Stochastic Multipliers: from Serial to Parallel

Multipliers are ubiquitous among the core components of multiple signal processing systems. Stochastic computing provides an alternative method to lower the design complexity and power consumption of multipliers, relying on independently and identically distributed bitstreams. In this chapter, several stochastic multipliers are reviewed, evaluated, and applied to image processing algorithms. Serial stochastic multipliers with serial bit-wise operations consume too long computing time and thus more energy. Parallel stochastic multipliers using hard-wired connections tackle this issue to shorten processing time, at a cost of occupied footprint. In addition, deterministic approaches, including relatively prime stream length, rotation, and clock division, are applied to stochastic multipliers to realize completely exact computing results, with double computing time.