A MLC STT-MRAM based Computing in-Memory Architec-ture for Binary Neural Network.

摘要

Due to add operations dominated computation and simplified network in Binary Neural Network(BNN), it is promising for TOT scenarios, which demand ultra -low power consumption and hardware area overhead. By means of exploiting the in -memory computing methods and high density of MLC STT-MRAM, this work designs a MLC-STT-CIM (Computein-Memory) based computing in -memory architecture to achieve add operations for BNN, to further reduce power consumption and area overhead