Abstract:The design of high-performance analog circuits is a time-consuming process that requires the involvement of highly qualified designers and frequent tedious interactions with circuit simulators. This sets a need for a method that accelerates the circuit design process and reduces unnecessary designer-simulator interactions. Existing methods are either slow or unable to deal with complex tasks. This paper proposes a simulation-based optimization method based on a deep reinforcement learning (DRL) agent to optimize analog circuits and accelerate the design process. In this method, the DRL agent exploits the soft-actor-critic algorithm based on an improved reward function considering multiple conditions to expedite the parameters optimization step of analog circuits. The designer provides the RL agent with the circuit testbench, design requirements and circuit parameters. The agent then repetitively interacts with the circuit simulator to tune these parameters. In this work, an incremental RL approach is also proposed to deal with complex circuits. Incrementality is found to be increasingly beneficial as the circuit complexity increases (increased number of design parameters and competing objectives). The effectiveness of the proposed method was validated on a complex circuit defined by a design space comprising as many as 36 dimensions (17 design variables and 19 competing objectives). The reported results demonstrate that our method can significantly accelerate the design process and yield high-performance optimized designs. Notably, the required number of simulation steps is on the order of 1e3, which is one to two orders of magnitude more efficient than similar state-of-the-art DRL approaches.

An Efficient Kriging-based Constrained Multi-objective Evolutionary Algorithm for Analog Circuit Synthesis Via Self-adaptive Incremental Learning

Fast Surrogate-Assisted Constrained Multiobjective Optimization for Analog Circuit Sizing Via Self-Adaptive Incremental Learning

Optimization Algorithm of Evolutionary Design of Circuits Based on Genetic Algorithm

An Efficient Reference-point Based Surrogate-assisted Multi-objective Differential Evolution for Analog/RF Circuit Synthesis

Analog Circuit Optimization System Based on Hybrid Evolutionary Algorithms

Surrogate-Assisted Multi-objective Differential Evolution based on Gaussian Process for Analog Circuit Synthesis

An Efficient Batch-Constrained Bayesian Optimization Approach for Analog Circuit Synthesis Via Multiobjective Acquisition Ensemble

Asynchronous Parallel Expected Improvement Matrix-Based Constrained Multi-Objective Optimization for Analog Circuit Sizing

A Gaussian Rocess and Multi-Swarm Optimizer Assisted Optimization Approach for Analog Circuit Design

An Incremental Evolutionary Algorithm Suitable for Industrial Platform for Analog Circuit Design

A Real Coded Vector Population-Based Incremental Learning Algorithm for Multi-Objective Optimizations of Electromagnetic Devices

Multi-objective Bayesian Optimization for Analog/RF Circuit Synthesis

A Multi-Algorithm Framework for the Evolvement of Analog Circuits

A Batch Bayesian Optimization Approach for Analog Circuit Synthesis Based on Multi-Points Selection Criterion

Analog Circuit Design Optimization Based on Evolutionary Algorithms

An Efficient Multi-fidelity Bayesian Optimization Approach for Analog Circuit Synthesis

Incremental reinforcement learning for multi-objective analog circuit design acceleration

An Efficient Multi-Objective Evolutionary Algorithm For Combinational Circuit Design

An Efficient Analog Circuit Sizing Method Based on Machine Learning Assisted Global Optimization

A Hybrid Genetic Algorithm with Hyper-Mutation and Elitist Strategies for Automated Analog Circuit Design

AutoCkt: Deep Reinforcement Learning of Analog Circuit Designs