Energy Efficient Implementation of MVM Operations Using Filament-Free Bulk RRAM Array

In this work, we present a hardware implementation of spiking neural network (SNN) model using trilayer bulk RRAM crossbar arrays for an autonomous navigation/racing task. We demonstrate multi-level switching in MΩ regime using trilayer (Al <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> O <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf> /TiO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> /TiO <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</inf> ) bulk RRAM devices without needing a compliance current. We also present a neuromorphic compute-in-memory (CIM) architecture based on trilayer bulk RRAM crossbars using energy-efficient voltage sensing and row differential encoding of weights to experimentally implement highly accurate matrix-vector-multiplications (MVM). Our results suggest that use of our bulk RRAM can reduce the energy consumption by more than 100× compared to other filamentary RRAMs. This work paves the way for neuromorphic computing at the edge under strict area and energy constraints.