A method to create new types of stochastic circuits and evaluate their performance has been developed at the Computer Science and Engineering department at the University of Michigan. Stochastic circuits reuse conventional logic circuits with probabilistic data encodings and simplify arithmetic operations. While currently limited in applications, stochastic circuits have been shown to outperform conventional circuits in certain image-processing tasks.
The aforementioned method of designing and evaluating the performance of stochastic circuits distinguishes several types of stochastic information encodings that use signal correlations. The technology involves both circuits that work with stochastic encodings, and algorithms for designing and optimizing such circuits. The technology also involves methods for timing analysis (including the use of known mathematical results and algorithms) that estimate the number of signal transitions to produce sufficiently accurate results.
The main advantages of stochastic circuits over traditional circuits are smaller size and lower cost and power. Semiconductor technology scaling can make stochastic circuits more competitive from a speed perspective in future years, and the circuits and design techniques proposed enhance these advantages. Further, the timing analysis is the first-of-a-kind technique to evaluate the speed of stochastic circuits without the need for simulation.
- Multimedia processing
- Real-time processing
- Reduced size and cost compared to traditional digital circuits
- Low power usage
- Circuit speed evaluation without the need for simulation