Oxide-based Resistive RAM Analog Synaptic Behavior Assessment for Neuromemristive systems

Abstract

Resistive RAM (RRAM) synaptic plasticity behavior is essential to enable future power-efficient brain-inspired accelerators. In oxide-based RAM (so-called OxRAM), synaptic plasticity is reflected in the OxRAM conductance modulation capability, making the technology well suited for neuromorphic applications. This chapter assesses the conductance modulation of HfO2-based RRAM and leverages experimental measurement results to establish the inherent synaptic behavior of the technology. The OxRAM conductance modulation capability is demonstrated based on a 1T1R elementary memory array test chip. The memory array OxRAM cells are operated using two distinct programming schemes. The first one is based on RESET (RST) peak voltage modulation. The second one is based on SET compliance current modulation. Both methods show that OxRAM conductance modulation is highly impacted by the variability of the technology. Hence, variability happens to be one the most challenging limitations for OxRAMs used in neuromorphic applications.

Analysis of Conductance Variability in RRAM for Accurate Neuromorphic Computing

“…It is referred to as the coefficient of variation CV. The formula for calculating CV is given in (1).…”

“…However, not solving the issues related to no-idealities such as the variability in the electrical parameters of RRAMs (e.g., conductance variability) may block further development of the technology. In RRAM-based neural networks (NN), conductance variability results in weight variability [1,2]. Weight variability can affect the network during training and inference, affecting the network ability to make precise predictions.…”