Today's microprocessors are designed around logic, and handling 1s and 0s - and they're very good at it. However, they aren't so good at probability calculations and dealing with matters of uncertainty.
Of course, these sorts of problems pop-up quite a lot in everything from carrying out search queries to financial modeling and even in error-correction calculations for flash memory.
Starting from scratch
A start-up called Lyric Semiconductor has announced this week that it has gone back to basics and rethought the way that processors should be designed. The result is that the company has developed logic gates that are designed to specifically handle probability calculations.
The end product is a probability processor that can perform these sorts of tasks in a fraction of the time and with far less circuitry than a traditional x86 chip.
As a part of the announcement, the company unveiled the LEC, its first chip using this design, which is intended for use in SSDs. Lyric claims that it is 30 times smaller and uses 12 times less power than current error-correction chips based on the same manufacturing process. Furthermore, the processors have a higher throughput, removing a potential bottleneck from the equation.
Probably the future of probability
Error correction is needed because, according to the company, one in every 1,000 bits of data on an SSD will be incorrect. This makes error-correction circuitry essential to maintain data integrity. As the density of flash storage increases, this problem will only get worse. The LEC is available for licensing now and could start showing up in retail products within the next year.
The company isn't content to stop at flash chips, though, and with good reason. Thanks to a $20m investment from the US Department of Defence's Defence Advanced Research Projects Agency (DARPA), Lyric is building a general-purpose probability processor called the GP5. This chip will be capable of being used for a wide variety of different applications and the company believes that it could speed up probability calculations by several orders of magnitude. If all goes to plan, samples of the GP5 will be sent out in 2013.
