Data Augmentation and Deep Learning for FET Modeling

In this phase of research, we are addressing the challenges of FET modeling with a limited dataset of approximately 80,000 data points from 27 transistors. Our current focus is on leveraging data augmentation techniques to enhance the dataset. These techniques range from traditional interpolation methods to advanced approaches using Generative Adversarial Networks (GANs), Autoencoders, and other modern augmentation methods.Once the dataset is augmented, we aim to build and refine deep learning models, and potentially experiment with reinforcement learning algorithms, to develop accurate and robust FET models. This project has implications for improving semiconductor device modeling and integrating these models into tools like SPICE or Verilog-A.Participants will:

• Explore and implement advanced data augmentation techniques.
• Develop and evaluate deep learning or reinforcement learning models for FET modeling.
• Gain exposure to semiconductor device modeling and the integration of compact models with CAD tools.

Students with machine learning experience or those who have completed CS412 are encouraged to apply.