Selecting test patterns with the pool-based sampling method

A precisely calibrated process model is required for the optical proximity correction purpose. To fit the model, a test pattern mask needs to be prepared and the lithographic experiment should be done. It is beneficial to select fewer test patterns without compromising the accuracy of the lithographic model fit with them. We apply the autoencoder and core set method to select parts of gauges out from the existing test patterns set. The models obtained with the selected data achieve nearly the same precision on the original set of test patterns compared with the model fit by the whole set data.