Neuromorphic computing offers an enticing solution to reduce artificial intelligence (AI) size, weight, and power (SWaP) through hardware-algorithm co-design inspired by working principles of neurophysiology and biological intelligence. However, the majority of today's neuromorphic computing research focuses on models of the nervous system that are too simplistic (e.g. point neurons, rate coding, digital designs) to capture biology's full computational richness and efficiency. The goal of this project is to demonstrate the superiority of neuromorphic systems that emulate the fully analog and non-linear nature of nervous tissue for complex spatiotemporal information processing. In particular, we aim to design and fabricate a fully analog neuromorphic chip with novel active dendritic processing and non-linear synaptic devices for incremental learning. The capabilities of the chip will be demonstrated on a number of scientific discovery tasks such as classification and time series forecasting.
