Compressed Sensing Algorithm for Memristor-Based Computing Architectures

Compressed sensing theory offers a novel paradigm for high-dimensional signal processing through sparse sampling and nonlinear reconstruction. However, its efficient hardware implementation remains constrained by the memory-computation separation inherent in the von Neumann architecture. In this paper, a TaO (Tantalum Oxide) memristor-based in-memory compressed sensing framework was proposed. By establishing a MATLAB-LTspice co-simulation platform and performing circuit-algorithm co-optimization, a cross-layer equivalence is constructed between the sparse weight matrix in compressed sensing and the conductance parameters of the memristor array. This approach enables analog-domain computation with accuracy comparable to digital circuits while eliminating data transfer latency. Experimental results demonstrated that the average reconstruction signal-to-noise ratio (ARSNR) of the memristorbased system deviates by no more than 1 dB from the ideal reconstruction.