Adaptive Computing for RF Device and Component Modeling

While circuit simulation tools have revolutionized the art of RF and microwave design, these tools are only as good as the component values they employ. The increased demand for reduced design cycles and easier to manufacture systems has generated the need for accurate and reliable tools. Current circuit simulation tools use component models that do not always account for parasitic coupling, non-standard shapes, and numerous other effects present in a real world design. A general and more efficient approach to creating such high-accuracy models is proposed herein. We propose to develop a neural network-based tool to provide accurate and complete component models needed for efficient and accurate circuit simulation. The advantage of this approach is that the models are universal, highly efficient, and simple to use. Under this Phase I project, we propose to investigate the usefulness of neural network-based models for a particular circuit simulation task. We will use state of the art electromagnetic field simulators to create the data needed to train the neural networks to simulate certain circuit elements that are widely used today. The Phase I work will show the potential of this approach for use in future microwave CAD tools.