Exploiting the State Dependency of Conductance Variations in Memristive Devices for Accurate In-Memory Computing

Dec 1, 2023 by A. Vasilopoulos, J. Büchel, B. Kersting, C. Lammie, K. Brew, Samuel Choi, T. Philip, N. Saulnier, Vijay Narayanan, M. Le Gallo, A. Sebastian (IEEE Transactions on Electron Devices)

DOI 10.1109/TED.2023.3321014

We exploit the state dependence of conductance noise in PCM memristors to optimally map weights across multi-device synapses so most devices sit in stable SET or RESET states, cutting MVM error and boosting resilience to retention drift. Applied to MNIST and ImageNet workloads this simple mapping raises experimental and hardware-realistic accuracies while slashing run-to-run variability, making AIMC with noisy devices actually predictable.

source S2, crossref