An Optimized Stochastic Computing Architecture With Reduced Correlation Induced Errors

Abstract

Stochastic computing (SC) offers a hardware-efficient alternative to binary arithmetic by encoding data as probability bit streams processed with simple logic gates. However, SC accuracy is highly sensitive to correlation between bit streams generated by Linear Feedback Shift Registers (LFSRs), particularly in cascaded DSP systems. While prior work has studied LFSR order in isolation, a systematic analysis of correlation effects at the filter level — and the extension of SC to bioinformatics — remains limited. This paper makes two contributions. First, we propose a correlation-aware SC framework for FIR and IIR filters that mitigates correlation-induced error without modifying the underlying stochastic arithmetic units, achieving up to 70% lower Error Percentage (EP) than state-of-the-art SC FIR designs and 26–68% improvement over SC-IIR lattice designs at zero SCC, while eliminating DSP48E1 usage and cutting MUX count by up to 99.2%. The 3-tap and 7-tap configurations offer the best accuracy-area tradeoff among all designs compared. Second, we present a stochastic-computing implementation of the Nussinov algorithm for RNA secondary structure prediction, reducing FPGA slices by 26.7–50.0%, LUTs by 5.6–14.7%, and power by 5.4–18.2% relative to a deterministic implementation, with gains increasing for longer sequences. MATLAB simulations and FPGA synthesis confirm the practicability of SC for energy-efficient biomedical computing. Finally, this research makes a contribution to Sustainable Development Goals by offering energy-efficient and scalable hardware solutions for both biomedical and computational tasks, illustrating the potential of stochastic computing for interdisciplinary accelerators.