A Hafnium-Oxide Memristive Dynamic Adaptive Neural Network Array

Power and precise scaling are significant restraining elements to advancements in computing. This paper presents a power efficient memristive device technology in the design of a neuromorphic architectural model that promises to overcome many of the performance limitations of conventional Von Neumann systems. The resulting memristive Dynamic Adaptive Neural Network Array (mrDANNA) addresses contemporary application challenges while also enabling continued performance scaling. The mrDANNA system is a mixed-mode neuromorphic computing system built with reconfigurable structure, dynamic adaptation, low-power operation, and is well suited for processing spatio-temporal data. This work is specifically based on a HfO2 memristor device, experimental results for which are presented in this paper. Proof-of-concept simulation results for a mrDANNA pattern recognition network are also presented showing high accuracy for recognizing basic shapes.