Keysight Technologies, Inc. announced the release of a new machine learning toolkit included in the latest Keysight device modeling software suite. This new solution reduces model development and extraction time from weeks to hours, enabling faster delivery of process design kits (PDKs) and design technology collaborative optimization (DTCO) applications. Machine Learning Toolkit reduces model development time from weeks to hours.
The semiconductor industry is undergoing rapid transformation, driven by advanced architectures such as gate-all-around (GAA) transistors, wide bandgap materials such as GaN and SiC, and heterogeneous integration strategies such as chiplets and 3D stacking. While these innovations improve performance, they also create complex modeling and parameter extraction challenges. Traditional workflows rely on compact physics-based models and manual parameter extraction, requiring engineers to tune hundreds of interconnected parameters across multiple operating conditions, a process that can take weeks and often struggles to achieve optimal results.
As schedules become increasingly tight, faster, more predictive, and automated artificial intelligence/machine learning (AI/ML)-driven modeling solutions are essential. Keysight’s new Machine Learning Toolkit, featuring ML optimizers, automated extraction flows, and utilities within Device Modeling MBP 2026, tackles these challenges by introducing a framework that combines advanced neural network architectures with ML-based optimization. With this toolkit, automated extraction reduces parameter extraction steps from 200+ to less than 10, accelerating PDK delivery, automating DTCO, and reducing time to market.
Key features and benefits: Faster parameter extraction: Reduces hundreds of manual steps to 5-6 automatic steps, enables global optimization of 80+ parameters in a single run, and captures secondary effects, temperature fluctuations, and dynamic behavior. This solution eliminates the need for repeated manual adjustments and improves prediction accuracy across DC, RF, and large signal domains. Automated workflow: Seamlessly integrates with Keysight’s device modeling platform and supports Python-based customization and robust automated modeling flows.
Scalable across technologies: Workflows easily adapt to FinFET, GAA, GaN, SiC, and bipolar devices, ensuring repeatable and reusable flows across multiple process nodes. Keysight leverages AI/ML-driven modeling to help semiconductor companies accelerate innovation, reduce development risk, and stay competitive in rapidly evolving markets. Additional enhancements across other Keysight device modeling solutions include: Device Modeling MQA 2026: New rules related to aging model QA for OMI and MOSRA have been introduced.
Device Modeling WaferPro 2025: Introducing new remote control capabilities for remote low frequency noise testing with A-LFNA for increased flexibility and efficiency. A-LFNA 2026: Introduced new low-frequency noise stress test capabilities that enable seamless measurement from stress to noise tests. Resources: Simplify and automate compact model processes with machine learning.
White Paper: Accelerating Semiconductor Innovation with Machine Learning-Driven Modeling Modeling-Driven Modeling. Application note: Accelerating compact model parameter extraction with Machine Learning Optimizer. Blog: From equations to intelligence: integrating machine learning optimizers into compact model extraction.
