To protect the mesa's sensitive junctions, each transistor was packaged into a pea-size hermetically sealed metal can and then tested. Fairchild shipped the first hundred of them to IBM on schedule that July, billed at US $150 apiece. The next month, at the WESCON electronics trade show, the founders discovered to their delight that they were the only ones with silicon mesa transistors on the market. “We scooped the industry!” Noyce said, exulting at a Fairchild meeting a few days later.

About the only person at Fairchild not celebrating was Hoerni. A proud, charming, but irascible and often volatile man, the scion of a Swiss banking family, he was miffed that his p-n-p approach had been passed over. But he was also a hardheaded contrarian whose creative fires were stoked by adversity. Hoerni not only didn't give up, he set out to develop an even better transistor. Later that year, he returned to the ideas written down in the opening pages of his notebook. Could the oxide layer in fact be used to protect the sensitive p-n junctions? There were indications it might. That spring, reports had come in from Bell Labs that the oxide layer indeed protected the silicon underneath. Why not the junctions, too?

With a doctorate in crystal physics, Hoerni realized that the impurity atoms coming through the tiny openings in the oxide layer would diffuse sideways nearly as well as downward into silicon's crystal structure. Which meant that the junction interfaces would curl up under the oxide layer surrounding an opening, just micrometers farther out from its edges. If left in place instead of being etched away, he figured, the oxide layer could protect those junctions.

But the device Hoerni envisioned would not only be more difficult to fabricate, its structure flew in the face of conventional wisdom. Especially at Bell Labs and Western Electric, the oxide layer was considered “dirty”—filled with impurities after the diffusion process—and thus had to be removed.

Meanwhile, serious concerns began to emerge in late 1958 and early 1959 about the mesa transistors Fairchild was selling. Some of the devices were experiencing amplification instabilities, and others were malfunctioning. One important customer reported that a transistor had suddenly stopped working altogether. A Fairchild technician eventually traced the failures to tiny dust particles and solder fragments trapped inside the cans. The specks were attracted to the junctions by the strong electric fields there. In a subsequent quality-control procedure that became known as the tap test, workers would tap on the cans with pencil erasers, trying to dislodge any bits that might short out the junctions. If that happened, the transistor was discarded. Those were anxious days for the brash young firm, for such failures in its only product threatened its very existence.

Hoerni's single-minded pursuit of a more reliable transistor proved timely indeed. In what Moore described to me as a “kludge experiment” intended to assess Hoerni's ideas, a technician deliberately left the oxide layer on top of one of the p-n junctions in a mesa transistor. When tested, it had substantially better amplification stability—suggesting that Hoerni was truly onto something. On 14 January 1959, he had two of his notebook pages typed up as a formal disclosure and sent to John Ralls, Fairchild's patent attorney. Other than a few minor corrections and better drawings, it was identical to the notebook entry he had written more than a year earlier.

One problem with Hoerni's approach—and part of the reason nobody attempted it at first—was that his transistor structure was more complex than the mesa's, requiring a fourth photolithographic mask to fabricate it. Last and Noyce's step-and-repeat camera could accommodate only three masks. But that February, Last “jury-rigged a fourth mask” for this purpose, he recalled in a recent telephone interview. On 2 March, Hoerni wrote another entry in his notebook titled “A method of manufacture of PNP transistors with oxide protected junctions.” In two more pages of text and drawings, he indicated specifically how to fabricate such a device, though still stubbornly using silver for the electrical contacts on the top side. By then, his technicians were already transforming his novel ideas into actual fabrication processes.

But all that progress came at a time of upheaval at Fairchild. The same week that Hoerni was jotting down his fabrication ideas, Edward Baldwin, who had been hired from Hughes Electronics Corp. to serve as Fairchild's general manager, departed abruptly to found Rheem Semiconductor in Mountain View, taking with him five key people from the manufacturing division. After persistent urging by the other Fairchild cofounders, Noyce stepped up to replace him, and Moore took over Noyce's position as research director.

The following week, Hoerni invited several colleagues to watch a demonstration of his new prototype transistor. Under a microscope it appeared unlike any other Fairchild device. Less than a millimeter across, it was completely flat—no mesa protruded in the middle. All that was visible was a circular metallic dot with a metal ring around it, plus the oxide surface layer between them. It resembled a bull's-eye target with a portion of it pulled out like a teardrop, making it easier to attach a wire [see photos, “Silicon Flatland”].