Photoetching hot area detection method based on federal personalized learning

Abstract

The invention discloses a photoetching hot area detection method based on federal personalized learning. The method comprises the following steps that global model parameters returned by each node by a central server are aggregated, common characteristics of each node are fused, the global model parameters are updated, and the latest global model parameters to each node are fed back; each node downloads a global model parameter from the central server, and then trains a local model parameter by using local data to find the optimal local model parameter under the current global model parameter so as to overcome model heterogeneity and data heterogeneity of different nodes; and after the local model parameters are finely adjusted, the nodes train all the parameters by using local data to find the optimal current parameters for searching common features of different nodes. According to the method, the problem of model overfitting caused by too little local data is solved; data between chip design manufacturers is protected, and privacy protection is achieved; and the stability and the overall precision of the federal personalized learning model in the heterogeneous environment are improved.

Description

technical field [0001] The invention belongs to the field of machine learning, and in particular relates to a photolithography hot spot detection method based on federated individualized learning. Background technique [0002] Photolithography hotspots are areas of integrated circuit layout with manufacturing defects. How to quickly and accurately detect photolithography hotspots is an urgent problem to be solved. There are mainly four types of hot spot detection methods at this stage: [0003] 1. Lithography simulation, making full use of the partial coherence characteristics of the light source in the lithography system and the characteristics of one-dimensional chip graphics, and fast planar lithography simulation for one-dimensional chip layout. The lithography simulation method is composed of one-dimensional primitive graphics look-up table method, minimum look-up table and its edge extension and large-area layout simulation without cutting. Traditional lithography ho...

AI Technical Summary

Problems solved by technology

This method can solve the problem of data islands such as data incompatibility between chip design manufacturers, but the performance of federated learning is often poor when dealing with high-level data heterogeneity and asynchronous problems, and it is difficult to meet the detection accuracy standard.

Images