چكيده به لاتين
Stochastic computing (SC) proposed in the early of 1960s as a low-cost alternative to weighted binary radix (WBR) processing. This approach is "unique" since it treats data in the form of bit-streams. SC has simple processing units and consumes low power. For instance, the multiplication of two numbers in the SC is performed using a single AND gate, and the sum of two numbers is made using a two-to-one Multiplexer. In spite of the mentioned advantages, as well as inherent error-resilient, SC has high energy consumption due to the relatively long processing time in the high precisions. So this approach, in some circumstances is impractical. However, recent technological achievements in various fields of studies, including artificial neural networks (ANNs), image processing and signal processing, provide the possibility of inaccurate processing to gain more efficiency in the other design parameters like area occupied by the circuit and power consumption.
Nowadays, there is growing attention to the theory of Fuzzy-logic and its applications. An efficient design of fuzzy-inference hardware becomes the only solution in many high-performance applications like machine learning. Since fuzzy variables have real values in [0,1] interval, this dissertation proposes to apply the concept of processing in unary bit-streams into the fuzzy-logic controllers. This work proposes an idea to tackle the long processing time, which significantly reduces the latency at no accuracy loss. In the case of implementing a fuzzy-inference engine with 81 fuzzy-inference rules, the proposed architecture provides up to 82% savings in the hardware area, 46% reduction in the power consumption, and 67% saving in the energy consumption compared to the traditional weighted binary implementation.
