Analysis of the Thermal Degradation Effect on a HfO2-Based Memristor Synapse Caused by Oxygen Affinity of a Top Electrode Metal and on a Neuromorphic System

Abstract

The thermal budget problem needs to be considered for a neuromorphic system based on a memristor to be compatible with silicon-based devices. The thermal degradation caused by postmetal annealing (PMA) was investigated for HfO2-based memristor (HM) devices at temperatures above 300 °C to ensure the thermal stability of memristor devices and analyze the effect of thermal degradation on the neuromorphic system. As thermal degradation is caused by oxygen atom movement between the interlayer and the top electrode (TE), the thermal stability of the memristor can be improved by adjusting the oxygen affinity of the TE metal. By changing the TE metal from titanium (Ti) to tantalum (Ta), the resistance of the high-resistance state and resistance variability after PMA at 400 °C for 1 h decreased from 516 to 10% and from 21 to 4%, respectively. In addition, pattern recognition simulation was performed using an artificial neural network consisting of a memristor device. Although the pattern recognition simulation with the Ti TE-HM showed a pattern recognition accuracy of 82.3% after PMA owing to thermal degradation, the simulation with the Ta TE-HM showed a high accuracy of 51.7% even after PMA. This experimental approach can facilitate the development of neuromorphic systems with good thermal stability up to 400 °C.