Achieving ultra-slow timescales in neuromorphic circuits - Application to neural bursting

Abstract

Neuromorphic engineering seeks to replicate the brain's computational power and energy efficiency in hardware. Current neuromorphic designs, however, face challenges in achieving ultra-slow timescales critical for replicating biological neural behaviors such as realistic bursting patterns. This thesis focuses on addressing these limitations through the design and simulation of neuromorphic circuits capable of ultra-slow dynamics while optimizing area efficiency. Using the Cadence Virtuoso software and a general purpose development kit (GPDK), the work reproduces a reference circuit which mimics biological homeostasis, and incorporates this system to an existing neuron circuit, leading to a new modulable neuron design. Key advancements include the combined use of a differential pair integrator (DPI) and an automatic gain control (AGC) loop to achieve ultra-slow temporal filtering and new neuromodulation capabilities while avoiding the need for excessively large capacitors. Simulation results demonstrate significant improvements in achieving the desired dynamics with enhanced area efficiency. This work represents a step towards more practical large-scale neuromorphic hardware capable of mimicking complex neural behaviors.