<p>As semiconductor manufacturing moves toward fine feature sizes, precise and efficient resist model calibration has become crucial for optical proximity correction to ensure pattern fidelity. However, traditional calibration methods struggle with efficiency and scalability and are prone to becoming trapped in local optima. Herein, we propose a surrogate-assisted genetic algorithm (SAGA) that integrates Kriging interpolation-based surrogate models and dynamic adaptive mechanisms to optimize resist model coefficients, convolution kernel parameters, and aerial image settings jointly. By leveraging surrogate models to predict high-performance solutions and adaptively adjusting crossover/mutation rates, SAGA balances global exploration and local exploitation, achieving rapid convergence and superior model accuracy compared with other algorithms. Experimental validation across three resist cases demonstrates that SAGA outperforms conventional genetic algorithms and grid search. Compared with other algorithms, SAGA not only achieves higher accuracy but also converges faster, with its optimization trajectories stabilizing earlier in the iterative process. These results highlight SAGA’s potential for efficient and high-precision resist calibration in computational lithography.</p>