By Bridgit Ekland
This article is from the June 15 and July 1, 2001, issue of Red Herring magazine.
Dave Patterson is a master at pulling off technology coups. In the early '80s, the University of California at Berkeley computer science professor invented a microprocessor design called RISC, which replaced long sets of processing instructions with smaller and faster sets. It's the engine driving many of the large servers in operation today. A short time later, he invented a more reliable -- and now ubiquitous -- data-storage technology called redundant array of inexpensive disks, which offers fast, reliable, cheap mass storage.
Now Mr. Patterson is on the verge of announcing yet another engineering feat. This time, he's going after the brains of computers with a chip design he calls intelligent random access memory (IRAM). Simply put, IRAM defies conventional computing economics by combining a microprocessor and a memory chip on a single piece of silicon.
After five years of work, the professor and his team of ten graduate students have handed a detailed design of IRAM to IBM (NYSE: IBM), which will fabricate the prototype chip. The plan is to begin testing the prototype this fall, in applications like multimedia and portable systems.
SPEED OF BYTE
Mr. Patterson's invention doesn't mark the first time engineers have tried to marry a microprocessor and a memory chip. Similar ideas, like graphics chips integrated with data-storage devices, have filtered onto the market, particularly in Sony PlayStations and set-top boxes. But these haven't approached the promise of the faster and more efficient IRAM chip. If the IRAM design takes hold in the chip industry, Mr. Patterson's invention may accelerate the market for a new generation of handheld computers that would combine wireless communications, television, speech recognition, graphics, and video games. "I believe in the post-PC era and the gadgets, cell phones, and PDAs," Mr. Patterson says.
One application Mr. Patterson has in mind is to leverage IRAM so that a handheld like the Palm can be used as a tape recorder with speech recognition and file-index capabilities. For example, the device would enable someone to locate and hear what a colleague, "John," has said about "computer privacy." The user would simply repeat those words, and the IRAM technology in the Palm would recognize the voice command and find the specific passages.
Mr. Patterson is essentially attempting to remove a thorn in the side of the microprocessor industry: the bottleneck that has long restrained processing speeds. Over the last two decades, the speed of microprocessors has increased more than 100-fold. But while memory chips, known as DRAMs, have kept pace in terms of capacity, their speed has increased only by about a factor of ten. As a result, microprocessors spend more time waiting for data and less time doing valuable computations. And as the gap between speeds grows, methods to help alleviate the problem, like memory caching, are being maxed out.
Mr. Patterson believes that any IRAM-designed microprocessor could potentially access memory 100 times faster than is currently possible. The performance in, say, a wireless device would be comparable to that of the average PC. Eventually, other chip giants will try to place microprocessors alongside DRAM on a single chip. But Mr. Patterson is impatient; he thinks it can be done now, intelligently, and with benefits far greater than others might imagine.
Mr. Patterson, a former college wrestler who bench-pressed 350 pounds on his 50th birthday, isn't one to shy away from tough odds. While the expected payoff is a faster chip, IRAM is a huge gamble, given that Intel (Nasdaq: INTC) chips dominate the market. Another hurdle: if Intel or any other company adopts the new chip, engineers would have to learn how to program it. "Almost all of these new types of media processor-kind of chips are difficult to develop software for," says Pete Glaskowsky, senior analyst for MicroDesign Resources, a market research firm. "They all say that their product is easy to develop software for. I've learned to discount that statement."
If history is any indicator, IRAM might level the playing field, just as RISC once allowed Sun Microsystems (Nasdaq: SUNW) and IBM to challenge Intel.