Toward Universal Multiplexer Multiply-Accumulate Architecture in Stochastic Computing

Stochastic Computing (SC) has recently gained attraction due to its inherent error tolerance and extremely simple arithmetic hardware, making it particularly effective for accelerating modern applications such as neural networks on resource-constrained devices. Traditional SC architectures often ado...

Full description

Saved in:
Bibliographic Details
Main Authors: Yang Yang Lee, Zaini Abdul Halim, Mohd Nadhir Ab Wahab, Tarik Adnan Almohamad
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10878982/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Stochastic Computing (SC) has recently gained attraction due to its inherent error tolerance and extremely simple arithmetic hardware, making it particularly effective for accelerating modern applications such as neural networks on resource-constrained devices. Traditional SC architectures often adopt binary computing principles, relying on dedicated hardware, i.e. AND gates for multiplication and multiplexers (MUX) for addition. However, SC’s mathematical foundation enables the fusion of complex operations into remarkably simple hardware. Several SC studies demonstrated the potential of MUX-based architectures to perform multiply-and-accumulate (MAC) operations, but existing designs face correlation complication, scaling problems, and limited application scope. This paper introduces an auxiliary logic block to address the complexity of MUX select inputs, significantly enhancing the scalability of MUX-based MAC operations. The proposed approach has been validated through SC image processing tasks, including grayscale conversion and Sobel edge detection, achieving up to 75% reduction in hardware resource utilization on field-programmable gate arrays (FPGAs) and up to 96% improvement in computational accuracy compared to traditional AND/XNOR-based SC multipliers.
ISSN:2169-3536