Frustrated, Hoff went to Noyce. “I think we can do something to simplify this,” Hoff said. “I know this can be done. It can be made to emulate a computer.” Hoff sketched out his thoughts, which would have required only one microprocessor and three other chips (two memories and a shift register). Noyce, who never would have claimed to know anything about computers, kept pushing Hoff, asking question after question, all of them so basic that Noyce was almost apologizing for his lack of knowledge. “Um, can you tell me the functions of a computer operating system?” What did it mean, precisely, to build a computer that acted like a calculator? What was it, exactly, that the chip would need to do? It was the same Socratic method of forcing people to “argue ourselves into some smart things,” as Vadasz put it, that had worked so well in the lab.27
At the end of the conversation, Noyce told Hoff, “Why don’t you go ahead and pursue those ideas? It’s always nice to have a backup position.”28
With this single seemingly casual remark, Noyce directed someone who did not officially report to him (Hoff says he nonetheless “tended to report to Noyce”) to do something that was not in his job description—something that ran counter, in fact, to what the customer’s representatives wanted—at a time when the company had yet to generate revenue and when its key technical employees could ill-afford distractions. And Noyce did this simply because he thought Hoff’s ideas were intriguing. That is one take on Noyce’s telling Hoff to go ahead. It would have been Andy Grove’s take at the time.
A second interpretation would look like this: Noyce told Hoff to pursue his ideas precisely because Hoff was so important to the company. At Shockley, Noyce had seen how a boss’s lack of interest stunted Noyce’s own work, cut the company off from important ideas, and devastated his young researchers’ commitment to the firm. He was not going to risk doing the same thing to Hoff.
The first interpretation certainly has merit. Even Noyce would have admitted in 1969 that work on the Busicom chip was orthogonal to Intel’s primary business of building semiconductor memories. And that business was not going particularly well. The excited call Moore placed to Noyce’s Aspen hospital bed was a bit premature. For several months afterwards, yields on the silicon-gate MOS memory hovered around two working chips per wafer, less than 10 percent of what the company thought it needed to be successful. Every time even a single good memory came off the line, the silicon-gate MOS team announced it over the company intercom. The MOS chip underwent nearly 20 design changes. The engineers and designers altered the brand of bottled water they used. They varied the acid dips. They reverse engineered a competing chip from a company called Mostek, which had spun out of Texas Instruments. They hung a rubber chicken over an evaporator to serve as a good-luck talisman.29
The breakthrough came in late 1969, when someone changed the formula of the acid dip yet again. Suddenly the yield went from two chips per wafer to 25. The employee in charge of sorting good dice from bad started yelling, “Holy hell, look what’s going on here!” People came pouring out of their offices, urging the MOS team to test another wafer, and then another. When it was ascertained that the key change was the acid dip, an employee recalled, “Les [Vadasz] was so excited he started jumping up and down and yelling, ‘It’s a sooper dip,’ with his Hungarian accent, ‘It’s a sooper dip, it’s a sooper dip,’ over and over again. Someone heard him and went back and marked the container of acid, ‘Super Dip.’ So even though it wasn’t a precise chemical formula, for over a year that container just sat there with that label ‘Super Dip’ on it.”30
In September 1969, Intel introduced its MOS chip, a 256-bit memory called the 1101. It was not the “just right” solution Intel had hoped for. The chip proved too slow and, priced from 20 to 60 cents per bit (depending on the quantity purchased), too expensive to replace cores, which cost about four cents per bit, as a mainframe memory. Even when Intel reduced its initial price on the part by 75 percent, sales were sluggish. Only when Intel applied the basic MOS structure to shift registers, an already established market, did silicon-gate technology make a profit, albeit a small one, for the company. But this me-too business was not what Noyce and Moore had planned for Intel, nor was it a strategy for extended long-term growth.31
Meanwhile, Ted Hoff had been working on the novel general-purpose logic chip for Busicom. He did this in the time not spent on his “real” job managing applications research. Outside of the MOS team, with whom Hoff, who had not previously designed an MOS circuit, needed to consult fairly regularly, Noyce was his main contact on the Busicom project. (“He was always very encouraging, always very helpful, always had an idea,” Hoff recalls.) Moore was also intrigued by Hoff’s vision, though he admits “my enthusiasm may not [have been] so obvious at times.” In late August, an engineer named Stan Mazor began helping Hoff with the processor chip. With Mazor on board, the two men completed a block drawing of the architecture within two weeks. Throughout this time, the Busicom engineers were working on their own design.32
In August, Noyce sent a note to Busicom president Yoshio Kojima, warning him that the “complexity of the circuits for this machine” meant that there was “no possibility that we could manufacture these units for $50/kit, even for the simplest kit.” If Intel built the calculator set according to Busicom’s design, Noyce estimated that the final cost per kit would be around $300. (He hastened to add, “I do not criticize the design of the … calculator.”) Noyce ended with a question: “Is it reasonable to proceed with this development on the basis of this design, or should the project be abandoned?” Kojima must have immediately responded that he wanted to continue some sort of relationship, because on September 16, Bob Graham followed up by letter, again emphasizing the complexity of the Busicom chip set and ending with a proposal: why not try a different design, an Intel design, in fact?33
In October 1969, a pair of Busicom executives came to Intel to decide which chip set to pursue—Hoff and Mazor’s or the Busicom engineers’. The Intel design required four chips, with about 1,900 transistors on the most complicated chip, at a cost of roughly $155 per kit. The Busicom design would cost twice that and require 12 to 15 chips with about 2,000 transistors each. The Busicom executives chose the Intel design with its general-purpose logic chip over their own engineers’ designs—a decision Ted Hoff declared “a bit of a coup.”34
The Busicom team returned to Japan. But it was February 1970, before the agreement between the two companies was formally signed. From October to February, Hoff turned his attention to a contract for a central processor for Control Terminal Corporation, and no one at Intel was designated to lead the work on the Busicom chip set. Perhaps it is not surprising that the chip set lingered untended, since both Noyce and Hoff, the two men within Intel most excited by it, enjoyed coming up with ideas more than implementing them. In March 1970, a Busicom executive sent a gently worded letter to Noyce wondering why he had heard nothing about their calculator chip set. He requested an update and specific signs of progress.35
The letter got results. Intel hired Frederico Faggin, an MOS process expert from Fairchild, to work on the Busicom chips. Faggin immediately began refining Hoff’s architecture and implementing the design into silicon, working in close consultation with the Busicom engineers, who had flown back across the Pacific to assist with the effort.
Noyce also went into Andy Grove’s office and sat down on the corner of his desk, a move that immediately raised Grove’s suspicions. Any time Noyce affected a faux-casual air, Grove knew he was not going to like the message. Noyce looked at Grove more from the corner of his eyes than face-on. “We’re starting another project,” Noyce said with a little laugh. Grove remembers thinking, “Go away, go away, we don’t have time for this.” What kind of company started new projects when its very survival was at stake?36
Intel was indeed struggling. One internal company timeline shorthands the first months of 1970 with the phrase, “Management Near Panic.” The next-generation silicongate MOS device was proving as
difficult to build as the 1101, even though almost everyone in the company had been assigned to work on it, and even though Intel was collaborating with computer manufacturer Honeywell to increase its chances of building a MOS memory with desirable features. At the end of the second quarter, Intel laid off about 20 people and reined in the most ambitious of their expansion plans—“probably a more painful task for Noyce than for Moore,” explained one contemporary account.37
The energy and optimism that ended the 1960s had cratered under their own weight, leaving only surplus inventory and employees without work to do. Headlines on a single page of an industry newsletter in July 1970 read “Raytheon Profits Off in Quarter,” “GT&E Quarter Net Slumps; Sylvania Nosedives 63%,” “Itek Earnings Skid 30.5% for Quarter.” The year 1970 threatened to end with Intel almost $2 million in the red.38
Arthur Rock felt a sense of urgency—even impending disaster—about Intel that he was not certain the management and directors shared. “Rock was quite comfortable making [the situation] sound quite severe and bad in order to make people be more realistic,” recalled one board member. “Noyce and Moore weren’t good penny-pinching, pencil-pushing types. Rock was a fiend of that sort.” Intel needed a source of significant revenue, Rock said—and quickly. They needed to put the next-generation silicon-gate MOS memory into production, even if it was not perfect. In October, Intel officially introduced the 1103, its second memory chip built using the silicon gate process, with a full-page ad blaring, “THE END: Cores lose Price War to New Chip.” The 1103 held four times more data than the 1101. It was also the first semiconductor memory that could be made in volume and thus the first that could really challenge cores on price.39
To be sure, the device was far from perfect. Among the 1103’s many failings known to Intel was the fact that, in Andy Grove’s words, “under certain adverse conditions the thing just couldn’t remember”—a problem for a memory. Some 1103s failed when they were shaken. A few developed moisture inside the glass used to seal them. Often no one knew why the devices would stop working. The problems inspired Ted Hoff to write a 28-page memo explaining the 1103’s operation and quirks.40
Andy Grove had nightmares that boxes and boxes of 1103s would be returned to the company for defects—and would ruin Intel entirely. Gordon Moore, on the other hand, wondered if, in some perverse way, the 1103’s problems made it easier to convince customers to use the device. Engineers who specialized in core memories recognized analogs in the 1103. Both suffered from voltage and pattern sensitivity, which means that their performance was affected by other nearby electronic devices. The 1103 “refreshed itself” every thousandth of a second, an operation that regularly caused problems; cores did something similar called “destructive read.” “All these things made the 1103 more challenging and less threatening to engineers [at customer companies],” Moore explains. “We did not plan it to happen this way, but I think that if [the 1103] had been perfect out of the box, we would have had a lot more resistance [to it] from our customers.”41
Even with its problems, the 1103 represented a real technical breakthrough. As Steve Jobs once said, when the light bulb was invented, people did not complain that it was too dim. A firm called Microsystems International Limited (MIL), the manufacturing arm of Bell Canada, approached Intel about serving as a “second source” on the 1103. Customers recognized that building state-of-the art semiconductors was a temperamental business, and so no customer was willing to leave an important order in the hands of only one supplier. They wanted an officially sanctioned alternative source of product.42
Noyce told MIL that if the company wanted to second source 1103s, all it had to do was “pay us our net worth and we’ll all teach you what we know.” This is essentially the arrangement to which MIL agreed. In exchange for $1.5 million (roughly Intel’s net worth at the time), plus royalties on MIL sales until the end of 1972, Intel would provide the know-how, technical information, and licenses necessary to produce 1103s. Intel would also send a team to Ottawa to help MIL set up a fab to build the devices. If the new MIL fab met certain production goals, another $500,000 would be transferred from MIL to Intel. Noyce told Intel’s investors that “[MIL’s] payments under the agreement will substantially improve our already good cash position, [and] the implied endorsement of our process excellence will also be a valuable marketing aid for us.” Noyce later said that the MIL deal was one of his favorites of his career because “it’s not every day you get a chance to double your net worth.”43
Operations manager Andy Grove, whom Noyce had not consulted on the second source plan, was irate when he learned of it. “We were seeing one or two [working] dies per wafer of 1103s. It’s hard to describe what shit we were in with that product. [This MIL deal meant] I would have to improve yields without a manufacturing manager or chief technologist,” both of whom would be part of the team transferring technology to Ottawa. Grove, who could not believe Noyce would even consider such a proposal, had tried several times to dissuade him from this plan of action. When spring turned to summer, and the deal seemed imminent, Grove tried again. He went into Noyce’s office and found him there, talking with Moore. Grove immediately began arguing against the plan “as aggressively as I was capable—which was very.” He insisted that Intel could not increase the 1103 yield, maintain the fab, satisfy customers, and find new prospects with half the staff in Canada. “This will be the death of Intel!” he yelled at one point. “We’ll get the $1.5 million, and then we’ll sink.”44
Noyce waited until Grove finished. Then he looked at the younger man, his eyes hard. “We have decided to do this,” Noyce said slowly. “You need to put your energies into figuring out how to do this.” Grove could bluster all he wanted, but Noyce ran Intel at this stage. Grove left, as he put it, “with my tail between my legs.” The MIL deal was inked in July, and at the very end of 1970, Intel received its first payment from the Canadian firm—$500,000—which lowered Intel’s losses for the year to $1 million.45
In the end, the arrangement with MIL could not have gone better if Intel had scripted it. The initial technology transfer was successful enough to earn Intel the $500,000 incentive bonus and allow the transfer team to return home in the spring of 1971. Intel paid for the transfer team members and their wives to fly to Hawaii for a three-day party. Noyce arrived in time to share a fancy dinner of chateaubriand and copious volumes of liquor with the celebrants at the Kauai Surf Hotel. After a few hours listening to the jazz band, he joined a group stripping to their undershorts for a quick swim at the beach. Noyce threw himself headfirst into the water with characteristic vigor—only to discover the tide had gone out. He emerged from the shallow waters covered in scrapes, cursing and laughing.46
The dive in the ocean is vintage Noyce. He was trying, as he always did, to be “just one of the guys.” Any sort of hierarchy—but particularly one that placed him at the top—made him nervous. He once said that what he loved about madrigal singing was its blend of multiple voices into one, with no single voice dominating: “Your part depends on [the others’ and] it always supports the others.” When asked in social situations about his profession, he would say that he was a physicist. At Intel, Noyce spoke of “hierarchy power” and “knowledge power” and firmly believed that when it came to technical decisions, the word of the person with the most knowledge ought to trump the opinion of the one with the higher title. Intel board meetings were not the typical two-hour drone of rubber-stamp motions and resolutions. Noyce insisted that operating managers present to the board—both to keep the directors informed and to teach the operating staff how to field tough questions and make compelling arguments.47
When an early employee wanted to see Intel’s organization chart, Noyce drew an X in the middle of a circle, and then drew seven more Xs along the perimeter of the circle. As the amazed employee looked on, Noyce proceeded to connect the center X to each of the other Xs in the system so the drawing resembled a wagon wheel. The X in the center, said Noyce, was the empl
oyee asking the question. He added, the other seven Xs “are me, Gordon, Andy, Les [Vadasz], Bob [Graham], Gene [Flath] and other people you’ll be dealing with.” Noyce’s point was that he expected a reciprocal relationship between all employees—from the center to the perimeter, and back again—and that formal reporting structures and hierarchies were largely irrelevant.48
“You never heard him say, ‘I did this, I did that,’” recalls Judy Vadasz, who is married to Les. “It was always, ‘we did this.’” Intel’s far-reaching stock-option plan can be seen as a reflection of Noyce’s democratic tendencies, which were also shared by Moore. The founders’ beliefs on this matter further led them to refuse to set aside parking spaces for executives, and to decide, when Intel moved to a new building in the mid-1970s, that everyone—including the founders—should work in essentially identical cubicles.49
A FEW MONTHS AFTER STARTING INTEL, Noyce told a reporter, “One of the reasons I wanted to set up this two-headed monster is so that either Gordon or I can feel free to take off without severe guilt feelings about leaving a job undone behind us.” Noyce “took off” to lead Wednesday evening rehearsals for his madrigal group. He permitted himself long weekends away from Intel during the ski season and a week-long family vacation. He and Betty sold the ski cottage she disliked. Instead they began an annual tradition of renting the entire lodge at Alpine Meadows near Lake Tahoe for several days every winter and inviting nearly two dozen people to join them. Noyce also served on the boards of three startups—including Eugene Kleiner’s short-lived company, Cybercom, which designed and manufactured computer peripherals such as printers and display terminals—each of which held monthly meetings.50
In 1967, Noyce was a founding director of Coherent Radiation, a company that developed and manufactured lasers for use in scientific instruments and as machine tools in industrial applications. Noyce had learned about the company from Charles B. Smith, an associate with the investment arm of the Rockefeller family, who were willing to invest in the company if Smith could find a “man on the West Coast to keep an eye on things.”51
The Man Behind the Microchip Page 29