TR10: Probabilistic Chips
(Page 2 of 2)
Applications where the limits of human perception reduce the need for precision are perfectly suited to PCMOS designs, Palem says. In cell phones, laptop computers, and other mobile devices, graphics and sound processing consume a significant proportion of the battery power; Palem believes that PCMOS chips might increase battery life as much as tenfold without compromising the user’s experience.
PCMOS also has obvious applications in fields that employ probabilistic approaches, such as cryptography and machine learning. Algorithms used in these fields are typically designed to arrive quickly at an approximate answer. Since PCMOS chips do the same thing, they could achieve in hardware what must be done with software today–with a significant gain in both energy efficiency and speed. Palem envisions devices that use one or more PCMOS coprocessors to handle specialized tasks, such as encryption, while a traditional chip assists with other computing chores.
Palem and his team have already built and started testing a cryptography engine. They are also designing a graphics engine and a chip that people could use to adjust the power consumption and performance of their cell phones: consumers might choose high video or call quality and consume more power or choose lower quality and save the battery. Palem is discussing plans for one or more startup companies to commercialize such PCMOS chips. Companies could launch as early as next year, and products might be available in three or four years.
As silicon transistors become smaller, basic physics means they will become less reliable, says Shekhar Borkar, director of Intel’s Microprocessor Technology Lab. “So what you’re looking at is having a probability of getting the result you wanted,” he says. In addition to developing hardware designs, Palem has created a probabilistic analogue to the Boolean algebra that is at the core of computational logic circuits; it is this probabilistic logic that Borkar believes could keep Moore’s Law on track. Though he says that much work remains to be done, Borkar says Palem’s research “has a very vast applicability in any digital electronics.”
If Palem’s research plays out the way the optimists believe it will, he may have the rebel’s ultimate satisfaction: seeing his heresy become dogma.