We are to develop a neuromorphic chip technology with RRAM-based multi-level synaptic weights having 10x efficiency in energy and 10x improvement in density over conventional neuromorphic memory (e.g., SRAM-DRAM) enabling low-power neuromorphic computing system.
With the ultra-low power, yet good pattern classification or recognition capability, we hoped to explore application possibilities that include robotics, unmanned IOT sensors or security sensors able to operate over a long duration, wearable devices for healthcare etc.
I2R’s involvement include:
With focus on mapping a trained deep learning network onto Spiking Neural Network (SNN) suitable for realization on the Neuromorphic chip and exploring solution to adapt a few layers of the network to allow for on-site customization.
This includes interface to event-driven sensors, multi-chip interface to a normal processor and integrating the solution into the application demonstrator.