Effective accuracy estimation and representation error reduction for stochastic logic operations

Stochastic Logic provides with a methodology that allows the execution of analog operations by means of digital logic. In this work we calculate the expected values and the estimated error for number representation in stochastic logic operations. From these two quantities, we calculate the effective number of bits, which represents the effective resolution of the number in stochastic logic operations. Finally, a random number generator (RNG), fundamental block for number representation in stochastic logic and of stochastic logic operations, capable of reducing the representation error is presented and evaluated.

[1]  L. A. Goodman.  On the Exact Variance of Products , 1960 .

[2]  Brian R. Gaines,et al.  Stochastic computing , 1967, AFIPS '67 (Spring).

[3]  M. Loève,et al.  Elementary Probability Theory , 1977 .

[4]  Bruce F. Cockburn,et al.  Stochastic circuit design and performance evaluation of vector quantization , 2015, 2015 IEEE 26th International Conference on Application-specific Systems, Architectures and Processors (ASAP).

[5]  J. Neumann.  Probabilistic Logic and the Synthesis of Reliable Organisms from Unreliable Components , 1956 .

[6]  Josep L. Rossello,et al.  Ultra-Fast Data-Mining Hardware Architecture Based on Stochastic Computing , 2015, PloS one.

[7]  Zhongfeng Wang,et al.  Area-efficient scaling-free DFT/FFT design using stochastic computing , 2016, 2016 IEEE International Symposium on Circuits and Systems (ISCAS).