The theory and method of artificial intelligence (AI) driven by both data and knowledge using symbolic regression have shown promising prospects in exploring potential relationships in integrated circuit (IC) design. This paper develops automatic IC design and optimization techniques using AI. By improving machine learning and coding mode, AI-based automatic design and parameter optimization of ICs are applied to CMOS circuits. The symbolic regression algorithm developed is focused on mining the relationship between the hidden parameters of the circuit being designed. Firstly, the theoretical parameter expression of the circuit is obtained through knowledge of the circuit. Then, a multi-task evolutionary algorithm is used to generate data to optimize the parameters for improved performance. The data are pre-processed and the implicit relation of the parameters is mined by symbolic regression. Compared with manual tuning, the error of the design fitting is shown below 10%, while correlations are 0.98 and 0.84. At the same time, the regression resultant formula can be used to clean the unprocessed data under the same parameters.