the wire alloying it to the sample. I then connected it to my oscilloscope expecting to see a short circuit but instead I observed the rectifying characteristic of a good transistor. I immediately grounded the aluminum base contact, varied the emitter current and measured an alpha of 0.98. Fantastic! Apparently, as the aluminum wire solidified, it created a PN junction with the heavily doped N-type layer permitting excellent transistor action. If I had tried the “obvious” method first, instead of assuming it couldn’t work, I would have saved myself several weeks of frustration!
I flew home–without getting a speeding ticket. Fortunately, the bridge game was over and I could tell Charlotte what had happened. I could hardly sleep that night. I needed to get back to the lab and make sure I still had a transistor and not a dream. It was still there! My colleague, Don Thomas, measured its frequency response at over 100 megahertz, higher than any germanium transistor then in production. (Charlie Lee, in the lab next door to me had very recently made the first germanium transistor using solid state diffusion by an elegant experimental technique and obtained frequency responses of over 400 megahertz but it never reached production because of the better fundamental properties of silicon.)
I called my supervisor, Morgan Sparks (who had made the first germanium junction transistor) and showed him the results. He soon had the top management of Bell Labs in my lab, with the exception of Jack Morton, the Vice President of Device Development, who was on a business trip in Europe. When the news reached Jack, he cancelled his trip and rushed back to Bell Labs. Soon, his division was devoted to silicon. In the interim, we published our work in the Bell System Technical Journal, and a symposium on silicon technology was held for all our semiconductor licensees. I continued to use our technology to make related devices, most importantly the pnpn switching diode in collaboration with John Moll, Jim Goldey and Nick Holonyak from the Device Development Division.
Staying with Bell Labs
It was during this period that Bill Shockley left Bell Labs to form Shockley Semiconductors. Bill called me and offered me a job at twice my present salary. I agonized for a week and finally decided that I was having too much fun at Bell Labs to leave. That could have been my first -- and only -- opportunity to be a billionaire had I gone with Bob Noyce and Gordon Moore when they left Shockley to form Fairchild Semiconductors and then Intel. I had met Bob Noyce earlier when he was being interviewed for a job with Shockley at Bell Labs which he accepted. Shockley later hired him to work at Shockley Semiconductors before he ever arrived at Bell Labs. I also met Gordon Moore whom Shockley had hired and sent to visit me and learn about our silicon technology. I’ve never regretted my decision to stay at Bell Labs.
I was soon promoted to become a Subdepartment Head and charged to establish a group to search for other interesting electronic solid state materials. Carl Frosch who had discovered the effect of silicon dioxide in masking the diffusion of donor and acceptor impurities into silicon, was part of my responsibility. That was to become a critical process in the invention and production of integrated circuitry. We also did some early work on the synthesis of the III-V compounds of GaP and GaAs. After a year or so, I moved over to the Metallurgy Research Department. That was the beginning of another story.