Reinforcement Learning-Driven Autonomous Optimization of Multi-Parameter Power Efficiency in High-Dimensional VLSI Circuit Design Architectures

The study focuses on the use of one RL method to minimize power consumption in VLSI circuits. In this paper, the RL model is trained with DQN algorithm where the VLSI circuit design is optimized in terms of power consumption by controlling certain parameters without compromising performance. The RL agent communicates with a developed simulated environment utilizing open-source VLSI design tools in order to utilize the DQNs optimization capability in design spaces that are dynamic. It is important to point out that the entire process of power consumption forecasting, utilizing the proposed architecture, is validated using a single, standardized power consumption dataset. TensorFlow is used for modeling while simulating and validating the circuits would require Magic VLSI tool. This approach has shown the usage of the RL-based optimization for arriving energy-efficient the VLSI design.