Silvaco Integrates NVIDIA AI and Accelerated Computing for Semiconductor Digital Twins

Silvaco and NVIDIA announced a collaboration to integrate NVIDIA accelerated computing and AI with Silvaco’s physics-based simulation portfolio for digital twins in semiconductor design and manufacturing.

Silvaco is combining its physics-based modeling with NVIDIA’s accelerated computing, CUDA-X libraries, PhysicsNeMo, Omniverse libraries, and Nemotron open models. The effort aims to support development of digital twins that predict, optimize, and validate semiconductor systems.

The partnership focuses on several areas. Silvaco plans to use NVIDIA accelerated computing and CUDA-X libraries to speed up its semiconductor device, process, photonics, and multiphysics simulation tools. In one example, Silvaco completed a fully scaled 3D FDTD simulation of a photonic edge coupler with 3.2 billion mesh nodes on 32 NVIDIA GPUs connected with NVLink in under four hours. The workload did not converge on CPUs, and the result showed less than 0.15 dB difference between measurement and simulation.

Silvaco intends to use NVIDIA PhysicsNeMo to develop AI surrogate models that work alongside high-fidelity physics simulation. It also plans to link its digital twin environment with NVIDIA Omniverse libraries and NVIDIA Cosmos for visualization and simulation across semiconductor fabs, manufacturing systems, robotics platforms, and related infrastructure. Addit...