Split weight distribution learning for binary neural networks exploiting nano-memristive graphene/sumanene/graphene devices

Research in artificial intelligence (AI) and neural network (NN) technologies has been boosted by recent hardware advances, while the emergence of self-synthesized materials such as nanocarbons has attracted much attention to overcome the limitations of Si-CMOS technologies. Prototypes of novel nanocarbon devices consisting of sumanene (S) sandwiched between graphene (G) sheets have recently been demonstrated, but the potential application of such novel devices, e.g., vector-matrix multiplication (VMM) for NNs, has not yet been explored. In this study, we propose learning methods to obtain binary weights having split weight distribution suitable for inference in binary NNs (BNNs) to exploit G/S/G devices. The demonstrated learning method was evaluated in BNN simulations, showing inference improvement from 77% to 94% for the MNIST dataset. Furthermore, robustness against imperfect device yield was confirmed with respect to conventional NNs that combine continuous weights and analog computing.