September 2001 |
||
 |
|
|
|
||
|
||
|
||
|
||
|
||
|
Written by Kris B. Mamula Photograph by Michael Olfert |
|
Something was going on in a storage room at Intel’s Jones Farm campus in Oregon. But what? For months, five computer engineers showed up at the room every day. One of them would slip a bent wire into the jamb to unlock the door. The others then piled inside, dragging with them a few borrowed marker boards. A security guard once warned them to be careful. Stuff was piled so high they could get hurt in there, he said. There were no phones. No windows. No distractions. Just five gifted engineers and lots of blue-sky imagination. Few others knew in the early days of 1991 when the meetings began, but the group was dreaming up a way to make computers go faster than ever before. Sometimes it got a little heated. “If someone had looked in,” says Michael Fetterman, one of the five, “they would’ve thought we were going to eat each other.” They broke it off after two hours each day. The rest of the day was needed to copy the scrawl from the marker boards, unwind, get ready for the next round. Steering these meetings was Pittsburgh native Robert P. Colwell (Engineering ’77). Hired by Intel just a few months earlier, Colwell was assigned to oversee the development of a new computer chip. Chips, of course, are the brains of the computer, processors that control virtually every computer operation. Colwell’s job was to make a faster chip. A faster chip means computers can better handle memory-intensive files like graphics. Speed would become even more important in the next few years as the use of the Internet became more widespread. Dubbed P6, Colwell’s new chip was designed to be twice as fast as the company’s ground-breaking Pentium. The size of a postage stamp, the Pentium boasted the computing power of 3.1 million transistors—a thousand times as many transistors as Intel’s first chip in 1971. Although it wouldn’t be unveiled until 1993, the team knew that the Pentium could outrace everything before it by a factor of two times. Their P6, which would be marketed as the Pentium Pro, would be twice as fast as that. The storage room was an idea factory. Fiery arguments are part of what Intel rather delicately calls its culture of “constructive confrontation.” But these engineers were a breed apart, even for Intel. Fiercely independent, they believed that innovation bubbles up from the bottom, that work should be directed by those doing it—not corporate bigwigs. The group’s reputation as “cowboys” suited the plainspoken Colwell just fine. “Who else can better see how to get there?” says Colwell, the son of a milkman who worked his way through Catholic high school typing in the school office. “Tell us where to go, and get out of the way.” David J. Papworth, another member of the team, agrees, saying Intel had the “substantially good judgment to leave the group alone." So the team huddled every day in a storage room crowded with the walls of unassembled office cubicles. Fetterman and Andy Glew were fresh-faced kids barely out of college. Papworth and Glenn J. Hinton were old hands. So was Colwell. In his pre-Intel days, Colwell kept a personal computer on his nightstand so he’d never be far from his work. During those years, Colwell says he learned humility. It’s so easy for an engineer to promise more than he can deliver. He learned something else before Intel. It became his mantra: If you haven’t tested it, then it doesn’t work. Period. Colwell was single-minded, reining in the guys “way out there on the bleeding edge.” But there is more to Colwell than single-mindedness, according to longtime friend and Pitt classmate Kaigham J. Gabriel. There’s a maverick who is not afraid to speak his mind. “There is nothing that’s going to own 100 percent of Bob Colwell,” Gabriel, now a professor at CMU, says. As senior Pentium Pro architect, Colwell was also careful to separate the work from the worker. When problems arose, he was quick to remind his team that the “problem was in the code, not the person,” says Donald Parker, who joined the Pentium Pro team when it was a year old. Colwell says his strength was in distilling what was valuable during the “anything-goes phase of design.” Colwell is quick to credit coworkers: “I had raving geniuses at what they do.” “The experiment was a success,” says Papworth in what is perhaps the biggest understatement of the decade. The Pentium Pro went on sale in November 1995. Four-and-a-half years had passed since the first brainstorming session in the storage room. The Pentium Pro was the best the company had achieved on silicon. It was also a commercial hit. For a while, the chip was selling at an astounding $200 over list price as the company struggled to keep up with roaring demand. Intel’s earnings jumped 41 percent in the first year of sales. Sales of the Pentium Pro grew faster than any previous Intel processor, partly spurred by things outside the company’s control. Browsing the Internet didn’t exist as we know it when the Pentium Pro team was formed in 1990. Netscape introduced its popular browser late in 1994. By August 1995, some 18 million people in North America were surfing the Net. Over the next eight months, that number exploded to 34 million. At the same time, personal computer sales boomed— up 26 percent between 1994 and 1995 to 60.2 million worldwide. The Pentium Pro became the baseline for Intel’s processors for the next six years—an eternity in the world of computing. In recognition of his work, Colwell was named an Intel Fellow in 1996. A year later, Colwell bought a sleek, Oxford blue BMW with a bonus he received for his Pentium Pro work. Life was good. From the dawn of the wire and bead abacus to the release of Intel’s screaming Pentium 4 last fall, speed has been the one constant goal in computing. The race for speed developed a wrinkle in 1971 when Intel introduced the first computer-chip processor. Suddenly, size mattered. The Intel 4004 was as powerful as the US Army’s giant Electrical Numerical Integrator and Calculator (ENIAC), built in 1946 in Philadelphia. ENIAC was no slouch. It was a thousand times faster than anything before it. But when this 30-ton hog was turned on, it caused brownouts in the city. Intel’s first chip burned about as much electricity as a tiny blinking Christmas tree light. What’s more, it was smaller than your thumbnail. The difference? ENIAC was powered by 18,000 vacuum tubes. The Intel 4004 used 2,300 tiny transistors on a silicon chip. For decades, processors have been doubling in speed about every 18 months. At the same time, they’ve been getting smaller and smaller. Everyone agrees the trend will end, but no one knows exactly when. In the meantime, the question isn’t getting a whole lot of attention. Worldwide processor sales totaled $33 billion in 2000 as chipmakers like Intel, the biggest in the world, firehosed the market with faster and faster chips. The larger semiconductor industry, which didn’t exist before 1950, employs a medium-sized city population of 284,000 today. Worldwide sales of silicon chips totaled $204 billion last year. Few disciples of processor speed were more devout than Colwell. He was already a veteran in the field when he arrived at Intel’s Jones Farm complex in 1990. Within two months, Colwell, and chip architects Papworth and Hinton had settled on the “out-of-order engine” as the way to approach their quest for a faster computer chip. In theory, the engine boosted computer speed by working ahead, doing the immediate task while also skipping ahead and following instructions that had to be done in the next few seconds. This was new; the idea had never been proven. Colwell and the others decided that the engine would become the brain of the Pentium Pro. The new engine promised unheard of speed. Still, there was confusion over exactly what Colwell was cooking up at Jones Farm. “When I see an ‘out-of-order’ sign on a pop machine,” one manager once told Colwell, “I don’t put my money in.” Tongue-in-cheek, Colwell says that “out-of-order engine” sounded better than “disordered engine.” The Pentium Pro was certainly speedier than anything before it. But it was also completely different. So different, in fact, that many people inside Intel doubted whether it would work. Outside Intel, previous tries at making an out-of-order engine had foundered. “There was very much a dare in the air,” says Fetterman. “It was ‘us against the world’—that’s what it felt like,” says Hinton. “Everyone viewed this like a moon shot.” Colwell loves analogies. Here’s one he uses to explain why the Pentium Pro was so fast. Think of the processor’s job as a trip to the grocery store, he says. The idea is to pick up what you need—comparable to the fetching of instructions for the processor—then check out, which is like a processor doing the tasks that it’s told to do. But the Pentium Pro was like having three shoppers with one grocery list. The earlier Pentium had only two carts, tied together. It also had a helper who looked ahead to the next item on the list, and pointed to the aisle. But the Pentium Pro shoppers figured out the most efficient way to get around the store before they even got started. They moved independently, and each one could read ahead to see what was needed next. What’s more, Pentium Pro shoppers got help from several clerks. For instance, the clerks guessed what the shoppers were going to buy, based on past purchases, then pulled these items from the shelves and tossed them into the carts as the shoppers passed. If the clerks guessed wrong—no problem. The items were simply discarded at the check out. And unlike the Pentium, the Pentium Pro didn’t have to wait for some bit of information—an out-of-stock item—fetched from the main memory. Instructions could forge ahead with data from the processor’s cache. Imagine the clerks dropping out-of-stock items into the carts as soon as delivery trucks bring them. Colwell sees his job as part artist, part urban planner. Chip architects decide the best places to put things like warehouses and power plants in cities no bigger than a credit card. They take the big view of some really small places. Art has long influenced science. Galileo’s study of perspective drawing, for example, led him to the conclusion that the moon was mountainous because of the shadows he saw. So, too, with chip architecture. “You don’t just want a pure engineering solution,” says Atiq Bajwa, who joined the team in 1990, “but an elegant solution as well. And elegant solutions tend to work better. Good technical solutions are aesthetically pleasing at an intellectual level.” Papworth compares his work to that of a building architect. “From an architect’s point of view, we estimate approximately what can be created from the materials we have,” he says. So it wasn’t enough that the Pentium Pro would work well; it had to be “graceful” and “elegant.” Colwell uses these words a lot when talking about chip design. Colwell also describes himself as a “paranoid engineer.” (Maybe it’s catchy: Former Intel CEO Andy Grove, whom Colwell admires, wrote a best-seller in 1996 called Only the Paranoid Survive.) What’s more, Colwell brought an unusual talent to the task. Gabriel says Colwell has an innate understanding of electrical circuitry. Growing up, Colwell once wired the family’s Christmas tree to flash eerily and emit an alien howl when someone walked by. The first person to activate it was his mother, a bank manager, who arrived home late one night from work. “She knew instantly who did it,” Colwell says, laughing. To many, computer engineers are a “bunch of drones turning a crank,” he says during a more serious moment. “But an artist would recognize much of what a chip engineer does. There is a consistency, a wholeness to a bridge or house—or computer chip—that is well designed and constructed.” Jones Farm is a patch of faceless, yellow buildings in Hillsboro, about 18 miles west of Portland. Inside the complex is a blur of blue office cubicles. Intel offices worldwide have the same numbing design. Still, creativity squeezes through. During the Pentium Pro years, Colwell assigned people to cubicles arranged to mirror functions on the new chip. Question about the chip’s memory or processor? Finding the person who had the answer was just a matter of walking to the right part of the floor. That only worked for a little while though. The sheer size of the team, which eventually grew to over 200 people, forced people into increasingly remote places. Colwell’s cubicle has always been stuffed with books. A favorite is a thick volume about the space shuttle Challenger, which vaporized in a ball of flames 73 seconds after liftoff on January 28, 1986. Every Challenger part was built to required specification. Rocket assembly at the pad went smoothly. But those who gave the go-ahead for the launch didn’t have all the facts. They didn’t know about the problems with rocket seals. Missing was the big view. Colwell has learned well the lessons of the Challenger. He never doubted whether the Pentium Pro would work. But the testing he demanded bordered on obsessive, others on the team said. Colwell counts Western Pennsylvania native and bridge builder Washington A. Roebling among his heroes. Obsessed with being intimately involved with his work, Roebling nearly died in 1872 while working deep in the foundation of his biggest project—the Brooklyn Bridge. Forty-six year old Colwell is a fast walker. He is tieless this day, wearing black rubber-sole shoes that hush his footsteps. Since development of the Pentium Pro, Colwell has a new idea about what speed should mean. Sure, computers should go faster, says Colwell, an affable man with a dry sense of humor. But the industry has painted itself into a corner by chasing processor speed at the cost of just about everything else. Industry is not entirely to blame. Long ago, the public grabbed onto processor speed as an easy yardstick for quality. Faster equals better. Industry’s response was zippier chips. The paradox, Colwell says, is that faster processor speed has not always translated into faster, more reliable computers. Take booting up, which happens when your computer is switched on. Colwell is infuriated by the warm-up wait, which he calls “hourglass time.” “That’s my time being wasted there,” he says. “I find it particularly grating.” Computer reliability, too, has lagged. When was the last time you heard someone use the word “crash” to describe an automobile accident rather than a broken computer? In 1965, Intel co-founder Gordon Moore predicted that processor speed would double about every 18 months. The prediction has proven remarkably accurate. But in recent years, stress cracks have developed in Moore’s Law. “As the sizes involved in chip designs close in on atomic measurements,” the 72-year-old Moore said in announcing his retirement in May, “it will become harder to innovate.” Colwell’s forehead creases into four neat wrinkles. That day arrived some time ago, he says. Colwell is greeted like a long lost friend as he walks through Jones Farm. He smiles and nods, occasionally shaking hands, his grip sure. His fingers are those of an accomplished guitarist, an instrument he has played since age 6. He plays classical and steel string guitar at the Catholic church he attends Sundays with wife, Ellen, a 1978 Pitt computer science graduate, and the couple’s three children. “He has a better ear than I do,” says Colwell’s brother, Denis, who is music director of Pittsburgh’s River City Brass Band. Colwell stops outside a dark, first-floor room. This is where it all began, he says softly. In 1990, the hubbub was down at company headquarters in Santa Clara, California, where work on the Pentium was going well. Oregon—well, who really knew what those cowboys were up to? At first, being ignored by the company brass made the team feel slighted, Colwell says. Soon, they came to appreciate their good fortune. They were left alone to dream and design. Colwell tries the door. Locked. Wrinkles crease his forehead. It was always locked back then, too, he says. But you could count on Papworth to jimmy the lock in no time. Not today though. In February, Colwell came back to Pitt’s campus to be honored at the annual Honors Convocation as one of the 2001 Distinguished Alumni Fellows. Speaking to a standing-room-only audience of Pitt engineering students, he talked about what he sees as “trouble in paradise.” He began by relating to the crowd an old children’s story about Mike Mulligan, the steam shovel operator. Determined to prove his steam shovel can compete against more cost-effective gasoline and diesel earthmovers, Mulligan offers to dig a foundation for a town hall for free. But in his fervor to prove himself, he forgets to plan a way to get his steam shovel back out of the hole. “This is really an insidious trap to fall into, even for guys who make a lot of money,” Colwell says, bringing the point around to the chase for computer speed. “The idea is to reach the people who’ve never seen a personal computer, not just make processors faster.” Some users will always need more speed, but not a majority. What’s more, faster processors have created an “ugly looking curve” in increasing chip problems. (Last year, computer bugs forced Intel to recall a version of its Pentium III chip, and nearly one million motherboards using its 820-chip set.) “We don’t have any big ideas,” an emphatic Colwell told the engineering students. “We have 50 small ones. And they have to work together. That’s a nightmare. We’re into diminishing returns—big time.” Later, thinking of the students and his role in making computers go faster, he says, “I wonder if they realized the irony.” It’s Friday afternoon, and Colwell is headed home. The fragrant warmth of late spring washes through the BMW’s sunroof. The consummate engineering team player listens to a solo steel guitarist on CD. The engine murmurs, hinting at its thundering power. Traffic cops love these kinds of days. Open road. The start of the weekend. People behind the wheel. Speeding. But not Colwell. He’s driving carefully, working through the gears well under the posted limit. The push for processor speed that began 11 years ago for him has come full circle. Instead of desktop things wired to keyboards and TV screens, chip architects should start with everyday people, Colwell says. Plumbers. Senior citizens. Barbers. Soccer moms. How can silicon better each of their lives? What about a headset and pocket-sized device that amplifies sound for the hearing impaired? Imagine this: Chip technology is out there to allow users to “replay” part of a conversation they missed, Colwell says. Or, how about a pager-like tracking device for children so parents could know where they are at any given time? The man whose life has been spent souping up the tiny motors that power personal computers now questions where his industry is going. “When you step back from the computing industry and try to see the big picture,” Colwell says, “you realize that what needs fixing isn’t the speed of the computer.” Maybe it’s time that personal computers stop driving the industry, he says. “What would happen if we quit thinking of them as computers?” It’s a good question. And you get the feeling that Colwell is going to be in on whatever answer comes up. Look Mom, no keys You walk to your car. The door unlocks just as you get within a few feet of the vehicle. No fumbling for keys. A sensor inside the door recognizes a tiny computer chip that’s attached to, say, your wristwatch. Or a piece of jewelry. Getting Small For years, the tiny motors that power computers have been getting faster and smaller. Now, Pitt engineering students have the opportunity to make these electronic gadgets out of silicon wafers in the School of Engineering’s new Micro-Electrical Mechanical Systems (MEMS) Lab. How about a device the size of a single grain of rice that controls when and how much insulin a person with diabetes receives? How about a mechanism the size of a pollen grain that would keep time, making quartz-powered wristwatches as obsolete as wind-ups? These are among the tantalizing possibilities for micro-electromechanical systems. Here’s how they’re made. In the sixth floor, Benedum Hall lab, various coatings will be layered onto silicon chips. Contamination-free suits are mandatory to keep the chip from being ruined by dust. The chips and coating will then be programmed for a specific purpose. “What you can use these things for is almost limitless,” says Michael Lovell, associate professor of mechanical engineering. Bioapplications, such as the administration of medicine, will be the lab’s focus, Lovell says. The new lab is one of four research facilities that make up the Swanson Center for Product Innovation. —KM |
Pitt Magazine Home | Past Issues | Contact Us | Top of Page