Convolutional Neural Network for Mask Resist Pattern Prediction

The patent replaces conventional mechanical/optical simulation methods with a machine learning model (neural network). The system inputs optical image data and resist condition data, and the neural network automatically outputs the resist pattern without requiring complex physical simulation calculations. This substitution of the simulation mechanism with an intelligent algorithm resolves the contradiction by providing high accuracy through learned patterns while eliminating the complexity of traditional simulation processes.

The system employs self-service through automatic threshold determination. The neural network model automatically learns and determines the optimal threshold values for resist pattern prediction by training on input data (optical images and resist conditions). This eliminates the need for manual threshold setting by operators, simultaneously improving prediction precision through data-driven optimization while greatly simplifying the operation process. The system serves itself by automatically configuring parameters that were previously required manual intervention.

The system performs preliminary action by pre-training the neural network model with extensive simulation data and resist condition information before actual mask design work. Once trained, the model can rapidly predict resist patterns for new mask designs without requiring extensive repeated simulations. This preliminary preparation of the intelligence model enables high accuracy in subsequent design work while dramatically improving design efficiency, as the model instantly applies learned knowledge rather than performing computationally intensive simulations for each design iteration.