IBM, Stanford craft next-generation chips
- 28 April, 2004 07:24
IBM and Stanford University have announced a new research group dedicated to the emerging science of spintronics, with the goal of creating prototype CPUs that complete computations through magnetism instead of today's electrical charge.
"We're trying to do something that could be as significant as the launch of the transistor 50 years ago," vice-president of IBM's personal systems and storage, Robert Morris, said.
He directs the Almaden Research Center here, where Big Blue and Stanford luminaries announced the new joint development effort.
The IBM-Stanford Spintronic Science and Applications Center (SpinApps) could be key to future advancements in processor technology, according to Stuart Parkin, an IBM Fellow and manager of the Magnetoelectronics group at the center.
Processor manufacturers had traditionally increased CPU performance by shrinking a chip's circuitry so it can run faster, he said. This method had dramatically increased computing power over the past few decades, but the process was hitting physical limitations. Specifically: The necessary electrical charge causes those faster chips to get too hot to handle.
Processor giant, Intel, has noted the growing problem of heat in recent years, too.
Parkin and his associates hope spintronics will help create ever-more-powerful processors, so manufacturing roadblocks won't slow the pace of technological evolution.
Today's processors use an electrical charge to create on and off states. A processor based on the principles of spintronics could control the spin (or magnetic orientation) of electrons and create two possible states: up or down. The result: atomic-size structures that offer enormous computational capabilities while generating very little heat.
Parkin estimates that current processor manufacturing technologies could yield five to ten more years of performance improvements. However, that's not much time to develop a new way to build CPUs.
"It takes a long time to go from theory to product," Parkin said.
In fact, that's about how long it had taken another spintronics-based technology to go from idea to product, he said.
Magnetic random access memory (MRAM) had been in the works for more than nine years. Several companies were now readying products that use the technology, and could ship them as early as next year.
Cheap, high performance, and non-volatile, MRAM represents a dramatic improvement over today's two most common memory standards: dynamic RAM (DRAM) and static RAM (SRAM, or flash memory), Parkin said.
He likened today's DRAM to a leaky bucket that must be constantly refreshed to maintain its contents.
Meanwhile, SRAM doesn't leak, but it must be a much larger bucket. MRAM is like a small bucket that doesn't need constant refilling. IBM and Infineon Technologies announced in 2000 plans to co-develop MRAM products.
While CPUs and MRAM are still on the drawing board, IBM has already introduced its first spintronics-based product. In fact, it appeared back in 1997 as the first hard drive to use the giant magneto-resistive (GMR) head. The GMR technology brought about a 40-fold increase in data density over the past seven years, and helped fuel the massive growth in hard drive capacities, according to IBM representatives.
By enabling manufacturers to create large, cost-effective hard drives, spintronics technology had helped the Internet expand, Parkin said. With lots of cheap storage available, the Web has had more room to grow.
Parkin and his team - made up of more than 25 research specialists including Stanford professors, graduate students, postdoctoral researchers, and IBM employees - clearly have lofty goals for this emerging technology.
In his closing comments at the announcement, IBM's Morris outlined the technology's potential to impact everything from computational computing to storage.
He said, in what could turn out to be an understatement, "we may be on the verge of something extremely important."