AI Physics Simulations Enhance Understanding of Complex Systems

Technology Computer-Aided Design simulations are crucial for modern semiconductor manufacturing, enabling virtual manufacturing that allows engineers to design, build, and test transistors and integrated circuits digitally.

This approach significantly reduces development time from years to months and saves billions of dollars in experimental manufacturing costs. However, TCAD simulations are computationally intensive, often taking weeks to complete, which can delay manufacturing deadlines.

To address this challenge, NVIDIA introduced AI-augmented TCAD solutions, specifically through the PhysicsNeMo framework and NVIDIA Apollo. PhysicsNeMo allows developers to build high-fidelity surrogates for engineering and science simulations, while Apollo provides domain-specific, pre-trained models.

SK hynix, a leading memory chip manufacturer, is leveraging these AI physics tools to accelerate device and process simulations. By utilizing the PhysicsNeMo framework, engineers at SK hynix have advanced proprietary AI models that enhance innovation in semiconductor design and manufacturing.

TCAD is divided into two parts: process TCAD, which models the physical and chemical steps of chip manufacturing, and device TCAD, which models the electrical behavior of the final 3D structures. AI surrogate models developed with PhysicsNeMo dramatically reduce simulation times from hours to milliseconds, enabling engineers to explore a wider range of design possibilities.

The framework provides Python modules for scalable and optimized training and inference pipelines, allowing users to focus on domain expertise rather than developing from scratch. SK hynix engineers have made significant strides in developing AI surrogate models for the etching process, a critical step in semiconductor manufacturing.

They have adopted Graph Network-based Simulator architectures to enhance the accuracy of their models, addressing challenges such as data scarcity. This AI-augmented TCAD is seen as a key enabler for research productivity in the semiconductor industry, allowing engineers to evaluate thousands of process cases from various recipe combinations.

The PhysicsNeMo framework is positioned as a powerful tool for TCAD application developers and AI physics researchers, facilitating rapid development of enterprise-scale Physics AI solutions.