Device Parameter Extraction Method for Training Performance Predicting Models

For performance modeling using machine learning (ML), parameters that characterize transistors can be extracted and used as data for an ML model to predict circuit performance. However, generating and providing proper training data (e.g., device parameters for given bias voltage levels of the transistor's drain/gate/source) is crucial for the accuracy of the performance predicting model, as the prediction outcomes may vary depending on how the training data is generated. In this paper, we validate that the prediction error of an ML-based performance model can be reduced by employing the proposed method that extracts device parameters by choosing five bias voltage levels based on the threshold voltage (V<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</inf>).