Stochastic Computing: Systems, Applications, Challenges and Solutions

Modern day applications demands high speed and robustness with small integration area and low power consumption. Due to the advancement in technology, the dimensions of transistors goes on reducing thereby makes it possible to incorporate more and more number of transistors on a chip. However, reduction of the size of transistor increases leakage currents hence power consumption and unreliability. To address these problems either transistor has to be replaced by other device or classical way of computing has to be replaced by an alternative computing technique. One such unconventional form of computing is Stochastic Computing. Stochastic computing is probabilistic in nature. While weighted binary computing provides high speed and accuracy at the cost of large integration area and huge hardware cost. Also, these computing techniques have small noise robustness. Stochastic computing is efficient in terms of integration area, hardware cost and has high noise robustness. While the speed of stochastic computing circuits is comparable to weighted binary computing systems for small applications, their accuracy is far less then weighted binary computing circuits. This paper describes the key concepts of stochastic computing, its variants, conversion process and stochastic circuits. Main challenges to stochastic computing and their possible solutions are also highlighted in this paper.