FeFET-Based Binarized Neural Networks Under Temperature-Dependent Bit Errors

Ferroelectric FET (FeFET) is a highly promising emerging non-volatile memory (NVM) technology, especially for binarized neural network (BNN) inference on the low-power edge. The reliability of such devices, however, inherently depends temperature. Hence, changes in temperature during run time manifest themselves as bit error rates. In this work, we reveal temperature-dependent model FeFET memories, evaluate its effect BNN accuracy, and propose countermeasures. We begin transistor level accurately impact rates FeFET. This analysis reveals asymmetric Afterwards, application level, errors accuracy BNNs. Under errors, drops to unacceptable levels when no countermeasures are employed. two countermeasures: (1) Training BNNs tolerance by injecting flips into data, (2) applying rate assignment algorithm (BERA) which operates layer-wise manner does not inject training. experiments, BNNs, applied to, effectively tolerate entire range operating