Reinforcement learning (RL) based federated automated defect classification and detection
AssigneesImec Vzw, Katholieke Universiteit Leuven
InventorsBappaditya Dey, Enrique Dehaerne, Sandip Halder
Life of this patent
Estimating the term…
Abstract
A method for training a local machine learning model is provided. The method includes receiving a scanning electron microscope (SEM) image of semiconductor features. The method additionally includes determining a location and dimensions of a bounding box within the SEM image. The method yet further includes determining, whether a defect feature exists within the bounding box, based on an unsupervised object detection process. The method also includes, if the defect feature exists within the bounding box, receiving positive rewards. The method also includes, if the defect feature does not exist within the bounding box, receiving negative rewards.
Claims19
A method comprising:
- training a local machine learning (ML) model, wherein the training comprises:
- receiving a plurality of scanning electron microscope (SEM) images, wherein the plurality of SEM images vary in at least one of:
- resist profile or resist thickness;
- and
- for each SEM image of the plurality of SEM images:
- determining a location and dimensions of a bounding box within the respective SEM image;
- determining, whether a defect feature exists within the bounding box, based on an unsupervised object detection process;
- if the defect feature exists within the bounding box, receiving positive rewards;
- and
- if the defect feature does not exist within the bounding box, receiving negative rewards.
Dependencies are inferred from the claim text.