Traditionally, there has been a clear separation between TCAD and full chip simulation tools. While TCAD is normally used during process development, it remains outside the realm of full chip corrections due to its long runtime requirements.
One of the key components of a model-based OPC tool is fast and reliable CD prediction of all features present in the design layout, usually at the best point in the process window. Such models exist today, and are routinely used in production. However, there is also a growing need to make more informed decisions about the tradeoffs between accuracy, correction and turn around time. For this reason we need to develop techniques that enable full chip simulations across a variety of process conditions. It was previously shown that a combination of optical vector and variable threshold models can be calibrated to predict well across multiple focus conditions, however dose predictions have not yet been studied.
In principle, the possibility of having models that predict process window behavior exists today by calibrating empirical models separately at every one of the process conditions under investigation. However, this method has two clear disadvantages. On the one hand, it cannot guarantee that such models can be extrapolated to conditions other than those used for their calibration, thus not making it possible to provide models "on demand" for arbitrary focus and dose values. And on the other hand, a substantial additional effort is required for creating models at more than one process condition.
This work concentrates in listing the requirements to evaluate the robustness of any process window model as well as showing how a well-calibrated compact model can be used to predict -within metrology uncertainty- dose and defocus induced changes for a wide variety of features. While such capability has a number of applications, we will
describe a methodology for IC-product verification.