Proceedings PaperNeural-network-based approach to resist modeling and OPC
|Format||Member Price||Non-Member Price|
Resist modeling based on aerial image parameters is an attractive approach to account for resist effects in optical proximity correction. The goal of this work is to introduce neural networks as a means to tackle this problem. We first discuss some of the issues associated with resist modeling based on a fixed, predetermined set of aerial image parameters such as the maximum aerial image intensity. This methodology is found to encounter difficulties if used in conjunction with resolution enhancement techniques such as sub resolution assist features. More specifically we find that layouts characterized by identical values in the aerial image parameters used for modeling experimentally do not always require the same resist correction. As a result modeling errors are introduced that can only be resolved by searching for additional parameters. We have made an attempt to develop an alternate methodology with higher flexibility within the generic framework of a mapping technique. The model uses aerial images taken at a predefined set of sampling points as input parameters. A neural network is used to model the resist effects, taking advantage of the nonlinear non local capabilities of such a system. Using the well defined training methodologies available for neural networks resist models can be calibrated in a fashion similar to standard fitting routines. We first optimize the structure of the neural network based on simulations data derived from a lumped parameter model. A two- layer, non-linear network is found to provide good modeling capabilities for a wide range of resist conditions as well as real 193 nm resist data.