Research on Wafer CMP Temperature Online Detection Compensation Algorithm Based on GA-BP Improved Neural Network

In chemical mechanical polishing (CMP), temperature is a primary factor that determines the rate of chemical reactions, which directly influences the polishing efficiency and quality of wafers. Therefore, constructing an online monitoring system for wafer temperature to enable real-time feedback control of key node temperatures is beneficial for improving the polishing yield of semiconductor wafers. This paper addresses the challenge that non-contact infrared temperature sensors encounter in the CMP process due to their susceptibility to environmental temperature interference. We propose a backpropagation (BP) neural network temperature compensation model optimized by a dynamic adaptive genetic algorithm. The improved genetic algorithm-optimized backpropagation (GA-BP) neural network model incorporates a dynamic nonlinear probability adjustment mechanism and a fitness calibration mechanism. Compared to traditional BP and standard GA-BP models, the compensation accuracy has improved significantly. Experimental results indicate that the mean absolute error (MAE) of the test set has decreased to 0.2381°C, representing reductions of 82% and 68% compared to BP and GA-BP, respectively. Furthermore, the maximum relative measurement error of the temperature sensor within the environmental temperature range of 3.8 to 75.4°C has decreased from 14% to 2.59%. This model effectively mitigates the shortcomings of the traditional BP network, such as slow convergence rate and vulnerability to local optima, thereby providing a novel approach for sensor temperature compensation in ultra-precision wafer polishing applications.