Vector-Matrix-Multiplication Acceleration with Multi-Input Pr0.7Ca0.3MnO3 based RRAM for Highly Parallel In-Memory Computing

In this work, we propose multi-input memristorbased vector-matrix-multiplication (VMM) acceleration in memristor crossbar arrays via bitgrouping. First, we demonstrate parallel processing and multiply-and-accumulate(MAC), which is fundamental to VMM, using standard 2-terminal(T) PCMO-RRAM devices. Second, we introduce 3T PCMO-RRAM with two-input/one-output terminals. Two input terminals per RRAM enhance parallel computation by grouping MAC input bits in single device. Lastly, computation with two 3T-RRAMs demonstrating 16 possible outcomes with high MAC linearity (average 3% deviation) is shown.