Fast and Flexible Decision Making using CMOS Hardware

Abstract

This thesis focuses on the study and implementation of neuromorphic circuits capable of demonstrating multistability and hysteresis, inspired by biological neural processes. The main objective is to design a hardware system that mimics fast and flexible decision-making behavior. The work begins with an introduction to CMOS transistors, the circuits used and their operation. This is followed by a detailed study of bifurcations, feedback loops and the mathematical principles that lead to the existence of several equilibria. Possible equilibrium changes are explored and supported by numerical simulations performed in Julia. These principles are then implemented in an electrical circuit using the Cadence Virtuoso tool. The different simulations demonstrate the behavior of bistable and tristable circuits, with a particular focus on hysteresis loops that highlight state-dependent transitions and region of multistability. Experimental results confirmed theoretical predictions, showing multistable behavior with equilibrium states dependent on initial conditions and input current. This demonstrates the potential of CMOS neuromorphic circuits for low-power, biologically inspired decision-making. In conclusion, the combination of theoretical modeling, circuit simulations, and experimental testing validates the approach and provides insights into the design of energy-efficient and adaptive neuromorphic systems.