Learned fabrication constraints for optimizing physical devices

A computer-implemented method for training a neural network included in a fabrication constraint model to model fabrication constraints of a fabrication process, the computer-implemented method comprising:

• receiving pre-training data including pre-fabrication designs and simulated images representative of the pre-fabrication designs physically fabricated with the fabrication process;
• pre-training the neural network with the pre-training data;
• receiving training data including pre-fabrication structures representative of target designs and post-fabrication images representative of the target designs physically fabricated by the fabrication process;
• training the neural network with the training data after the pre-training;
• selecting a first set of the pre-fabrication structures and the post-fabrication images included in the training data that the neural network indicates uncertainty for fabricability;
• and
• training the neural network with the first set of the pre-fabrication structures and the post-fabrication images after the training of the neural network with the training data.

A computer-implemented method for modeling fabrication constraints of a fabrication process, the computer-implemented method comprising:

• receiving paired training data including pre-fabrication structures and corresponding post-fabrication structures representative of the pre-fabrication structures fabricated by the fabrication process, wherein the pre-fabrication structures includes fabricable and non-fabricable structures;
• training a neural network with the training data to configure the neural network to output a fabricability metric in response to receiving an input design, wherein the fabricability metric is related to a probabilistic certainty that the input design is fabricable by the fabrication process;
• generating a plurality of training fabricability metrics with the training data using the neural network;
• selecting a first set of the pre-fabrication structures based on the plurality of training fabricability metrics, each of the pre-fabrication structures included in the first set having a corresponding one of the training fabricability metrics between a first threshold and a second threshold, wherein the first threshold and the second threshold correspond to a range of values in which the fabrication constraint model is uncertain as to whether the first set of the pre-fabrication structures are fabricable or unfabricable;
• receiving secondary training data including a plurality of secondary pre-fabrication structures, wherein the secondary pre-fabrication structures represent design variance identified from the first set of pre-fabrication structures;
• and
• performing additional training of the fabrication constraint model with the secondary training data to further optimize the neural network to model the fabrication constraints of the fabrication process.