Speaker
Description
Biologically-inspired computation schemes are more effective than standard digital-based approaches when dealing with complex, unstructured tasks as image recognition. In particular, systems of frequency-locked, coupled oscillators exhibit associative memory capabilities encoded in the phase difference of the signal. We are using oscillating neural networks as hardware accelerators for image recognition. In this work, nanometer scale relaxation oscillators are built using the insulator-metal transition of VO2. Our experiments show that the relative phase of coupled oscillators can be configured with the tuning of the coupling strength, i.e. the magnitude of the coupling resistor. This offers the perspective of realization a compact, computational network of oscillators. Mathematical simulations prove the computing capabilities of these networks when scaled to larger sizes.
