Abstract
Neuro-inspired computing systems provide an opportunity to mitigate the bottleneck of the Von Neumann architecture. A memristor-based reconfigurable neuromorphic array (RNA) can facilitate implementations of the neuro-inspired dynamic architecture (NIDA). In this work, current-controlled HfO2 -based memristive devices are considered in the construction of an RNA-based system, which shows reliable functionality compared to the state-of-the-art. An analog leaky integrate-and-fire neuron with absolute and relative refractory provides more bio-inspired mechanisms for adaptation. Time-division multiplexing (TDM) is also leveraged in the system-level implementation in order to improve the overall footprint and power consumption of on-chip interconnection logic. An intra-cluster communication protocol treats the spikes between neural cores as spikes, much like a human brain, saving energy and peripheral resources by eliminating encoding and decoding operations. A system-level implementation of the proposed architecture built using 65 nm CM OS and HfO2 -based memristive technology is also described.