Fig. 4: Graphene memristor based vector-matrix multiplication (VMM) using k-means clustering.
From: Graphene memristive synapses for high precision neuromorphic computing
a Memory architecture for executing VMM operations. Drain voltages (V1 and V2) are used as the input vector and graphene memristor conductance values (G1 and G2) are used as the weight matrix. The output current (IOUT) is used as the output vector. b Colormap of the expected output current corresponding to different input voltage vectors for G1 = 215 µS and G2 = 155 µS. Experimentally obtained output currents when these weights are rounded to the nearest conductance states offered by the respective graphene memristors with uniformly distributed memory levels for (c) N = 2 and (d) N = 4. Error between the expected and actual output current for (e) N = 2 and (f) N = 4. g Experimentally obtained output current and (h) error when the weights are rounded to the nearest conductance states following k-means clustering.
