Abstract
Bandgap voltage reference (BGR) circuits are widely adopted in analog and mixed-signal systems for a stable reference voltage supply; however, performance tuning is usually time-consuming. To address this challenge, this brief introduces an AI-based design method to improve the design efficiency of two classical BGR circuits. A simulator is used to build schematics and scan performance metrics for training neural network (NN) models. The genetic algorithm (GA) identifies optimal design parameters. By balancing the weights in multi-objective fitness functions, temperature coefficients (TC) of 7.76ppm/◦C and 18.27ppm/◦C for the two types of BGR circuits are achieved, respectively. Additional refinements can be made by adjusting weights to meet specific criteria. The execution time of this method is 425.16 s. The proposed method is applicable not only to BGR circuits but also adaptable for real-time performance tuning across various analog integrated circuits, providing a robust framework for parameter determination.