O-RRP: Optical Proximity Correction Runtime and Resource Prediction using Machine Learning

Growth of compute resource demand in Optical Proximity Correction (OPC) is driving the adoption of cloud-based workflows, making accurate prediction of runtime and memory usage crucial to optimize resource allocation and scheduling. This paper presents a machine learning-based method for predicting OPC runtime and memory usage using data from the large number of worker jobs contained in a single representative OPC run. Regression models are trained on extracted geometrical features to capture the relationship between layout characteristics and compute resource consumption. Experimental results demonstrate total runtime prediction within 11.6% and peak memory prediction within 5.3% of actual measurements across multiple design layouts and recipes. This method can be easily integrated into existing OPC pipelines, allowing seamless deployment of optimized hybrid multi-cloud environments for substantially faster tape-out cycles through improved infrastructure utilization.