The technology, called "proximity communication," aims to let one chip transmit signals directly to another next to it, instead of through the tangle of pins, wires and circuit boards employed today. If successful, the technique could greatly alter many aspects of computer design.
Performance, for instance, could greatly escalate because the speed of transferring data among chips and the number of channels for the transfers would increase. Energy consumption could also decline. Just as important, overall costs could fall, because defective chips could be removed like Scrabble tiles.
"There is a huge need for higher-bandwidth kind of chips," Robert Drost, a senior researcher at Sun Labs, said at an open house last week. "Rather than have the chips soldered onto a printed circuit board, the printed circuit board is taken out of the system."
The technology is an integral part of a supercomputer Sun is creating for a supercomputer bake-off sponsored by the Defense Advanced Research Projects Agency (DARPA). The government agency wants to see US manufacturers establish a new generation of supercomputers by 2010. IBM and Cray are also building supercomputers in the project. In about two years, DARPA will select one of the three entrants to develop its machine more fully.
Although the performance of processors has steadily increased in the past 20 years, the performance of the input-output paths that connect these chips to the rest of the computer hasn't, resulting in well-documented bottlenecks.
In part, the problem arises from how the chip connections are built. Processors are inserted into a package that contains metal pins. The pins secure a chip to a board and create the electrical connections.
Unfortunately, the pins are rather large to be electrical and mechanical devices. Only a few hundred fit on a package that contains a processor with several million transistors. Bandwidth, thus, is constrained.
"It is not that on chip wires are evil. It is just that they are large," Drost said. "The number of pins to get to the outside world has gone up only 5 (percent) to 10 percent a year."
By contrast, proximity communication relies on capacitive coupling -- the ability of two electrically charged devices close to each other to interact. Transmitters on one chip can send signals to another. These signals are then amplified. A much higher number of transmitter/receiver pairs than pins can be inserted in a specific area, which allows for more simultaneous connections.
"There is a small, but noticeable voltage change in the receiver level," Drost said.
The technique could also allow designers to remove the cache -- the large pool of memory currently found on the processor -- and put it on a separate chip. Caches were integrated onto processors to amplify bandwidth. Adding cache, however, bumps up manufacturing costs, as it greatly increases the number of transistors. With the bandwidth constraint gone, caches could once again be made independent without it having an impact on performance.
One of the chief difficulties in developing the technology comes from the environment where computer chips live. Heat and vibration in this environment can cause chips to get out of the precise alignment needed for proximity communication. Sun is currently tinkering with different techniques and different packages to prevent, or correct, these effects.
"There are some very interesting technical problems that must be solved, such as, how do you maintain alignment?" said Nathan Brookwood, an analyst at Insight 64. "But if it does pay off, it could be quite revolutionary."
