Influence of Parameters of Crossbars with STT-MRAM on the Accuracy of Analog Neural Network

Matrices of STT-MRAM cells assembled into a 3 × 3 crossbar architecture were studied. We experimentally found how interconnect resistance affects the fidelity of the combined multiply-accumulate operations. The results obtained were compared with the simulation results in the CAD Cadence. The results of the work allow us to evaluate how the fidelity of combined MAC operations contributes to training analog neural networks.