The US Naval Research Laboratory (NRL) is engaging in multiple efforts to realize neuromorphic, or brain-inspired, computing hardware, to meet the needs of various Department of Navy (DON) systems such as airborne, submersible, or space-deployed remote and autonomous platforms, where size, weight, and power (SWaP) limitations are mission-critical. The orthogonal requirements of the DON to limit SWaP while offering orders of magnitude additional computational capabilities motivate efforts to develop a new paradigm for computing devices and architectures, independent of industry driving forces.
In this talk, I will describe NRL’s efforts to develop neuromorphic hardware elements using transition metal-oxide memristors, nanodevices which have demonstrated electrical response properties strongly analogous to those of synapses that in a Si-based architecture would require circuitry consuming hundreds of µm2 of die area to replicate. The talk will include discussion on materials development to enable more precise control over the memristor switching properties, electrical testing results from device pairs that exhibit multi-terminal latching, and efforts towards integration of multiple devices to emulate neuron functions such as programmable spiking behavior. The ultimate goal of this research program is the realization of a memristor-based, fully non-digital, neuron equivalent that can function as a unit cell in a cellular neural network.