By now, most of us are tolerably familiar with the basics of how computers process information, and with the idea that numbers and other bits of information are represented as a pattern of 0s and 1s. You may be watching a YouTube video on your PC, but deep under the covers, the processor(s) involved are just carrying out a lot of arithmetic and logical operations on a string of bits (from binary digits).
Now, however, an article in Technology Review discussed a new type of processing chip that has been developed by Lyric Semiconductor, based in Boston. The company has developed what it claims is the first probability processing chip. Instead of using solid-state components to implement digital logic operations, Lyric’s chips use them to represent Bayesian probabilities.
Whereas a conventional NAND gate outputs a “1” if neither of its inputs match, the output of a Bayesian NAND gate represents the odds that the two input probabilities match. This makes it possible to perform calculations that use probabilities as their input and output.
The company, which was founded in 2006, has mostly worked out of the public eye, with most of its funding coming from DARPA [the Defense Advanced Research Projects Agency]. The company has now announced its first commercial product, a probability processing chip that, it claims, can perform error checking and correction for flash memory devices using less than 10% of the power of traditional digital logic, in about 1/30 of the space.
Potentially, this might be a very valuable niche product. As flash memory components get smaller and smaller, the absolute amount of electrical charge used to represent a 0 or 1 bit gets smaller, too. According to Ben Vigoda, CEO of Lyric, on whose PhD thesis the technology is based, the difference between a 0 and 1 on some devices is only the charge of 100 electrons. So error correction is of great importance if storage densities are to be increased.
The company also plans to introduce a General-Purpose Programmable Probability Processing Platform [GP5] chip that will be able to handle general probability problems, as well as a programming language [PBSL Probability Synthesis for Bayesian Logic]. There may well be a fairly formidable learning curve associated with employing this technology in a useful way; still, there are enough probability-based computing problems — ranging from Amazon’s book recommendations to credit card fraud detection — that the idea of the approach is intriguing.