Project

DESCRIPTION:

Effective power management is essential for balancing energy efficiency and performance in modern microprocessors. Fully integrated digital low-dropout (DLDO) regulators provide a scalable and compact solution; however, conventional architectures such as Computational DLDOs (CLDOs) rely on fixed-state, sequential logic. This limits their ability to adapt to fast-changing load conditions and process variations, resulting in suboptimal transient response and regulation accuracy. This project proposes the development of a Reinforcement Learning (RL)-based DLDO architecture that replaces the conventional finite state machine (FSM) controller with an adaptive, learning-based agent. The RL controller will enable rapid, single-step transient recovery and enhanced robustness across diverse operating conditions, paving the way for more intelligent and efficient power regulation in next-generation integrated circuits.