| CPC G03F 1/38 (2013.01) [G03F 7/0002 (2013.01); G06N 3/04 (2013.01); G06N 3/063 (2013.01); G06N 3/08 (2013.01)] | 12 Claims |
|
1. A method comprising:
generating a first plurality of digital pattern descriptions representing desired binary patterns representative of desired binary patterns of photomasks to be produced in a process for producing photomasks according to digital pattern descriptions;
obtaining a physical model using which a predicted binary pattern of a photomask can be derived from a given digital pattern description, wherein the predicted binary pattern is a prediction of a binary pattern that would result from inputting the given digital pattern description to the process for producing photomasks;
training a reinforcement learning agent to derive corrected digital pattern descriptions from respective digital pattern descriptions, the training comprising, for each digital pattern description of the first plurality of digital pattern descriptions, the reinforcement learning agent iteratively updating a current candidate corrected digital pattern description based on a similarity between a predicted binary pattern of a photomask derived from the current candidate corrected digital pattern description using the physical model and a desired binary pattern represented by the digital pattern description, and updating the reinforcement learning agent, thereby generating a trained reinforcement learning agent, wherein the reinforcement learning agent is arranged to iteratively update the current candidate corrected digital pattern description in such a way that a long term similarity between the predicted binary pattern and the desired binary pattern is prioritized;
generating a second plurality of digital pattern descriptions representative of desired binary patterns of photomasks to be produced in the process for producing photomasks; and
for each digital pattern description of the second plurality of digital pattern descriptions, deriving a corresponding corrected digital pattern description using the trained reinforcement learning agent, thereby generating training data for training a convolutional neural network for deriving corrected digital pattern descriptions from digital pattern descriptions.
|