Enabling Automatic Model Generation of RF Components: A Practical Application of Neural Networks

Abstract

Electronic computer-aided design (CAD), or electronic design automation (EDA), has achieved great importance in RF integrated circuit (RFIC) design in the past decade as semiconductor technology evolves to meet the needs of high-speed and high-reliability communication. Today, EDA tools are considered essential for a standard design flow. In many cases, the first-pass success of a modern RF design relies on the proper use of software tools, along with an accurate representation of the involved semiconductor technologies. For example, technology CAD (TCAD) can model the semiconductor device-fabrication process steps (such as diffusion and ion implantation) and simulate/model device operation by solving fundamental physics equations (with inputs such as geometrical dimensions and the doping profiles of the device). A process design kit for a specific technology contains a set of components and models that can be used for circuit simulation and layout design. In the competitive RF power business, where a short design cycle is desired and time to market is critical, a new device?s technology may need to be evaluated through a standard design flow, even before the parts are fabricated for testing. This has been made possible with the advent of EDA tools and advanced modeling techniques [1].