Computer Researchers Look Past Silicon

Signal, Jun 2008 by Ackerman, Robert K

The crystal ball is cloudy when it comes to the nature of future chips.

Scientists are racing against the clock to develop a means of defeating an enemy that threatens to stop computer technology progress dead in its tracks. The threat is not terrorism; it is Moore's Law, as complementary metal oxide semiconductors are nearing their size and performance limits as defined by the laws of physics.

Many computer experts have been predicting since the last millennium that the technology underpinning the information revolution was nearing the end of its progressive development. Scientists have been able to prove those earlier prognostications wrong as silicon chip density and power continued to increase geometrically. Now, however, researchers are seeing the beginning of the end as physical limitations are becoming evident. Within 10 to 15 years, some experts say, the doubling of computing power every 18 to 24 months with complementary metal oxide semiconductor (CMOS) chips will come to an end.

Instead of trying to fight nature, scientists are exploring new materials and techniques that could be exploited to produce new generations of processing chips. No clear contender as an heir to silicon has emerged yet, as researchers must consider scalability, fabrication and economics in their quest to find the next type of microprocessor.

This may involve changing the nature of the processor itself. Existing CMOS devices feature thousands or millions of charge-based transistor switches on a single chip. Future devices may incorporate spin-based, optical or molecular switches.

But none of these technologies is likely to pan out in the near future, warns Shankar Basu of the National Science Foundation (NSF). These are long-term research projects, says Basu, program director, Computing and Communication Foundations (CCF) division, Computer and Information Science and Engineering (CISE) directorate at NSF. He points out that the computing industry cannot wait for these blue-sky technologies to become economically viable, so near-term research aims to exploit CMOS technology by enhancing its products.

One near-term fix may be multicore chips. Describing it as a band-aid solution to the problem, Basu nevertheless says that it might permit growth in processing power by embedding many processors in a single chip. These multiple processors would run in parallel on silicon CMOS.

Many existing consumer computers have two or four cores on a single chip. The near future may see 64 cores on a single chip, reports Charles Ying, program director in the Materials Research division, Mathematical and Physical Sciences directorate at NSF.

However, Ying warns that programming challenges begin to emerge when that many cores are structured to operate in parallel. Two cores can split tasks between them pretty easily, but many more cores cannot divide tasks quite so easily. That programming problem is a hurdle that may be more difficult to overcome than the fabrication challenge. It will require a new form of theoretical thinking, Basu adds. "It opens up a new can of worms even for theorists," he says.

"Parallel computing was the 'in' thing to do many years ago," Basu relates. "It did not quite pan out, but maybe this time it will because the hardware is supporting it."

Because the issue threatening to derail chip progress is centered on its underlying makeup, one of the key NSF research areas focuses on chip materials. The foundation's Materials Research division works with its NSF Engineering directorate counterpart, the Electrical, Communications and Cyber Systems (ECCS) division, which concentrates on the device and systems levels. As materials research bears fruit, the systems division works to escalate that technology into a full device.

Ying analogizes that CMOS is the ground floor of a hundred-story building. Each chip has many layers atop that CMOS material, and technology advances may occur at many of those levels. Many issues involving power consumption, leakage and heat generation confront researchers.

Small-scale improvements already have taken place, such as replacing aluminum wiring with copper, which conducts heat much better. Another innovation replaced the silicon gate material with hafnium, which reduces current leakage. IBM and Intel last year announced the development of hafnium transistors, which Intel is including with some of its next-generation multicore processors. IBM also has developed a way to replace silicon material between wiring with polymers, which also will reduce heat generation.

Power management in that hundred-story structure might help control heat, Basu points out. If one part of the processor begins to overheat, its power might be cut and functions distributed elsewhere in the device. Sophisticated techniques such as economic game theory are being applied to processor power management, he adds.

Superconducting materials do not offer much promise in chip technology, but fast conducting materials have potential, Basu allows. Electrical resistance contributes to the heat generated by a computer, and this heat threatens to derail computer operations. Room-temperature fast conducting materials have lower electrical resistance and thus generate less heat.