Professor's Idea for Speedy Chip Could Be More Than Academic

By DEAN TAKAHASHI 
Staff Reporter of THE WALL STREET JOURNAL
August 28, 1998 pages B1, B4

August 28, 1998 As a young computer-science professor at the University of California at Berkeley, David Patterson championed the cause of a radical approach to designing computers and chips that simplifies their internal tasks. His concept, reduced instruction set computing, or RISC, transformed the chip industry and became a mainstay for several big computer companies.

Then Mr. Patterson was a force behind a new alternative to big, costly disk drives, dubbed RAID, or redundant arrays of inexpensive disks. RAID-based products are now commonplace.

Now the professor has another pet project; and, given his track record, Silicon Valley is paying heed. Mr. Patterson, with a handful of graduate students, is laying the groundwork for a new type of hybrid computer chip, breaking from a tradition that has shaped the balance of power among semiconductor companies. The new idea, aimed at fueling a new generation of powerful hand-held computers, also puts Mr. Patterson at odds with an old and powerful nemesis -- Intel Corp.

 

"'Like Hell You Can't' "

Some people say we can't compete with Intel," says Mr. Patterson, a weight lifter and former wrestler who delights in confrontation. "I say, like hell you can't. We'll do something more interesting than them with a dozen students."

The professor wants to produce a single piece of silicon that combines the two most important and lucrative types of chips: microprocessors, which act as the brains in personal computers, and the data-storage devices known as dynamic random access memory, or DRAMs. The idea, which he calls "intelligent" random-access memory, or IRAM, could help clear a bottleneck that has long slowed down chip speeds.

The problem: The speed of microprocessors increases 60% a year, but the time it takes them to retrieve data from separate memory chips increases only about 10% a year. That's like having a record-setting drag racer that runs out of gas every 50 feet. If the circuitry holding data were to be located on the same chip, a microprocessor could theoretically access memory 100 times faster.

"Theoretically" is a key word. Mr. Patterson, who has yet to make working IRAM chips, must prove that they can justify changing billion-dollar chip factories around the world.

Mr. Patterson recalls some "very negative" reactions from executives at Intel, the world leader in microprocessors. Sunlin Chou, a manufacturing vice president at Intel, says combining microprocessors and DRAMs on the same chip results in a "more complicated and lot more expensive" manufacturing process and a different set of machinery. "It's a nightmare process with many more steps to it," Mr. Chou says.

Yet, in a testament to Mr. Patterson's clout, Intel is among a group of companies that is putting up $1 million annually to fund the professor's IRAM research. The others are LG Semicon Co., Microsoft Corp., Mitsubishi Electric Corp., Micron Technology Inc., Silicon Graphics Inc., Sun Microsystems Inc., International Business Machines Corp. and Hitachi Ltd. Most see the project as a long shot but aren't ready to bet against a man in the forefront of two earlier revolutions that succeeded.

"He is the best at organizing teams of students and having major successes with real-world projects," says David Ditzel, a former student who is now chief executive officer of Transmeta Corp., a chip designer in Santa Clara, Calif.

John Hennessy, dean of the College of Engineering at Stanford University and another influential RISC advocate, thinks Mr. Patterson's style will help advance his latest project. "A new technology always needs a champion because it's so easy to kill ideas," Mr. Hennessy says.
 

Autograph Requests

The two men collaborated on a book on computer design that turned into a bestseller on college campuses. Even now, Mr. Patterson says he gets requests for autographs and nit-picking electronic mail challenging some ideas in the book and its follow-up.

Mr. Patterson says he honed skills of explanation and debate around the family dinner table. The oldest of four children, he would often explain homework problems to his younger siblings. His competitive nature spawned an interest in weight lifting (he bench-pressed 325 pounds on his 50th birthday) and his participation in college wrestling.

He began writing about IRAM in 1990 but took it more seriously after an article he wrote in 1995 prompted his Berkeley students to demand a class on the subject. The class drew wide industry notice since notes about it were posted on the Internet.

Mr. Patterson is betting, in part, on the hope that makers of memory chips are hurting so badly from a global slump they might be willing to try something radical to siphon off some of Intel's huge profit.

"Dominant companies usually don't change unless they're forced to do so," Mr. Patterson says. "I'm looking for the company that is going to conclude, 'Damn, this is something that we'll have to do or someone else will.' "

Similar ideas already have filtered into the marketplace. Some makers of graphics chips, which are composed of tiny circuits that are similar to those found in microprocessors, have already made some chips that are combined with DRAM.

"There are a lot of hurdles" for IRAM, says Thomas Liao, a vice president at Mitsubishi, which sells graphics chips combined with memory. "We are going in the same direction, but Mr. Patterson is more aggressive ... because he is a researcher."

Rather than take on Intel's dominance of desktop machines, Mr. Patterson believes IRAM could accelerate the market for a hand-held computer combining television, wireless communications, speech recognition, graphics and video games in a single device. He says it should be possible to cram 32 megabytes of memory into an IRAM microprocessor, as much as today's average PC.

Intel, of course, could eventually co-opt the IRAM movement by adding memory to its own chips. To neutralize the challenge from RISC chips, Intel adopted some of its competitors' design techniques while investing heavily in manufacturing advancements that helped it stay abreast in the speed race.

Mr. Patterson is still satisfied that his ideas in semiconductors, as well as data storage, pushed technology forward. That these same ideas didn't make him a wealthy man but enriched entrepreneurs in several start-ups is as it should be, he believes.

"The money has gone to people who took the ideas and did something with them," says the professor, who is married and has two adult children and a grandson. "I picked personal happiness over wealth."

Still, Mr. Patterson's work hasn't been without financial rewards. After working as a consultant for Sun, he received a generous amount of stock in the Palo Alto, Calif., company. He won't say exactly how much stock he got but says it will pay for his retirement.