earticle

영어

Recently, neuromorphic system has been actively studied as hardware-based neural networks using memristive devices to overcome the limitations of Von-Neumann architecture. In particular, on-chip learning has been proposed and optimized because hardware itself can learn based on weight-update linearity characteristics. In this study, we demonstrate the effect of weight overlap region on on-chip learning when designing the learning characteristics of devices. We use identical potentiation/depression pulse of fabricated Pt/Al2O3/TiOX/Ti/Pt stacked memristors to study the effect of conductance overlap region on the recognition accuracy for modified national institute of standards and technology (MNIST) dataset. The learning characteristics of memristive show a characteristic that is highly dependent on the overlap range.