Reinforcement Learning-based Analog Circuit Optimizer using gm/ID for Sizing
DescriptionThere are various machine learning methods to reduce the high time cost of design. However, it is difficult for learning agents to learn circuit operation due to its high non-linearity. Therefore, we trained the agent using reinforcement learning based on the Gm/ID design methodology, which more easily considers short-channel effects and is directly related to circuit performance. It can also determine the size using the Gm/ID Lookup Table. Furthermore, we rearranged the data so that the agent could learn the principle of the circuit more efficiently. Experiments can show how much the number of simulations has decreased.
Event Type
Research Manuscript
TimeTuesday, July 11th3:55pm - 4:10pm PDT
Location3002, 3rd Floor
Analog CAD, Simulation, Verification and Test