New Locally-Active Memristor for Neuromorphic Application

Artificial synapses and neurons-based neuromorphic computing have garnered significant attention due to its lower energy consumption with exceptional computational efficiency. The locally-active memristor (LAM) is a potential candidate for artificial neural networks, highlighting its potential applications in neuromorphic computing. This study proposes a memristor with locally-active characteristics to implement its typical behaviour in neuromorphic computing. Several numerical and theoretical analyses exhibit the behaviours of the proposed LAM model, including non-volatility, and local-activity. The validation of the proposed memristor is physically experienced through the Arduino Uno. Ultimately, the designed LAM is employed in a memristor crossbar, which is used as a kernel of convolutional neural network (CNN) to recognize handwritten digits. The accuracy reaches 98.28%, which displays the effectiveness of the proposed LAM in real-time applications.