He began talking things over with Noyce, who surprised him by suggesting he try using aluminum to make the contacts to both the P- and N-type silicon. In the semiconductor industry of the late 1950s, this was a nearly preposterous suggestion. Everyone knew that aluminum acts as a P-type impurity, which means that while it makes an ideal contact for P-type silicon, it is a terrible choice for contacts to N-type silicon. In fact, aluminum contacts to N-type silicon tended to add another P-N junction to the transistor, rendering it useless. Of all the people at Fairchild, Noyce, the most experienced, should have known better.6
Noyce was unswayed, insisting that if Moore used a new process developed by their colleague Jean Hoerni, then aluminum should work as a contact for N-type silicon as well as P-type. Noyce knew more than Moore about the surface of semiconductors, which was where the connection would be made, and the Hoerni process looked promising. Since Moore was running out of alternatives, he decided to try using aluminum as the only contact metal, even though, he recalls, “all the conventional wisdom said it wouldn’t work.”7
What happened next perfectly illustrates Noyce’s approach to invention. He left his idea in Moore’s hands and moved on to something new. He began sketching ideas for a semiconductor switching device and for a scanning device that could “serve in many of the same applications in which an electron beam device is used, such as display, camera tubes, beam switching, amplifiers of the traveling wave variety, etc.” He patented both of these ideas.8
Meanwhile, Moore sweated out the details of the aluminum contact process, laboring alone and in occasional consultation with Noyce, Jean Hoerni, and Sheldon Roberts. Every time one problem seemed solved with the aluminum contacts, another arose. Wires connected to the contacts fell off indiscriminately. The junctions (the meeting points of the P- and N-type regions) began leaking a tiny bit of current in the wrong direction. The metal pulled away from the silicon. Noyce was a willing consultant to Moore in these travails. He tried to imagine how the pulling-away problem could be turned into an asset, and he suggested that Moore try plating the back of the transistor with nickel to solve the leaking junction problems. But it was the big idea—why not try aluminum?!—not the details of making that idea work (“reducing the idea to practice,” in scientific parlance) that Noyce cared about.9
Moore’s work paid off on May 2, 1958, when he could finally report that “pure Al[uminum] works very well on N-P-Ns in all respects.” The combination of aluminum contacts and nickel plating yielded transistors with hard junctions and contacts large enough to minimize spreading resistance but not so large that they shorted across the junctions. Noyce and Moore filed for a joint patent on the process. It was one of the most significant early patents at Fairchild Semiconductor, since aluminum became the standard metal for contacts in the semiconductor industry.10
In this case, as in many others, the Fairchild researchers did not understand precisely why this innovation worked. In an academic environment or at Bell Labs, which was the research arm of a regulated monopoly, this question would have been paramount. At Fairchild Semiconductor, however, why something worked was far less important than the fact that it did. In a newborn company with only one customer of any significance (IBM), pursuing science for its own sake was an ill-afforded luxury. Thus early research at Fairchild Semiconductor was almost all process oriented, with building a saleable product the fundamental goal of the research lab. Noyce and several of the other researchers directly refer to the IBM specifications in their scientific lab books, giving as much weight to performance targets as they did to the science they needed to reach those targets. Noyce, who believed that “the only thing that’s technologically exciting is something that has a need for it,” was the ideal man to oversee this work in the lab.11
Of the seven patents Noyce filed in his first 18 months at Fairchild, the best known is #2,981,877 for “Semiconductor Device-and-Lead Structure.” Fairchild called the product developed on the basis of this patent, which Noyce filed in 1959, a “monolithic integrated circuit.” Years later, John Bardeen, co-inventor of the transistor, would call it an invention “as important as the wheel.” The integrated circuit made electronic devices smaller, faster, and cheaper than ever before. Every modern computer, microwave, airplane, traffic light, missile, telephone, ATM, and automobile in use today has at its center direct descendants of the integrated circuit Noyce sketched in his device-and-lead patent application.12
An integrated circuit is a complete electronic circuit built on a chip of silicon small enough to be carried off by an ant. Every electronic circuit is actually an interconnected series of discrete components that serve specific functions—resistors to control current, diodes to block it, transistors to amplify it. When these discrete components are strung together, the resulting circuit can do anything from adding millions of numbers to sensing when coffee is done brewing. Before the integrated circuit, these components were attached to each other one at a time, by hand, in a process fraught with errors and failures. With the integrated circuit, by contrast, the components could be printed and connected to each other simultaneously in a reliable process that resulted in a complete circuit no larger than any one of the components taken individually.
In the 1950s, semiconductor firms manufactured hundreds of identical discrete components (transistors, for example) side by side on a silicon wafer. All these companies used a process similar to the one Fairchild Semiconductor employed to build its mesa transistors. Before the integrated circuit, the last step of the production process was a painstaking assembly line affair. Hundreds of women attired in identical lab coats sat side by side hunched over high-powered microscopes that magnified the wafer so that the women could slice apart individual components and attach leads and wires to them using tiny tweezers. The individual components were then tested, packaged, and shipped to customers. The customers would then reconnect various components to each other to configure a circuit.
Noyce, along with everyone else in the electronics industry, knew that this method of cutting apart components only to reassemble them later was inefficient. The ideal procedure would be somehow to connect the components to each other at the same time that they were built. Knowing this and doing it were two very different exercises, however. Under a microscope, the metal wires used for interconnecting discrete components, while thinner than a human hair, nonetheless resembled huge logs capable of flattening a component’s delicate architecture. No one could figure out how to deposit this metal on the surface of the semiconductor wafer in amounts small enough not to short out the circuit, but large enough to connect the components to a package, power source, and other devices. This is why the “girls” made the interconnections by hand, after the components had been safely cut from the wafer.
This cumbersome method of interconnecting components posed an additional problem: even if the components were themselves reliable, an accomplishment one could never take for granted, bad interconnections could render a circuit useless. Each component could be attached to many others, which meant that as the number of components in a circuit grew, the number of interconnections grew exponentially, until a circuit board could come to resemble a nesting ground for tiny wire-quilled hedgehogs. Circuits in the 1950s regularly consisted of hundreds, or even thousands, of discrete components, and people were predicting that the demand for “space-age electronics” would soon push the numbers of components into the hundreds of thousands, and the numbers of interconnections into the millions.
Although the women in the lab were chosen for their dexterity, it was physically impossible for anyone to solder millions of connections perfectly. This meant that given a big enough system with enough interconnections, even if every component in a system had a reliability of better than 99 percent, failure was statistically possible within the first two minutes of operation. The interconnections, often shorthanded the “tyranny of numbers,” were the industry’s Achilles’ heel. If the problem could not be resolved, real progress in elect
ronics would grind to a halt.13
The military desperately wanted to end the tyranny of numbers. One air force-sponsored effort tried to synthesize, atom by atom, a single piece of solid metal to achieve a complete circuit function. Another tried to build the wiring right into the discrete components and then snap the components together like a child’s plastic pop-beads. Neither effort proved commercially viable, nor did attempts to build tiny electron tubes or to grow a complete circuit with little or no wiring needed to interconnect the components. By the end of the 1950s, at least 20 companies, ranging from small component makers to huge equipment manufacturers, were on a quest to find a solution to the interconnections problem.14
Neither Bob Noyce nor anyone else at Fairchild Semiconductor set out with a grand plan to resolve the tyranny of numbers, though the integrated circuit accomplished precisely that. The course of events that led to the Fairchild integrated circuit is murky at best. Discoveries and ideas wind past each other and double back. Noyce’s involvement with the integrated circuit changed dramatically halfway through, when he was named general manager and left his lab bench, essentially forever. And the fame that the moniker “inventor of the integrated circuit” brought to Noyce left some other contributors to the invention resentful. The lawyers who defended Noyce’s patent, and the cutthroat pricing strategies that Fairchild adopted to ensure that their circuit became the industry standard, are also woven into the “invention” of the integrated circuit. This much is certain: the integrated circuit rests on the planar process invented by Jean Hoerni—no planar, no integrated circuit—and that invention, like so many at Fairchild, came in an effort to solve an immediately pressing practical problem.
In late 1958, several of Fairchild’s mesa transistors were returned to the company after random catastrophic failures. Tests in the lab soon revealed that it took nothing more than a sharp tap of a pencil against the side of a Fairchild transistor to make it stop working. The problem did not affect every transistor, but reliability was the single most important selling point of the Fairchild transistor, and so R&D had to solve the tap problem—immediately.
It is hard to appreciate how crude the techniques and analytical tools available to the Fairchild researchers were at this stage. Troubleshooting the tap problem meant taking a transistor, tapping it ten times with a pencil, recording whether it failed, pounding it on the table ten times, again recording results, and then finally opening up the failed transistors, at which point the diagnoses would read something like this from Gordon Moore’s notebook: “They each showed a fleck of crud (it looked like yellow metal) on the top edge of the mesa.”15
Eventually a skilled technician determined that during the process of sealing the cans, a tiny piece of metal was flaking off, bouncing about within the can, and eventually shorting out the transistor. Before he made this discovery, however, a few key lab employees launched a pull-everything-out-of-your-hats effort to solve the tap problem. Jean Hoerni began once again considering a subject that he had first begun to explore with Noyce and Moore at Shockley: how to protect a transistor’s junctions without contaminating them. The question had clearly preoccupied Hoerni since the very earliest days of Fairchild Semiconductor; in December 1957, when the company was scarcely two months old, Hoerni had proposed that “the building up of an oxide layer … on the surface of the transistor … will protect the otherwise exposed junctions from contamination and possible electrical leakage due to subsequent handling, cleaning, [and] canning of the device.” This was unusual thinking for the time. Most people thought that the oxides that grew naturally on the clean surface of a semiconductor needed to be washed away so they would not trap impurities between the oxide and the silicon. Hoerni instead wondered if the oxide might protect the surface—even from impurities as gargantuan and menacing as a rogue sliver of metal. About six months after Hoerni’s journal entry, several Fairchild researchers attended a conference where they learned that a group at Bell Labs had demonstrated that an oxide layer indeed could stabilize the surface of the semiconductor.16
At the end of Hoerni’s December 1957 entry, Noyce wrote “read and understood,” added the date, and signed his name. Scientists regularly asked their peers to “witness” their most important work in this way. In the future, the witnessed signature could help to document when precisely a researcher had first noted his ideas—an essential element for any patent application. But although Noyce understood Hoerni’s ideas and almost certainly knew of the Bell Labs oxide findings, neither he nor anyone else in the lab paid any special attention to Hoerni’s thoughts on oxide layers. Instead, the lab team focused on the IBM transistor and the other transistors and diodes that followed it. Solving the problems that led directly to more sales and building products that would make money was far more important than optimizing the Fairchild transistor process to an ideal level, which is what the oxide layer aimed to do. If it weren’t terribly broke, the Fairchild men saw no reason to fix it.
The tap problem, however, made it apparent that the Fairchild process was broken, or at least, that it was producing breakable transistors. For three weeks in January 1959, Hoerni thought about nothing but oxide layers. Imagine a cake that somehow grows its own icing. The surface of the cake is analogous to the surface of the semiconductor; the icing is similar to the oxide layer. Bell Labs had shown that the oxide icing would protect the surface. Now Hoerni wanted to develop a way to grow a perfectly consistent icing and then work between it and the cake—right at the all-important surface of the semiconductor, the same area that had interested Noyce ever since his dissertation research. Hoerni needed either to lift up the icing or drill through it to get down to the surface of the cake.
After what he described as an “epiphany” in the shower one morning, Hoerni realized that he ought to be able to use a mask to create an oxide layer over the entire wafer and then engrave a precisely located “window” in the oxide through which impurities could be diffused to form the base. At the same time as this first diffusion, another layer of oxide could be grown on the surface of the silicon. Then another window, another diffusion (this time to form the emitter junction), and another new layer of oxide. On January 14, 1959, Hoerni wrote out a two-page patent disclosure of this process. The next week he wrote another disclosure on a closely related process. Noyce witnessed this second disclosure and almost assuredly saw the first, as well.17
During these same remarkable weeks of January 1959, Noyce began sketching out his ideas about the integrated circuit, which he classed under the heading, “Methods of Isolating Multiple Devices.” He wrote, “In many applications now it would be desirable to make multiple devices on a single piece of silicon in order to be able to make interconnections between devices as part of the manufacturing process, and thus reduce size, weight, etc., as well as cost per active element.” Noyce imagined making an adder circuit from diodes and included thoughts on using P-N junctions to isolate components from each other so they would not interfere with one another’s electrical characteristics. (The work of Kurt Lehovec at Sprague had introduced Noyce to the possibility of using junctions to isolate devices.) Noyce included among the “important features” of such a circuit “use of the SiO2 [silicon dioxide] layer as an insulator to isolate contact strips from the underlying silicon” and “protection of junction at the surface with an oxide layer.” Noyce also refers to “impurities diffused through the holes in the oxide.”18
Noyce’s thoughts on the integrated circuit are thus directly and inextricably linked to Hoerni’s oxide work. In essence, Noyce was imagining that Hoerni’s process would make it theoretically possible to drop a relatively large bit of metal onto the surface of the semiconductor wafer, on top of the tiny holes etched in the icing. If done properly, precisely the right amount of metal would touch the silicon in precisely the right place, and any of the metal that happened to “overhang” the hole would benignly sit on top of the icing, unable to affect the rest of the circuit.19
“I don’t h
ave any recollection of a ‘Boom! There it is!’ light bulb going off,” Noyce later said of his ideas. Instead, he conceived of the integrated circuit in an iterative method he described thus: “[I thought,] let’s see, if we could do this, we can do that. If we can do that, then we can do this. [It was] a logical sequence. If I hit a wall, I’d back up and then find a path, conceptually, all the way through to the end. [Once you have that path], you can come back and start refining, thinking in little steps that will take you there. Once you get to the point that you can see the top of the mountain, then you know you can get there.”20
After noting his ideas in his lab notebook, Noyce did … nothing. He showed the entry to no one—he did not even have it witnessed—and failed to mention it not only to Hoerni, on whose ideas it leaned so heavily, but also to any other co-worker in the lab. Twenty-five years later, Noyce explained his inaction thus: “We were still a brand new company … worried about basic survival. That meant getting transistors out the door. The integrated circuit seemed interesting, it was something that might make you some money somewhere down the road, but that was not a period when you had a lot of time for it.” This comment offers a compelling reason for why Noyce did not push the lab to work on his ideas, but it does not offer any insight into why he did not at least mention their existence. Perhaps Noyce considered his entry as some form of theoretical doodling—recall that when he made his lab book entry, Hoerni’s process was simply an intriguing idea. There had been no evidence that it would actually work. And if it had not worked, Noyce’s integrated circuit ideas would have been moot.21
Several members of the founding group offer their own provocative explanation for Noyce’s silence: the idea was too obvious to bother mentioning. Of course, you would want to try to interconnect components over the oxide layer if the oxide layer stabilized the surface of a semiconductor. The idea was positively self-evident.22
The Man Behind the Microchip Page 16