Synaptic Behavior of Nanoscale ReRAM Devices for the Implementation in a Dynamic Neural Network Array

Abstract

Resistive random access memory (ReRAM) is a new form of non-volatile memory that has the potential to replace Flash memory or augment the current memory hierarchy. In addition, novel circuit architectures have been proposed that rely on newly discovered or predicted behavior of ReRAM devices. One such architecture is the memristive Dynamic Adaptive Neural Network Array (mrDANNA), developed to emulate the functionality of a biological neural network. This architecture relies on synapses which are capable of changing their resistance in an analog fashion by applying ultra-short pulses. We demonstrate ReRAM devices that show this tendency. The ReRAM devices shown here are based on an HfO 2 switching layer that sits on a tungsten bottom electrode, is covered by a titanium oxygen scavenger layer, a titanium nitride top electrode, and are structured to a size of 100×100 nm2. In this work, we show devices that exhibit incremental resistance changes in a synaptic fashion and can switch using pulses as short as 5 ns. A major hurdle is the variability observed with these devices and its effect on the designed mrDANNA architecture. One focus of the ongoing work is a simulation on the effect of the observed variability. For this purpose, a Monte Carlo simulation with extracted variability data are being performed to demonstrate the impact on this neuromorphic architecture.