Device Modeling IC-CAP 2025 Update 1.0 Product Release

The AI-enabled ML Optimizer reduces model extraction time from days to hours.

Enabling AI for Device Modeling

Device Modeling (IC-CAP) 2025 Update 1.0 features a Machine Learning (ML) Optimizer designed to streamline and automate the parameter extraction process for complex compact device models. It addresses the limitations of traditional gradient-based optimizers by leveraging derivative-free machine learning techniques to explore the complex, high-dimensional parameter spaces efficiently. This results in faster convergence, improved solution quality, and increased automation, reducing extraction time from days to hours. This release adds a new stop criteria designed to stop the Optimizer when the error does not improve, avoiding unnecessary iterations. A new formulation for the error makes it easier to prioritize certain plots during the extraction. The ML Optimizer is part of the existing W7010E IC-CAP Analysis product, and it can be applied to the extraction of any compact model. Several example projects are provided to get you started and help you integrate the new Optimizer into your extraction flows.

IC-CAP 2025 Update 1.0 also features significant improvements in the Model Generator (MG). In 2025, we added new tools to improve productivity, such as the Model Generator QA, and increased the speed of certain operations, such as loading data and creating scaling plots. MG now includes complete extraction example projects to provide a good starting point for learning the tool. In this update, we have added new tools to improve the MG usability. 

IC-CAP 2025 Update 1.0 now integrates and ships Python 3.12.11, including new and updated packages. For guidance, please refer to the official Python documentation at https://www.python.org/

Enhancing model accuracy with IC-CAP Hybrid-ANN modeling.

Hybrid ANN Modeling for GaN Devices

IC-CAP 2025 introduces the concept of Hybrid-ANN. The classic ANN approach creates a neural network-based model pre-trained on measured data. While this model can be very accurate, it is a black-box model and offers no specific insight on the device. On the contrary, Hybrid physical-ANN modeling offers a powerful new approach to device modeling, combining the advantages of physics-based models and data-driven ANNs. The original compact model is maintained and augmented with ANN-based elements carefully targeting second-order effects. The combined model offers superior accuracy while maintaining the structure and parameters of the original compact model.

Keysight's IC-CAP and the new ANN Modeling Toolkit provide a user-friendly environment for implementing this methodology. This approach enables faster model development and improved accuracy for complex device behaviors and novel technologies. It addresses modeling issues with minimal programming, enhances modeling accuracy quickly, and builds ANN frameworks quickly.

IC-CAP 2025 includes an example of a hybrid ANN model extraction applied to the ASM-HEMT industry-standard model for GaN devices. The W7009E IC-CAP ANN Modeling Toolkit is necessary to extract Hybrid-ANN models.