Years later Lynn Conway could still remember the moment she first laid eyes on the chip that would launch a new science. It was a week or two after Christmas 1979. She was seated before her second-floor window at PARC, which looked down on a lovely expanse of valley in its coat of lush winter green, sloping down toward Page Mill Road just out of view to the south. But her eyes were fixed on a wafer of silicon that had just come back from a commercial fabrication shop.
There were dozens of chip designs on the wafer, mostly student efforts from a Stanford course being taught under PARC’s technical supervision. They all strived toward an intricate machine elegance, comprising as they did tens of thousands of microscopic transistors packed into rectangular spaces the size of a cuticle, all arranged on a wafer that could fit comfortably in the palm of one’s hand. A few years earlier the same computing power could not have fit on an acre of real estate.
One design stood out, and not only because it bore along its edge the assertive hand-etched legend: “Geometry Engine © 1979 James Clark.” Where the others looked to be simple arrays of devices that formed simple digital clocks and arithmetic search engines and the like, Clark’s was obviously something more—larger, deeper, more complex than the others, even when viewed with the naked eye.
Clark’s got something really amazing going on in there, Conway thought to herself. But who knows what?
What Clark had going on, as it would turn out, was the cornerstone of an entirely original technology. The “Geometry Engine,” which he designed with the help of several of his Stanford students, was unique in compressing into a single integrated circuit the huge computing resources needed to render three-dimensional images in real time. After the appearance of Clark’s chip, the art and science of computer graphics would never be the same: The computer-aided design of cars and aircraft, the “virtual reality” toys and games of the modern midway, the lumbering dinosaurs of the movie Jurassic Park—they all sprang from the tiny chip Lynn Conway held by its edges that winter day.
With the Geometry Engine as its kernel, Clark founded Silicon Graphics Incorporated and developed it into the multibillion-dollar company it is today. But without Lynn Conway and PARC, he could not have built the Geometry Engine. The irony is that when Conway first proposed that PARC step into the vanguard of the science of designing such extraordinarily complex integrated circuits, many of her colleagues doubted it was worth doing at all.
Conway’s program would never even have gotten started had not Bert Sutherland decided that PARC needed a shot of “havoc.”
Sutherland had taken over management of the Systems Science Lab in 1975 after leaving Bolt, Beranek & Newman, the Boston consulting firm that had earlier given PARC Jerry Elkind, Bob Metcalfe, Dan Bobrow, and Warren Teitelman. Like them, he held strong views about research methods which did not always conform to PARC orthodoxy, especially as it was practiced in Bob Taylor’s Computer Science Lab. Sutherland believed that research conducted in a closed environment was doomed to suffocate, like an animal trapped in an airtight cage. He admired the Computer Science Lab’s work but regarded Taylor and some of his engineers as overly prone to facile prejudices and snap judgments—conditions, he thought, that deprived CSL of the necessary aeration. The harvest was its self-destructive elitism.
“They were the best and the brightest,” he said later. “That was the good news. The bad news was that they knew it.”
Sutherland did not allow SSL to become so sequestered. His policy was to keep its atmosphere enriched via continual contact with the outside world. One of his first acts upon succeeding Hall, for example, had been to send the engineers Tim Mott and Bill Newman on an “archeological dig” to Xerox’s copier sales office in Santa Clara, a few miles south of Palo Alto. The idea was for them to study how real office workers performed their daily routines, the better to design the equipment they would use in the future. This effort yielded OfficeTalk, a sophisticated and integrated system of office automation that heavily influenced the later design of the Star. Sutherland also recruited to SSL experts in cognitive science such as Stuart Card, Tom Moran, and John Seely Brown, whose research into how real people actually used computers, step by step and motion by motion, led to groundbreaking insights into man-machine ergonomics—insights that not even J. C. R. Licklider had anticipated when he wrote his own pioneering treatise on the subject in 1962.
At CSL, unsurprisingly, Sutherland’s democratic instincts provoked grumbling—wasting precious resources on anthropology, of all things!—even before he brought Carver Mead into the SSL tent. Then all hell broke loose.
Mead was one of the most popular and influential professors in the computer science department at California Institute of Technology, where Sutherland’s brother Ivan had recently become department chairman. Mead instantly struck him as the right person “to wander in and create some havoc” within PARC’s insulating walls. For sheer intellectual brio, Sutherland knew, Carver Mead could stand toe to toe with Butler Lampson and the rest of Taylor’s gunslingers any day. A compact, energetic man with a black mustache and goatee and lively, searching eyes, Mead possessed a confident mastery of electrical engineering, particularly at the extremes of the infinitely complex and the infinitesimally small—regions where ordinary engineers hesitated to venture but which he considered his personal preserve. He filled out that expertise with a breadth of interests that encompassed subjects as diverse as walnut farming and particle physics.
At the time of his first visit to PARC, he and Ivan Sutherland were deeply engaged in studying what happened to electronic systems at the edges of the physical scale—in other words, how minuscule a transistor could be without its becoming nonfunctional, and how large and complex a system one could build without its becoming unmanageable. At their core these questions were identical, for as transistors got smaller and more densely crowded on the silicon surface of an integrated circuit, the chip became more complex. The implications of this dual phenomenon were only just becoming understood when Bert Sutherland invited Mead to give a technical address at the Systems Science Lab in 1976. Mead’s formal topic was the design of silicon-based integrated circuits, but his real purpose was to propose a new way of thinking about computer design—one that threatened to make much of PARC’s work obsolete.
As Moore’s Law predicted, the technology of integrated circuits had been surging ahead ever since Intel—the company Moore co-founded—introduced its first microprocessor in 1971. The 4004 chip was fundamentally an arrangement of microscopic transistors that packed into the space of a matchbook cover the computing power of a mainframe—circa 1946. That was hardly an achievement to prompt a major reconsideration of computer architectures; but a year later came the 8008, which had twice the power, and in 1974 another doubling again.
There was no reason to think the trend would not continue well into the next millennium. From his academic aerie on Caltech’s Pasadena campus, Mead imagined the curve of shrinking transistor size and mushrooming density extending almost limitlessly into the distance. He believed that the traditional principles of computer design, of which MAXC and the Alto represented the intellectual pinnacle, were fated to fall off this curve well before it disappeared over the horizon. Both machines employed integrated circuits to help control their slowest peripheral devices, like the keyboard and mouse, but even those chips were of the passing generation known as MSI, or “medium-scale integration.” Mead had pioneered research into the next step—LSI, or “largescale integration”—and he was still thinking ahead. In partnership with Ivan Sutherland, he began exploring the difficulties and possibilities presented by the coming quantum leap in miniaturization, which would bring them to VLSI, or “very large-scale integration.” This was the gospel he came to preach at PARC.
Traditional computer design, he reminded his listeners, was essentially a mathematical exercise. One chose from the standard inventory of Boolean logic gates—ORs, ANDs, NORs, and so on—and arranged them to operate sequentially on a stream of bits. This worked fine as long as the logic elements (mostly transistors) were slow and expensive and the wires connecting them were relatively fast and cheap, as had been true throughout the history of digital computing. But it also meant that the blinding speed of digital computation was something of an illusion. The logic elements were such data bottlenecks that when you really examined what was happening inside the system, you could see that computers were still constrained “to perform individual steps on individual items of data”—that is, to do only one thing at a time.
The new technology would turn that architecture inside out. As silicon-based chips got smaller and denser, the microscopic transistors that were packed on them to make up the logic became faster and cheaper than the wires linking them. The wires became the bottlenecks. Soon the most important factor limiting the computer’s efficiency would not be the sequence of gates, but their geometric arrangement on a flake of silicon and the rising relative cost of transporting electrons over the minuscule pathways linking one to another. Computers were about to cease doing one thing at a time, in favor of doing many things simultaneously. Consequently, their architects would have to abandon the old methods of designing them simply as linear sequences of logical functions. They would have to also consider how to get bits from one logical function to another along the shortest path.
Traditional digital technology required designers to think like factory planners figuring out how to get raw materials in one end of a building and finished product out the other. Silicon, however, “forced you to think like an urban planner,” Conway said later. “You had to think hard about where the roads go.” Just as cities reaching a certain size suddenly find themselves threatened by highway gridlock, she observed, in VLSI, “if you weren’t careful you could end up having nothing but roads going nowhere.” Fortunately VLSI also offered a way out of that quandary: Because the logic gates and other devices were now so cheap, “it didn’t cost you anything to have more of them, if that paid you back by having less highway.”
For engineers who had reached the top of their game the old way, VLSI was full of murky ideas. Many doubted it was physically possible even to fabricate functioning devices as tiny as the ones Mead prophesied. Even those who thought VLSI an interesting idea with great potential questioned whether it would ever supplant the tried-and-true architectural structures that had brought them this far. In CSL the general opinion was that VLSI was more than they needed to have on their plates. “We didn’t have to be able to design chips,” Lampson said—not while the industrial designers at Intel and other chip companies were already hard at work on it.
In any case, PARC could hardly hope to contribute much to this nebulous science. At CSL “they were already out front in their own revolution,” one researcher later remarked. “To them VLSI was not really mainline, it was just this weird sort of thing happening somewhere else.”
But for two of Sutherland’s laboratory scientists, Lynn Conway and Douglas Fairbairn, Mead’s talk scored a direct hit.
Conway was a rarity at PARC—an accomplished designer of advanced mainframes who chose to give the hardware gurus of the Computer, Science Lab a wide berth. She ranked among PARC’s senior veterans, having joined in 1972 from IBM, where she had helped design a supercomputer at the Yorktown Heights lab, and Memorex, where she had worked as an architect of minicomputers. But at PARC she had played no role in developing the Alto or MAXC. On the contrary, something about the intellectual gunplay of CSL alarmed her, as did the intimidating presence of Butler Lampson.
“I always had a hard time dealing with Butler,” she recalled. “He had this complete photographic memory of all theory that ever existed about anything, but sometimes that can be kind of a mental block to being creative. You can be so confrontational and challenging about how smart you are that you can’t always see that somebody else has got this cool idea.”
Like Kay, Tesler, and Shoup, Conway found the ambiance more obliging among the Systems Science Lab’s lunatic fringe. “Taylor was someone who could manage the ‘neats’ and Bert could manage the ‘scruffles,’” she remarked. “In SSL I could survive. I could get all excited about an idea that was half-formed and go tell Bert about it, and he’d get all excited about it, maybe tell me somebody I should talk to about it. In CSL I’d be really afraid to present anything until it was perfect, and it would probably get immediately shot down anyway.”
Her inaugural assignment at PARC had been something of an acid test in the implementation of half-formed ideas. The job was to design and build a combination fax and optical scanning system known as Sierra, the aim of which was to transmit pages of mixed text and graphics at high speed via the trick of stripping off the text and sending it in compressed digital form, leaving only the graphics to be conveyed by conventional (and slower) fax. The entire page, it was hoped, would therefore transmit much faster than if faxed as a single coherent image.
Thanks to her big-iron training at IBM and hands-on experience at Memorex Conway was able to get the machine built in eighteen months, to everyone’s candid surprise. To their disappointment, it emerged as two gargantuan racks of special-purpose hardware that devoured so much power one could heat a building with it.
“You could make it, but you wouldn’t make any money off it,” she recalled wistfully. “It was such a giant, kludged-up thing with so many exotic little systems that all it demonstrated was that architects could envision and build useful systems that would take too much circuitry to be financially viable.”
Sierra would never be feasible as long as it came in such an unwieldy package. Intel’s new 4004, which packed thousands of transistors onto one chip—a full circuit board’s worth of her hard-wired machine reduced to something you could hold between thumb and index finger—provided Conway with the first hint of how the circuitry might eventually be reimplemented in a manageable package. The hint of a new class of architectures was somewhere inside there, whispering to her. “The itch,” she said, “was trying to be scratched.”
Doug Fairbairn, Mead’s second true believer, had arrived at PARC by way of the Stanford artificial intelligence lab, where he had worked with Kay and Tesler. “After getting my master’s at Stanford I’d gone to Europe,” he recalled. “After six months I came back. I wasn’t very driven to start a career but was thinking, what’s my next job? Then I heard about Xerox and thought, ‘If Alan Kay’s there, I bet I won’t have to wear a tie to the interview.’ And I didn’t.”
The interview was with Bill English. As usual, English was in desperate search of engineers to help him and Bill Duvall complete POLOS. Fairbairn spent three years entangled in POLOS hardware implementing the terminal system, which meant bringing together the TV display, keyboard, and mouse. (The ergonomic design of the latter consumed him for weeks. “I spent a lot of time on the cord. A normal cord would cause the mouse to move if you took your hand off. Then I found a wound cord that stayed put, but constantly unraveled. We ended up spending an incredible number of hours looking for the right insulated cord.”)
Bert Sutherland, who was more willing than Taylor to tolerate independent projects in his lab, but wielded an even more ruthless hatchet when they did not work out, canceled Sierra and POLOS within weeks of each other in 1975—the former because of its, impracticality, the latter because it was finally and unmistakably overtaken by the Alto. His two ace hardware designers were still looking for their next projects when Mead showed up a few months later. Whether it was their enforced idleness or their experience in building systems whose sheer size had gotten out of hand, both were captivated by his discussions of how to handle machine complexity.
“Lynn Conway and I,” Fairbairn remembered, “were the ones who said, ‘This VLSI is hot shit.’” They immersed themselves in the new technology, Fairbairn commuting weekly from Palo Alto to his parents’ home in Los Angeles so he could sit in on Mead’s classes at Caltech.
Mead was similarly seduced by PARC’s atmosphere of pure invention. Having spent years on campus and also been involved in commercial startups, he viewed PARC as a unique hybrid of both without the downside of either. “There was a lot more teamwork than in academia,” he said. “It was about getting things done, not about publishing papers.”
Nor was there the agitation to get product out the door he had observed at hard-charging enterprises like Intel. Instead he found himself surrounded by the enthusiasm for learning as an end in itself that drives people to come early to their labs and stay late into the night. Mead considered himself a pathologically early riser, but he could never remember a morning at PARC on which he was the first one in the building. “I’d get in at six in the morning,” he said, “and Alan Kay would already be there.”
He was even more profoundly impressed by the power of the integrated computing environment they had invented. “It was really obvious to me that this whole thing with the network and Altos and the file and printer servers was dynamite, and that it was going to be the way computing got done.”
For the next year Caltech and PARC educated each other. Mead transferred his theories about microelectronics and computer science, and Conway and Fairbairn paid him back by developing design methods and tools giving engineers the ability to create integrated circuits of unprecedented complexity on Alto-sized workstations. The science of VLSI was developing exactly as Mead had predicted. Systems that previously could be realized only as shaggy mats of diodes and wire strung on six-foot metal racks were getting reduced to filigreed etchings on the silvery surface of a silicon wafer—and they worked. They were approaching the goal of modularity, in which circuits that once required a square yard of schematic diagram could be reduced to a single compact chip for a computer designer to plug into a machine, like a simple building block from which a child can build a model skyscraper.
“This headed us in the direction of designing and building bigger, better, more elegant things,” Conway said. “Everybody’s ambition was cranking up month by month.”
They were a noisy group, given to loud and sometimes angry debates in the hallways that reminded people of a Dealer in full cry. Conway and Mead made for prickly teammates, sometimes collaborating, sometimes quarreling openly about how to organize and explain a technology moving ahead at breakneck speed. “Carver and I were both highly crazed by all this,” Conway recalled. “We’d compete and conflict with each other and there was so much noise around the project that it didn’t seem completely sane.”
VLSI also left some of their colleagues behind. The Computer Science Lab still held to the party line that VLSI was an untested technology and would remain so until there was proof the chips could be manufactured and exploited on a commercial scale. Mead was accustomed to such skepticism and on the whole untroubled by it. “At the time, the common wisdom was that if you make these things smaller and faster they’ll just melt,” he recalled. As early as 1971 he had written a paper predicting that the tiny chips would soon be part of every telephone, washing machine, and car. Nothing he had yet seen on the technical landscape suggested he should revise his opinion.
But Conway and Fairbairn were more sensitive to how their work was viewed by others at PARC. She felt CSL was not giving them the benefit of the doubt. “They didn’t seem to recognize that we were principled scientists who had our own self-check on things.”
She was right: CSL was profoundly dubious. “I didn’t like what Lynn Conway’s group was doing and I didn’t think it was very productive,” Lampson complained, troubled to see valuable PARC resources draining down a speculative rathole. Adding to the pain, Xerox was again tightening up the budget just as CSL was hoping to launch a few new initiatives.
“There was a zero-sum game in PARC resources and we thought there were all kinds of great opportunities for things we might do,” he recalled. “We wanted to get into databases and things like spreadsheets which we had completely ignored in the past. We wanted to do a lot of work on user interfaces and programming environments, all sorts of things. We did what we could, but it seemed clear that with more resources we could do a lot more.”
Conway started to feel that something had to be done to fight off CSL’s criticisms. Sutherland was a strong defender of her work, but by nature he was not a confrontational individual. If the computer lab—particularly Lampson, who commanded management’s respect—continued to carp at the money being spent on the hazy potential of VLSI, who knew how long she could survive at PARC? Especially since the power of the technology did not leap out at first glance. Compared to commercial integrated circuits, the schematics of VLSI looked simple and amateurish on the surface, because they employed novel, unfamiliar design techniques that had never been employed in building earlier generations of integrated circuits.
While discussing this one day with Mead and Fairbairn she realized the problem was not just scientific, but cultural. VLSI had not been around long enough even to generate textbooks and college courses—the paraphernalia of sound science that, she was convinced, would force everyone else to take it seriously.
“We should write the book,” she told Mead. “A book that communicates the simplest, most elegant rules and methods for VLSI design would make it look like a mature, proven science, like anything does if it’s been around for the ten or fifteen years you normally have behind a textbook.”
Mead was skeptical. They had no publisher and, given that they normally worked in two locations five hundred miles apart, no easy way of collaborating.
That’s where you’re wrong, she replied. What was the aim of all the technology that surrounded them at PARC, if not to facilitate just the project she was proposing? They had Altos running Bravo, a network to link long-distance collaborators, and high-speed laser-driven Dover printers to produce professional-looking manuscripts.
“With all that,” she said, “we can do it, and get it out there fast, and it’ll look just like a regular textbook.”
Their collaboration that summer on what became the seminal text of the new technology was only one of Conway’s efforts to distill and spread the VLSI gospel. The same year she agreed to teach a guest course at MIT (using the first few chapters of the still-maturing textbook), then printed up her lecture notes for instructors at an ever-enlarging circle of interested universities. By mid-1979 she was able to offer an additional incentive to a dozen schools: If they would transmit student designs to PARC over the ARPANET, PARC would arrange to have the chips built, packaged, and returned to the students for testing.
That summer her offer came to the attention of Jim Clark, then an untenured associate professor of electrical engineering at Stanford.
Clark had no prior expertise in integrated circuit design. “He’d never even worked in silicon before,” Conway recalled. But his expertise in computer graphics came from well within PARC’s universe: He had received his doctorate at the University of Utah, where his thesis advisor was Ivan Sutherland and his research funding had come from ARPA.
At Utah and later at Stanford, Clark was driven by the impulse to push the technology of graphics beyond the limits of existing hardware. As one of his Stanford students later recalled of a meeting in 1979, “The first day I went to speak to Jim, he pointed to a picture of an airplane he had up on the wall. ‘I’m going to make this move,’ he said.”
Like no one PARC had seen since Bob Metcalfe, he was also driven to explore all the commercial possibilities of his work, academia be damned. (“I love the metric of business,” he told an interviewer in 1994. “It’s money. It’s real simple. You either make money or you don’t. The metric of the university is politics.”)
Clark understood at once that the computing efficiency VLSI offered was the key to expanding the potential of computer graphics. That summer he essentially relocated to PARC, taking over a vacant office next door to Conway’s and steeping himself in VLSI lore. Within four months he had finished the Geometry Engine chip, the product of that summer’s total immersion.
Perhaps more than any other project, Clark’s chip fulfilled Conway’s quest to give VLSI credibility. Not only did it launch computer graphics as a profitable segment of high-powered computing, it showed that the unprecedentedly complex circuits could be designed rapidly, and then manufactured in huge quantities that would work flawlessly in an industrial context. The technology eventually matured into today’s generation of Motorola and Intel microprocessors, which run most of the world’s desktop computers, as well as a wide range of special-purpose circuits. Carver Mead’s prediction did come true. VLSI did turn every telephone, washing machine, and car—and thousands of other workaday appliances as well—into tiny computers. (Clark’s expectations were fulfilled too: Silicon Graphics Inc. made him a multi-millionaire.)
Carver Mead performed one more service for PARC after completing of the VLSI text with Lynn Conway. This was a visit he paid to Stamford to warn Xerox of the dangers of squelching the inventiveness at PARC.
The mission grew out of a conversation he had one day with Pake and Bert Sutherland. “I told them Xerox has got to get itself together,” he recalled, “because there’s no way a big company can take advantage of things moving this fast. People will get frustrated and start their own companies. Pake said, ‘You should talk to the people at corporate headquarters.’”
Mead’s familiarity with new-technology companies such as Intel won him a respectful audience from McColough and Kearns. “I spent a delightful morning with those guys,” he said. “I told them, ‘You’ll never have a better shot. If people leave because they don’t see anything happening, that’ll be like a bomb going off inside PARC. The only question is whether you participate and enable it, or let it happen for someone else.’”
“What do you suggest?” Kearns asked.
“Set up a venture capital arm,” Mead advised. “Smell out the technology, find it, incubate it. Take an equity position in things as they happen, otherwise it’ll all be gone and you won’t have any part of it.”
What he proposed would become standard operating procedure in American business twenty years later under the label “intrapreneuring”—a way to nurture innovation outside the dead hand of a corporation’s entrenched bureaucracy. In 1979, however, Xerox management regarded the concept as too elaborate to take seriously. That day over lunch Kearns confided to Mead that tradition’s hold on Xerox was almost too powerful even for him, its president and heir apparent to the chairmanship.
“Let me tell you a story about big companies,” Kearns said. Xerox employed a group of engineers to tear apart every new machine coming off the production line. Their goal was to figure out the most likely problems that would crop up under the stress and strain of daily operation, develop routines to fix them, and warn the design engineers of their mistakes. Yet every new model incorporated the same design blunders as the last. Finally the service engineers took matters into their own hands by designing their own machine. This was the Model 3100, a proposed desktop copier which, with its high reliability and decent resolution, was the closest thing Xerox could offer to compete with the Japanese models devouring its customer base. Yet instead of winning praise and rewards from the company, the bootlegging research engineers were widely vilified for interfering in the design process.
“You know what?” Kearns told Mead. “I spend most of my time trying to keep the rest of the company from killing those guys.”
Mead shook his head. No company so riven by tribal conflicts would ever bring itself to welcome the exceptional gifts of PARC. He returned to California with a mixed message, if a prescient one, about the likely fate of the powerful technologies he had himself used to such wonderful effect. He was sure they would sooner or later be developed and marketed for the world. But he was almost equally sure that when this happened, Xerox Corporation would be standing glumly on the sidelines.