Accurate Prediction and Reliable Parameter Optimization of Neural Network for Semiconductor Process Monitoring and Technology Development

Herein, novel neural network (NN) methods that improve prediction accuracy and reduce output variance of the optimized input in gradient method for cross-sectional data are proposed, variability evaluation approach inputs semiconductor process is suggested. Specifically, electrical parameter measurements (EPMs) power-delay product industrial high-k metal gate DRAM peripheral 29-stage ring oscillator circuits, including NMOS, PMOS, interconnects, focused on. The proposed find an to achieve a lower NN descent than one multilayer perceptron (MLP) mean ensemble MLPs even when considering variabilities devices interconnects. local optima problem MLP resolved by utilizing multiple trained with different train/validation data, their trimmed mean, additional learnable layer. Moreover, adding layer secures versatility various parametric datasets. (R2) 5.6–15.6% sparse space compared ensemble, decrease 73.0–81.6% successfully verified implementing it on EPMs 3977 test patterns 314 wafers 16 lots.