Workflow
Process Design Kit (PDK)
icon
Search documents
Keysight Unveils Machine Learning Toolkit to Accelerate Device Modeling and PDK Development
Businesswire· 2026-01-15 16:00
Core Insights - Keysight Technologies has launched a new Machine Learning Toolkit within its Device Modeling Software Suite, significantly reducing model development and extraction time from weeks to hours, which enhances Process Design Kit (PDK) delivery and Design Technology Co-Optimization (DTCO) applications [1][3]. Industry Overview - The semiconductor industry is experiencing rapid transformation due to advanced architectures like gate-all-around (GAA) transistors, wide-bandgap materials such as GaN and SiC, and heterogeneous integration strategies including chiplets and 3D stacking. These innovations, while improving performance, introduce complex modeling and parameter extraction challenges that traditional workflows struggle to address [2]. Product Features and Benefits - The Machine Learning Toolkit includes an ML optimizer and auto-extraction flows, which streamline the parameter extraction process from over 200 steps to fewer than 10, thereby accelerating PDK delivery and automating DTCO [3][7]. - The toolkit allows for global optimization of over 80 parameters in a single run, capturing secondary effects, temperature variations, and dynamic behaviors, which enhances predictive accuracy across various domains [7]. - The automated workflow integrates seamlessly with Keysight's Device Modeling platform, supporting Python-based customization and robust automated modeling flow [7]. - The solution is scalable across various technologies, including FinFET, GAA, GaN, SiC, and bipolar devices, ensuring repeatable and reusable flows for multiple process nodes [7]. - Improved DTCO efficiency is achieved by enabling faster feedback loops between device and circuit design, reducing PDK development cycles from weeks to days [7]. Company Perspective - Keysight's General Manager of EDA emphasized that AI/ML is transforming traditional workflows in compact modeling, allowing customers to deliver more predictive, higher-quality models in significantly less time, thus maintaining a competitive edge in the semiconductor market [5].