TaOx synapse array based on ion profile engineering for high accuracy neuromorpic computing

Resistive random-access memory is a promising candidate for high-density, low-power neuromorphic computing. Here, we demonstrate a thermally oxidized TaO x RRAM array with built-in oxygen concentration gradient, which ensures good linearity and symmetry and consequently better inference accuracy on both MNIST and CIFAR10 datasets after training with perceptron, LeNet-5 and ResNet-18 networks, thus illustrating the potentiation of applying RRAMs for high-density, high-accuracy, large-scale neuromorphic computing.