Performance Optimization of Atomic Layer Deposited HfOx Memristor by Annealing With Back-End-of-Line Compatibility

Hafnium oxide (HfO _x ) memristor has attracted enormous attention due to its high performance and back-end-of-line (BEOL) compatibility, thus providing a novel approach to implementing artificial intelligence neural networks. In this work, great performance optimization of HfO _x memristor has been achieved by using atomic layer deposition (ALD) method and post-metal annealing (PMA) process, in which both procedures are with low-temperature budget (< 300 °C) and are compatible with CMOS BEOL process. The device exhibits forming-free, high yield, good linearity, fast speed and non-volatile characteristics. Besides, the device conductance can be well modulated by using the most desired pulse protocol, namely the identical pulse with same pulse amplitude and width. More than 3bit stable conductance states have been obtained, indicating its great potential in practical memristor neuromorphic computing system.