Probabilistic Processors

Probabilistic processors calculate probabilities instead of certainties. They are therefore more prone to errors than a binary logic processor, but for certain applications, the increase in speed and decrease in complexity is well worth the trade. Lyric Semiconductor has successfully produced a small scale probabilistic processor that is in use today, and are in the midst of developing a larger scale chip (GP5) for more complex applications.


Probabilistic processors use analog values instead of digital, with the voltage on each bit (p-bit) being the probability that that bit is 1. The current technology allows for a resolution of 8 bits, meaning that each bit can represent 256 different voltage values.

The basic logic for these processors are Bayesian NAND gates, which have been developed to work with probabilities as Bayes defined them. He viewed probabilities as a measure of a state of knowledge instead of a frequency of some occurrence.



The motivation for the development of this technology is its speed and power consumption. For some problems it could provide speedup by the factor of 1000, while using a small fraction of the power and physical circuitry. This will also lead to a reduction in circuit complexity and will require many fewer transistors. []

Besides being able to use one bit to hold more data than a binary logic machine, probabilistic systems will be able to parallelize probability computations, instead of running them serially, providing another edge over standard CPUs for certain applications.


The major current use for probabilistic processors is Lyric's Error Correction (LEC) chip, used for error correction in flash memory. It does this with about 3% of the circuitry and 8% of the power required by previous binary logic error correction circuits. This will become much more important as solid state devices shrink, and the error rate rises correspondingly. []

Future Uses

Due to variations in voltage, these chips will be less precise than binary logic chips, but for certain applications, this will be unimportant. Some of these uses include spam filters, for calculating the probability that an email is spam, shopping pattern analysis, decoding the signal for cell phones, encoding or decoding video, or in situations where the data being handled is already uncertain. It has also been suggested that probabilistic processors would be useful in genome sequencing or for heightening security measures in password entry patterns.



The development of probabilistic processors requires development of new logic gates and programming languages as well as the main architecture. Lyric is currently on schedule to demonstrated their new CPU in 2013.