Nvidia’s AI-defined model targets semiconductor process simulation

Nvidia says Applied Materials can run topography simulations up to 35 times faster. Its Siemens integration delivers 17-fold acceleration; Lam Research demonstrates millisecond plasma prediction.

By Marcus Bennett5 min read

Why it matters

  • Nvidia’s November 18, 2026 fabrication model combines four technologies: accelerated computing, physics-based AI, AI agents and digital twins.
  • COMSOL uses Nvidia’s cuDSS library for up to 10 times faster semiconductor manufacturing simulations.
  • Applied Materials’ ACE+ platform reaches up to 35 times faster topography simulations with Nvidia PhysicsNeMo.
  • Siemens’ cuLitho integration with Calibre delivers up to 17 times acceleration, according to Nvidia.
  • Nvidia says Lam Research predicts plasma behavior in milliseconds rather than days, while Advantest cuts model-selection time from hours to minutes.

By Marcus Bennett

Nvidia says Applied Materials’ ACE+ multiphysics platform can run topography simulations (surface-shape calculations) up to 35 times faster when paired with Nvidia PhysicsNeMo. Applied Materials is working with Nvidia on atomic-scale materials discovery and process development.

On November 18, 2026, Nvidia outlined what it calls an “AI-defined model” for semiconductor fabrication. The approach combines accelerated computing (high-speed processing for engineering workloads), physics-based AI, AI agents (software that executes task sequences) and digital twins (virtual representations of physical systems).

Nvidia says the model targets engineering analysis that can take days and physical production experiments that can be costly.

Nvidia said semiconductor manufacturers face pressure from rising demand for advanced chips and increasingly complex processes. In that environment, even small errors in chip yield or process quality can have significant commercial consequences, according to Nvidia.

What does Nvidia’s “AI-defined model” include?

Nvidia organized the proposal around four elements. Each links a different point in the engineering process, from evaluating a process change to testing a factory-wide configuration.

  • Accelerated computing: Nvidia says this lets engineers test more alternatives before deciding which changes to try in physical production.
  • Physics-based AI: Nvidia says these models predict how materials and processes behave without running a full simulation every time.
  • AI agents: Nvidia says they automate sequences of engineering tasks that engineers have handled manually.
  • Digital twins: Nvidia says these virtual environments test equipment configurations, factory layouts and process changes before factory-floor implementation.

How much faster can chip-process simulation run?

COMSOL, a software provider, is using Nvidia’s cuDSS library to speed up semiconductor manufacturing simulations by up to 10 times, Nvidia said.

Nvidia reported a second Applied Materials result: the Ginestra simulation platform runs up to 10 times faster than CPU-only approaches (conventional-processor runs) when integrated with cuDSS. The up-to-35-times ACE+ figure uses PhysicsNeMo.

Siemens is integrating Nvidia cuLitho into its Calibre lithography software. Nvidia says this delivers up to 17 times acceleration on key simulation workloads, giving engineers more scope to verify designs before a wafer reaches print.

Within accelerated computing, Nvidia identifies multiphysics simulation—a model combining heat transfer, electrical behavior and mechanical stress—as a central use case. Nvidia says accelerated computing lets engineers test more alternatives before choosing which changes to try in physical production.

Computational lithography, which determines how circuit patterns are printed onto silicon, can strain conventional processors at production scale, according to Nvidia. Its calculations are highly intensive, and Siemens is integrating cuLitho into Calibre for that category of work.

For manufacturers, Nvidia says faster simulation creates room to compare more options before committing resources to physical trials. The company positions the approach as a way to manage greater process complexity without relying solely on slower, more expensive physical experiments.

What does physics-based AI change?

Nvidia argues physics-based AI models can predict how materials and manufacturing processes behave without a full simulation on every query. The company says the benefit is greatest for expensive calculations that remain demanding even on accelerated hardware.

Plasma modeling is one example. Nvidia says plasma is used in semiconductor etch processes, where models require equations that capture interactions among electrical, chemical and thermal effects.

Lam Research has shown that the Nvidia PhysicsNeMo framework can predict plasma behavior in milliseconds, compared with days for a full simulation, according to Nvidia. Nvidia says this gives engineers a faster way to explore process conditions without rerunning the complete model each time.

What do AI agents automate?

Nvidia says AI agents can automate sequences of intermediate engineering work that engineers previously handled manually. Engineers still decide. Many intermediate steps can consume hours, according to the company.

Advantest is using Nvidia NeMo Switchyard AI agents to select which AI model applies to a particular task. Nvidia says the deployment cuts response time from hours to minutes.

What can factory digital twins simulate?

Digital twins let teams test equipment configurations, factory layouts and process changes before carrying them out on the production line, Nvidia says. Applied Materials uses Nvidia Omniverse libraries to build physically accurate digital twins of entire fabrication plants, according to Nvidia.

Nvidia says those models simulate material movement, identify bottlenecks and assess layouts, throughput, cost and process control. The proposed scope extends beyond individual manufacturing steps to the operation of the factory as a whole.

Nvidia says the growing complexity of fabrication plants and the rising cost of unsuccessful process changes make that factory-wide scope more relevant. In its model, a digital twin provides a virtual test bed for changes before production-line execution.

What does the connected engineering loop change?

Nvidia presented the technologies as more useful when linked in a “closed engineering loop.” In that model, production data informs simulations and AI models, which then test alternatives before engineers make decisions on the factory floor.

Nvidia says the aim is shorter feedback cycles across semiconductor manufacturing. The company says engineers should identify problems earlier, evaluate more options and act on production insights sooner.

Nvidia positioned the push as a response to chipmakers and suppliers managing greater process complexity without relying solely on slower, more expensive physical trials. Under Nvidia’s account, production data feeds virtual tests, and engineers use the results when deciding what to do on the factory floor.

What could faster chip manufacturing change?

Nvidia’s outline joins five named implementations across the manufacturing workflow: COMSOL for simulation acceleration, Applied Materials for process development and plant twins, Siemens for lithography, Lam Research for plasma prediction and Advantest for model-selection agents.

Nvidia’s claimed gains range from 10 times to 35 times for the cited simulation platforms. Nvidia also cites 17 times acceleration for Siemens, millisecond plasma predictions against days-long full simulations and an hours-to-minutes reduction at Advantest.

For chipmakers, Nvidia frames the payoff as a shorter path from analysis to production. It says faster feedback can identify problems earlier, evaluate more options and reduce reliance on slow, expensive physical trials. Those gains matter because Nvidia says small yield or process-quality errors carry significant commercial consequences in that environment.

Nvidia did not provide deployment dates or independent performance data in the announcement. The announced speedups therefore remain claims that must be established in operating fabrication plants.

Nvidia’s proposal would make simulation a connective layer between production data and factory decisions, not merely a faster calculation. Its value will be measured by how much of the claimed 10-to-35-times acceleration survives at factory scale, where Nvidia says small process errors already carry commercial consequences.

Source: itbrief.com.au

Share this article:

More from Marcus Bennett

Marcus Bennett

Show full bio

Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

718 articles

Related articles

  1. Rockwell Brings NVIDIA Omniverse Into Emulate3D for Factory-Scale Simulation
  2. Nvidia backs Simplismart, software that runs AI on fewer GPUs
  3. AI's Power Hunger Is Reviving Optical Computing—70 Years After the Idea
  4. Nvidia launches open-weight Nemotron models for enterprise agents and AI PCs
  5. Industrial AI Moves From Demos to Tightly Controlled Deployment

« Previous articleNext article »