Photo: Wayne Miller/Magnum Photos/Fairchild Semiconductor

THE FAIRCHILD EIGHT: From left, Gordon Moore, Sheldon Roberts, Eugene Kleiner, Robert Noyce, Victor Grinich, Julius Blank, Jean Hoerni, and Jay Last.

That fall, the Fairchild founders worked feverishly to get everything up and running. Moore set up diffusion furnaces designed to impregnate silicon wafers with micrometers-thin layers of impurities—chemical elements such as boron, phosphorus, or aluminum that alter silicon's electrical characteristics to form a transistor's building blocks. Metallurgist Sheldon Roberts took on the task of growing high-purity silicon crystals from which the wafers could be sliced. Noyce and Last developed methods to do photolithography and oxide masking, by which they could define precise openings in a thin silicon-dioxide layer on the wafer surface; the impurities would diffuse through these openings into the underlying silicon. Other cofounders dug into manufacturing, testing, and selling the high-tech devices to aerospace customers.

And then there was Hoerni. A theorist with not one but two doctorates, from the Universities of Cambridge and Geneva, he had come to the United States to pursue postdoctoral studies at Caltech. In 1956, Shockley lured the 32-year-old physicist away from academia and assigned him to do theoretical calculations of diffusion rates. At first, Hoerni was cloistered in a separate office, but he kept coming around and snooping in the lab in the main building—which gave him valuable insights into solid-state diffusion. Later, at Fairchild, while the others worked on building or installing equipment, he mostly sat in his office and “scribbled in his notebook,” Moore told me.

On 1 December 1957, Hoerni grabbed his crisp new lab notebook and began writing an entry titled “Method of protecting exposed p-n junctions at the surface of silicon transistors by oxide masking techniques.” In a loose, fluid scrawl interspersed with three simple drawings, he described a revolutionary new way to fabricate transistors—unlike anything ever before attempted.

The most advanced silicon transistors at that time were called mesa transistors because they resembled the plateaus of the American Southwest, the impurity layers running laterally like the colorful rock strata [see illustration, “Mesa vs. Planar”]. These transistors basically consisted of three impurity layers piled up vertically, each rich in either electrons (n-type) or electron deficiencies, better known as holes (p-type). The main drawback of the mesa structure is that its p-n junctions, the interfaces between layers where the transistor's electrical activity occurs, are exposed at the edges. Bits of dust or drops of moisture can contaminate the sensitive interfaces and disrupt their normal electrical behavior.

Hoerni's idea was to protect the p-n junctions by keeping the oxide layer in place upon the silicon after the diffusion process; the standard practice at the time was to etch that layer away, baring the junctions. “The oxide layer so obtained is an integrant [sic] part of the device,” he wrote in his notebook that December day, “and will protect the otherwise exposed junctions from contamination and possible electrical leakage due to subsequent handling, cleaning, and canning of the device.”

It was a brilliant conception but too far ahead of its time. Hoerni's approach would require additional fabrication steps, and making mesa transistors was already at the limits of the possible. Bell Labs and Western Electric had produced prototypes of mesas, but no company had sold one on the open market.

In early 1958, Fairchild secured its first purchase order for silicon transistors from IBM's Federal Systems Division, which planned to use them in the onboard computer it was designing for the B-70 bomber. Fairchild, which didn't even have prototypes, faced the formidable challenge of delivering real working devices. To maximize the chances of success, the cofounders decided to develop two different kinds of mesa transistors. A group under Moore pursued the n-p-n transistors, which were thought to be easier to fabricate, while Hoerni formed another group to delve into the p-n-p versions.

Crucial to both efforts was the work Last and Noyce were doing on the optical methods needed to transfer the patterns defining a transistor's features onto the silicon wafer. On a trip to San Francisco, they purchased three 16-millimeter lenses from a camera store and used them to fashion a step-and-repeat camera, a contraption that produced rectangular arrays of tiny, identical images on photographic plates, called masks. Workers shone light through the masks onto a special photosensitive resin that had been deposited on the wafer's oxide surface layer. When they subsequently rinsed the wafer in a powerful acid, it etched the illuminated areas away, exposing the silicon beneath them. Thin layers of impurities were then diffused into the silicon through the resulting openings. Using such techniques, Fairchild could batch-process hundreds of identical transistors on a single wafer.

Another breakthrough was the use of a single metal to make the electrical connections to both n-type and p-type silicon, an approach that greatly simplified the manufacturing process. Moore had been struggling with this issue, trying many different metals, when Noyce happened by his lab early one day and suggested aluminum. As a p-type impurity, aluminum easily bonds to p-type silicon but often sets up a current-blocking p‑n junction when it is deposited on n-type silicon. Moore found a way around this problem by starting with n-type silicon that had more impurities than usual. Moore's group got its n-p-n transistors into production in May 1958, well ahead of Hoerni's team, which had opted to use silver for electrical contacts.