The Pure Stuff
Silicon is the stuff of the information revolution. It is the basic raw material of which are built the transistor and the integrated circuit and, indirectly, the computer and everything else made from microelectronic elements. Before any of those marvels of the age could be manufactured, there had to be elemental silicon; specifically, there had to be hyperpure silicon. Until the 1940s there was no way of obtaining hyperpure silicon in quantity; it was my good fortune to find the untrodden path that led to an answer. This is the story of the technology behind many technologies and the almost roundabout way that it developed.
Silicon, a gray, brittle solid, is the second most abundant element in the earth’s crust, after oxygen. But it always occurs in combined form, usually linked with oxygen, in rocks or silica sand. It can be liberated at high temperature by fusing sand with carbon in an electric furnace. Silicon that is 99 percent pure, easily produced that way, is quite adequate for metallurgical purposes and sells for well under a dollar a pound. But for electrical use, purity requirements are vastly different. Modern electronics depends on silicon with a purity to within one part in a billion.
The devices made of silicon today are the descendants of the rectifiers, or diodes, in early radios. Those were made of a crystal of lead ore contacted by a sharp wire known as a cat’s whisker, and they converted the high-frequency alternating current of a radio signal into direct current (the rectifier worked as a sort of turnstile for electron flow, allowing passage in one direction only). The cat’s whisker plus ore crystal was the original solid-state device. It was also hard to operate; success with a crystal radio set required tactile skill of a high order foi finding the sensitive spot on the mineral. The surface properties of each cleaved crystal were unique, so operators had to get to know the particular crystal they were using. With the development of rugged, inexpensive vacuum tubes that could do the same job much more dependably and economically and without manual dexterity, cat’s whisker rectifiers fell out of use.
Vacuum tubes had their own problems, however. The ideal rectifier has no moving parts, consumes no energy, and imposes no time delay on electron flow; a vacuum tube takes energy to heat its filament and a short but measurable time to move an electron stream between the elements enclosed in its glass envelope. And even when miniaturized, tubes take up valuable space and require forced cooling if arrayed in large number.
The need for a better alternative to the vacuum tube—and the resulting search for pure silicon—became urgent with the development of radar during World War II (see “The Road to Radar,” Invention & Technology , Spring 1987). Certain vacuum tubes were unusable at the high frequencies needed for radar, and their heat and bulk and weight would create serious problems anyway. The cat’s whisker-mineral junction had electrical characteristics well suited to radar circuitry, but it lacked ruggedness, exact reproducibility, and mechanical stability.
What was needed was a solid material displaying the electrical response of the old lead mineral without the accompanying shortcomings. Among the most promising candidates were the semiconductor elements germanium and silicon. Broadly speaking, a semiconductor is any solid whose electrical conductivity lies between that of a conductor and that of an insulator, providing nearly metallic conduction at high temperature and behaving as an insulator at low temperature. Semiconductors can be divided into the impure and the pure; the pure can be further divided into those that conduct through vacancies and those that conduct by free electrons. Silicon, when appropriately doped, can work either way.
If it was going to form the basis of an adequate rectifier, silicon was going to have to be highly pure. In 1940 pure silicon was not available at any price in quantities of more than a few milligrams. Faced with the demand for a reliable supply of this elusive element, chemists and metallurgists at the Massachusetts Institute of Technology, Bell Telephone Laboratories, and elsewhere undertook valiant efforts to purify the 99 percent product, either by leaching pulverized solid silicon with nitric and hydrochloric acid or by controlled, slow solidification from a melt, which provided segregation of the several impurities. It was found that repeated acid extraction followed by multiple recrystallizations could produce silicon from which diodes could be made, but the results were so erratic they satisfied no one. The search went on.
At the time, I was three hundred miles from MIT, at a modest Du Pont laboratory in Newport, Delaware, five miles south of Wilmington on the banks of the Christina River. There a group of chemists, physicists, and engineers were collaborating on the development of new or improved pigments for paints and other products. We were process-development people, seeking in particular better methods for producing titanium-dioxide white pigment.
By 1940 war on the high seas had jeopardized the continued availability of titanium-bearing ores from India, and production of the pigment was also threatened by a possible cutback in the supply of sulfuric acid, a prime reagent in the process but also in high demand for munitions manufacture. Our research management envisioned two possible ways around these problems: Either replace the sulfate process with a cyclic chloride process or, better yet, find a totally new pigment using raw materials abundant in the United States. None of this on the face of it had anything to do with silicon. None of us knew that research efforts to produce pure silicon were under way at other laboratories—or even that a need for pure silicon existed.
The chemistry of the proposed cyclic chloride process was straightforward: TiCl 4 + O 2 → TiO 2 + 2Cl 2 . The gaseous by-product, chlorine, would be recycled to make more TiCl 4 from titanium ores of lesser quality. On paper this new process looked like a good bet; in practice it needed some creative engineering. The reaction would take place readily only at flame temperatures of 1,000 degrees centigrade and above. Because both oxygen and chlorine are highly reactive at these temperatures, we would need extremely durable containment vessels, made of ceramics or fused quartz. As a result, I and one other chemist assigned to the project became selftaught quartz blowers. Glassblowing is a familiar operation to most chemists, but quartz fabrication requires the skill of a welder, and the chief tool for it is a hydrogen-oxygen torch.
Meanwhile, our other avenue of research, coming up with an altogether new white pigment, gave the Research Division a wide horizon along which to try all manner of unconventional combinations. Starting with a dozen or so elements available in the United States with which we could try to form binary or ternary compounds, the list of possibilities to explore grew very long. To be sure, many of the possible combinations were known and described in the literature, but the literature tended to be almost antiquated. Many of the references had been written in the nineteenth century. Had adequate steps been taken back then to ensure the purity of the starting materials? One thing we had learned in our experience with titanium dioxide as a pigment was that color-imparting impurities, such as trace elements like iron, had to be kept at an absolute minimum, a matter of parts per million, or the pigment would be badly discolored, just as a trace of free carbon can condemn an otherwise gem diamond to a life as an industrial abrasive.
To initiate a program of synthesizing binary or ternary compounds without an adequate supply of superpure starting materials would be folly. We began with a dozen elements, including such familiar names as calcium, zinc, boron, carbon, silicon, nitrogen, phosphorus, and sulfur. Most of them could be secured, adequately pure, from commercial sources. There was one startling exception—silicon. It seemed strange that the earth’s most abundant solid element couldn’t be obtained anywhere at better than 99 percent purity—which, after all, means ten thousand offending parts per million. If silicon was going to be a linchpin in the search for a new pigment, a first order of business would have to be to learn how to make it pure and not prohibitively expensive. Military necessity, diodes, and radar had nothing to do with it. We just wanted to be able to find a new pigment.
The first person to prepare silicon pure enough to allow even a first-order description of its physical and chemical properties was the Swedish chemist J’f6ns Jakob Berzelius, in 1824. His preferred procedure employed metallic potassium to react with potassium fluorosilicate, a solid white salt, according to the reaction 4K + K 2 SiF 6 → Si + 6KF. The reaction is energetic, the driving force allowing no give or take, and as a result, the silicon product emerges as finely divided granules mixed with residual salt.
Leaching the salt with water leaves the silicon difficult to consolidate, and this created problems. To complicate matters further, metallic potassium reacted with container materials like porcelain and quartz, forcing the chemist to resort to iron crucibles. And while iron is immune to attack from potassium, it does not fare well with molten salts at high temperature.
Containment schemes of various kinds were proposed in our lab, and I could easily digress here at length about whether any of them might have worked well. That became a side issue, however, for at about that time it occurred to me that a reaction between silicon chloride and zinc vapor in an all-quartz reactor had a chance of success. Some of my colleagues responded with substantial doubts about this proposal, mainly because it appeared that since the time of Berzelius’s first work, almost 120 years earlier, every avenue for making pure silicon seemed to have been explored, over and over again. In addition, the thermodynamics of the reaction looked unpromising. Zinc, a common metal of commerce, is a mild-mannered element readily melted in porcelain or quartzware and so unreactive that it can be used unprotected as the canister for dry batteries. As for silicon chloride, it was known to chemists as the volatile liquid used in World War I as the agent to form smoke screens.
The idea that no one had ever even tried the reaction between zinc and silicon chloride (2Zn + SiCl 4 → 2ZnCl 2 + Si) was hardly believable. A search of the chemical literature brought forth nothing; later a more comprehensive review by examiners in the U.S. Patent Office showed nothing in the prior art of U.S. or foreign patents. The equipment needed to make a valid test of the feasibility of this approach was disarmingly simple in concept (see diagram, page 62). Even the more elaborate large-scale units ultimately used for production would need no sophisticated engineering. And the boiling and melting points of the substances in the reaction were sufficiently far apart from one another to allow easy temperature control of the steps of the process.
We obtained “five nine” zinc (99.999 percent pure) in forty-five-pound slabs from a commercial producer and cut it into one-pound pieces for easy handling. These were melted in quartz crucibles, and the liquid zinc poured into a boiler through a heated trap. Vapor from the surface of the boiling pool of metal passed through quartz tubing into the reaction chamber. The silicon chloride was dispensed drop by drop as a liquid into a vaporizer and then passed through a quartz preheater on its way to the reaction chamber. To ensure a turbulent intermixing of the reactants, we put a pair of quartz plate baffles in the chamber. By-product zinc chloride and any excess silicon chloride would be vented through an exit tube and condensed in chambers beyond. Monitoring the flow of metal vapor at bright red heat is a formidable task; we avoided that problem by keeping the reactants in reasonable balance and noting any excessive residual liquid silicon chloride appearing from the water-cooled condenser. With experience, we found that a little extra silicon chloride was good, for under those conditions a larger and generally superior crystalline product was formed.
The very first trial of the zinc reduction process proved it to be a winner from the standpoint of purity over any other method known. To everyone’s surprise, the elemental silicon retrieved from the reactor comprised a loosely cohering, intertwined mass of acicular (needlelike) crystals some three-quarters of an inch long, with smooth, brilliant facets. As I recall, a laboratory assistant watching the scene instantly declared the experiment “thighthumping” and exclaimed, “Why didn’t 1 think of that?” As for the silicon’s quality, only very sensitive spectrographic methods could detect any impurities at all; the analyses came in at 0.001 percent impurity or less.
Among the first outside people to learn of our success with the zinc reduction was the physicist Frederick Seitz, who was at the time a professor at the University of Pennsylvania and who later became the longtime president of Rockefeller University. Because of his expertise in the field of solids and the interaction of radiant energy with crystals, Seitz had been retained as a consultant by Du Pont in its efforts better to understand the problems of pigments both white and colored. Light scattering by small particles (hiding power), photodegradation (chalking and flaking), and particle aggregation (settling) are but three of many physical phenomena always requiring attention in the development of pigments. Seitz was known among his peers for his knowledge of inorganic solids, metals, and alloys, and he had a wide acquaintanceship with physicists working in government-supported projects at the University of Pennsylvania, MIT, and elsewhere.
When Seitz saw our beautiful silicon crystals, he immediately reached for a telephone and called the Radiation Laboratory at MlT. Within one week Du Pont Pigments Research and MlT had an agreement in principle to supply silicon for use in diodes for radar. The efforts of the Radiation Laboratory became reliant on Du Font’s ability to supply the essential ingredient, silicon, for replacing inadequate or inappropriate vacuum tubes with solid-state diodes. And so three disparate elements came together to produce a viable technology: A need was provided, principally by the Radiation Laboratory; a technique was devised, independently, by us; and an appreciation of the unique product, allowing the need and the technique to find each other, was supplied by Seitz.
By late spring 1942 the demand for our pure silicon had so increased that a larger reactor was built, eight inches in diameter and six feet long, which could produce a hundred pounds a month. I must note that getting the pure product was a matter not only of a happy choice of reactants but also of maintaining an unbroken supply of containment vessels from start to finish. As the components required grew in size and weight, so did the challenges to those of us who had become homegrown quartz fabricators.
Technical people at the Radiation Laboratory selected several companies to make diodes from our silicon, including Westinghouse, Sylvania, and General Electric. Meanwhile, after Christmas 1942 I was on loan from Du Pont to join the Manhattan Project in its work on plutonium and did not return to Delaware until after the war was over.
When the war ended, so did the call for pure silicon. On V-J Day Du Pont had seventy pounds of pure silicon in inventory. Ten months then elapsed without a single request for any, and there were some inquiries about returning unused material for credit. After that, token sales were made to a few curious investigators. Du Pont Research exchanged some of the material for other substances from other laboratories.
Then, in 1947, the transistor was invented, by John Bardeen, Walter Brattain, and William Shockley (see “Sic Transit Transistor,” Invention & Technology , Summer 1986). The inventors used germanium as their semiconductor solid, and it was widely assumed that the electronics world would be vastly different from then on. In 1951, when Bell Telephone announced that it would license its patents and know-how, the Texas Instrument Company was one of the first to agree to the terms. By 1954, T.I. had learned how to make transistors that would sell for a price competitive with vacuum tubes and joined forces with a small but energetic radio manufacturer, Regency, to put the first transistor radio into the stores. It was a smashing hit. Soon IBM began to replace its vacuum-tube computers with transistorized models.
Though the germanium transistor and hand-held radios won over the public, the technical fraternity eagerly awaited a silicon transistor, which should tolerate higher operating temperatures. Even as T.I. was developing its germanium transistor, Dr. Willis Adcock, of T.I., ordered a supply of silicon from Du Pont and in a parallel research effort discovered the critical principles for a successful silicon transistor. That breakthrough was disclosed in May 1954, with the announcement that Texas Instruments was ready to go right into production with Du Pont supplying the material.
Demand for Du Pont’s hyperpure silicon began to increase steadily. By 1958 the company was turning out a thousand pounds a month, and production was moved to a new plant, in Brevard, North Carolina. Then, that year, another thunderclap struck: The integrated circuit on a silicon chip was invented.
As the use of the transistor had grown, so had the need for ever more complex circuitry. Standard components such as resistors, capacitors, diodes, and transistors needed to be connected in diverse ways to make useful circuits. Although each item could be mass-produced, wiring to make a complete assembly required time-consuming and expensive hand work. In the mid-fifties engineers were already planning the computers that would guide a rocket to the moon, but these would require up to ten million components. How could such an assembly be made affordable and compressed to fit into a rocket? A solution to this problem had become a numberone goal of researchers all over the world. The answer was found by two men: Essentially, Jack Kilby, at Texas Instruments, figured out how to put separate components on a single slice of germanium by selectively doping different areas on the chip, and Robert Noyce, of Fairchild Semiconductor, worked out a way of interconnecting the separate units. Similar procedures were used to make silicon chips.
Because successive doping operations followed by interconnecting steps subjected the silicon substrates to several heating cycles, it was imperative that the starting material be purer than ever. Du Font’s zinc process had served the industry well in the days of the diode and the discrete transistor, but now a better process would be needed. Customers were no longer happy with impurities in the parts-per-million range; now they wanted silicon with less than one impure part per billion. To serve this segment of the market, Du Pont established a new production line that depended on the decomposition of silane gas (SiH4) on the surface of a silicon rod heated to incandescence. Silicon is deposited on the rod and builds up there while hydrogen is released. Good quartz work was still essential.
Because the energy of many electronic signals—like that of an incoming radio or television signal—is remarkably weak, no deep layer of semiconductor material is usually needed to make a diode or transistor work. In fact, it can be less than a thousandth of an inch thick. Because of this, manufacturers soon found they could use a thin layer of very-high-purity silicon on a less pure foundation and both enhance quality and reduce costs. Thus the quality of silicon demanded by industry seemed to grow even faster than the quantity called for.
Forty-five years have passed since hyperpure silicon became readily available. Lately it has been challenged by semiconductor compounds (as opposed to single elements) whose properties in certain applications are more favorable. Just as the small time delay that electrons encountered in a vacuum tube became a tangible limitation, so has the mobility of electrons in the silicon lattice, in the race for ultrafast computer action. Electrons travel faster in some binary compounds, such as gallium arsenide, than in silicon. The race may not always go to the swift, but it tends to.
Du Pont enjoyed a market for silicon only until 1961. By then many customers, including the front-runner, Texas Instruments, had integrated backward to supply their own needs, and new companies had entered the competition. All the while the Pigments Department’s cyclic chloride process for making titanium dioxide had become increasingly attractive. It won the white pigment game and thereby laid claim to the lion’s share of the department’s investment dollars. Having no semiconductor products with which to integrate forward, Pigments withdrew from the silicon business in 1961.
It was fun while it lasted, and there is satisfaction in knowing that we scored a first in finding a process to make silicon really pure, first for radar and then for transistors. The spirit of J’f6ns Jakob Berzelius might smile, knowing that the element he described in 1824 was first produced pure, in quantity, on the banks of the Christina River just five miles upstream from the Rocks, where in 1638 the first Swedish colonists landed in America.