MOSFETs: a hopeless quest

by George Rostky

There's an old ditty that goes something like: The experts said it couldn't be done. And to look at the job, who wouldn't? But I tried the job that couldn't be done. And what do you know? It couldn't.

Well, Steven Hofstein and Frederic Heiman tried and, with the help of perseverance, luck and an encounter with Andy Grove in a Las Vegas swimming pool, they tried the job that couldn't be done and created the first practical MOS field-effect transistor without whose descendants we would not have today's ICs.

It all started in 1959, the year when Texas Instruments announced that Jack Kilby had invented the first integrated circuit, which TI called a "solid circuit," the ultimate in micro-miniaturization. In that year, Hofstein, with his brand-new BSEE from Cooper Union, and Heiman, with his brand-new BSEE from City College of New York, joined RCA's David Sarnoff Research Center as part of a work/study program. (The study part, toward an MSEE and PhD for each of them, was at Princeton.)

In time, they shifted to the Electron Devices Research Laboratory, where the group leader, Tom Stanley, offered a choice. "There is a lot of work going on to integrate transistors into a network on a chip," Stanley said. "Everyone is working on the junction field-effect transistor that Shockley invented in 1952. So if you accomplish anything with the junction FET, you will end up working as a technician for some senior engineer. But if you want to work on the MOS transistor, a hot potato that has killed everybody who tried it nobody can make it work the situation is different. Many people said that the MOS transistor would never work. There were problems with surface states, a problem Bardeen had uncovered at Bell Labs. If you succeed, you'll be a hero. If you fail, so what? You'll just go to work on a different project. What can you lose?"

Why a MOS transistor? Bipolar transistors, used in ICs in the early Sixties, were very complex devices. When you made a wafer with bipolar transistors, half of them didn't work. The bipolar transistor was an unpredictable device, so its fabrication yield was poor. If you tried to build a large array containing more than six or eight transistors into an integrated circuit, you were guaranteed failures. 100 transistors? A hopeless dream.

In contrast, the MOS transistor, whose physics was described by Julius Lilienfeld as early as 1925, was ultimately simple²in theory. The only problem was that nobody could make it. If you could learn how to make MOS transistors, theoretically producible, and reliably producible, you might indeed be able to build ICs with 100 transistors, maybe even 1,000. Wild dream.

The problem lay in growing the silicon-dioxide insulating layer, on which you would later evaporate metal. To grow the silicon dioxide, you normally placed a silicon wafer in a quartz tube, heated the wafer to about 1,200ƒC, then subjected it to steam or oxygen to grow the oxide. If you used steam, says Hofstein, the result was horrendous, but if you grew the oxide in oxygen, the result was merely horrible.

The going wisdom was that it was necessary to find a better way to clean the silicon before growing the oxide. There seemed to be a million cleaning recipes, Hofstein recalls, none of which did any good. Hofstein began to feel that no room-temperature cleaning would have a permanent effect after a wafer was subjected to 1,200ƒC.

A time-honored technique

He then remembered that, to improve surfaces, metallurgists annealed metals in a hydrogen atmosphere. He tried it. Hofstein ran hydrogen over oxidized wafers in an induction furnace, then lowered the temperature slowly. This was classical annealing. Then he evaporated the metal gate onto the silicon dioxide. Success! It was an exhilarating moment.

While everybody tried to figure out how to clean the silicon before oxidizing, the trick was to anneal after oxidizing.

Hofstein and Heiman were in inventor's heaven. They had created practical MOS transistors. Unfortunately, they weren't stable. The same applied voltage would yield different currents, depending, it seemed, on the state of the union or the phase of the moon.

While they worked closely together, Hofstein paid more attention to the transistor itself and the instability problem, and Heiman worked more on extending the device and developed the first MOS IC, a 16-gate array. That was unstable, too.

Meanwhile, at the other end of the country, Fairchild was desperately trying to figure out how RCA had overcome the problem of surface states, effectively electron traps on the silicon dioxide surface that limit conduction. Fairchild had discovered that sodium makes silicon unstable, but the company didn't know how to overcome those troublesome surface states.

Along about 1965, Hofstein, with the blessings of the IEEE, organized the Silicon Interface Specialist Conference. Since he came up with the idea for the conference, he was entitled to select the locale. Here, too, he made history. He chose Las Vegas.

Las Vegas! Of all places

This was ridiculous; some felt it was actually obscene. He organized the first professional conference ever held in Las Vegas, unknowingly sowing the seed for Las Vegas becoming a major site for technical conferences. The Stardust Hotel, thrilled at hosting real scientists, housed the conference attendees free. Hofstein suggests that the hotel kicked out the Elks, threw out the fancy ladies and swept the sawdust off the floor.

Now the situation at the conference is that, in one corner, RCA can make MOS transistors, but they are unstable. Fairchild can't make transistors, but they're stable. In one corner is RCA's Steve Hofstein and Fred Heiman; in the other corner is Fairchild's Andy Grove. Each had probably been warned to be wary of violating antitrust legislation.

In the final round, Hofstein and Grove are in the Stardust swimming pool, possibly chatting about the weather and the Las Vegas ladies. The conversation may have shifted to MOS transistors. Grove may have said that Fairchild had spent more than a million dollars tackling the surface-state problem. And Hofstein may have said that RCA had spent a similar sum on the instability problem. No one knows if or what was said in the pool, but it's possible that Grove may have said, "If you tell me how to get rid of those damn surface states, I might be able to help you with the instability." Or maybe Hofstein said: "If you tell me how to get rid of the instability, I might be able to help you with the surface states." The interchange might have reminded one of the old game of "I'll show you mine if you show me yours."

Two portentous words

At some point, it's possible that Hofstein said, "hydrogen," and Grove said, "sodium." Two million dollars and two words. Within a few weeks the industry benefited from two good sources of MOS field-effect transistors and, not much later, from a growing industry that made larger and larger MOS ICs. With quite a bit more than 1,000 transistors.

Hofstein left RCA in 1968 to found Princeton Electronic Products Inc., which produced the first silicon-target, scan-converter tube which, in turn, led to the birth of video imaging. In time, PEP provided the image computers for airport X-ray systems, for ultrasound medicine, for space image conversion and for the FBI's fingerprint-imaging systems.

Heiman left RCA in 1969 to join Hofstein at PEP, then, in 1972, joined candy giant M&M Mars as president of its electronics division. He moved to Intel, where he served as director of corporate planning from 1983 to 1986, then to Symbol Technologies, the giant in bar-code scanning, which was founded by an old college buddy, Jerry Swartz. There he started the company's RF Systems Division. He still spends one day a week at Symbol and serves on its board of directors.

But Heiman's major love these days is underwater videography. At a recent international film festival, he won the Stan Waterman award for excellence in underwater videography. The award is named for the man who is one of the world's foremost videographers.

Hofstein sold his equity in PEP in 1978, did some consulting and then, in 1990, founded Ascot Technologies Inc., now ATI Systems Inc. which is involved in advanced imaging, the application of Expert System technology to data recognition and data capture even with varied documents and varied formats. He's now president and chief technologist.