Semiconductor engineering depends on simulations that can take hours or days to resolve, limiting how quickly teams can evaluate materials, processes and chip designs. NVIDIA (NASDAQ: NVDA) said Samsung Electronics Co., Ltd. (KRX: 005930) is expanding its use of accelerated computing and AI physics across semiconductor research and development, with the goal of reducing that computational burden.
The collaboration covers materials research, process development and device engineering, according to NVIDIA. For enterprises that build advanced hardware, faster engineering analysis can change how many design options a team can test before committing fabrication resources. It can also bring simulation closer to the iteration speed expected in software-driven product development.
Samsung is working with NVIDIA cuISS and cuDSS for computations involving sparse linear systems, and the companies are extending GPU acceleration to process and optical simulations. NVIDIA also said Samsung’s sMOL platform has run selected quantum-chemical calculations up to five times faster end to end using the vendor’s cuEST technology. Those are company-reported results for selected workloads, rather than a universal measure of semiconductor-engineering performance.
The companies are also applying NVIDIA PhysicsNeMo to chip-scale thermal and mechanical analysis. NVIDIA described one AI-homogenization case that combined AI-predicted effective material properties, coarser simulation grids and GPU acceleration. In that specific case, the company said the workflow reduced a fine-grid solve on a 32-core CPU from about 18 hours to 41 seconds on a GPU-accelerated coarse-grid configuration—roughly a 1,600-fold speedup in a production engineering setting.
That comparison is striking, but it needs the right context. A coarser-grid AI-assisted approach does not necessarily substitute for every high-fidelity simulation, and the value of a faster calculation depends on the accuracy and validation requirements of the engineering decision. NVIDIA did not provide broad benchmark details for all Samsung workflows. Engineering teams will need to test whether an AI-based approximation preserves the fidelity required for their own materials, thermal and mechanical analyses.
Still, the combination points to a practical use of AI beyond text generation. Physics-informed models can help researchers estimate properties or behaviors that would otherwise require more expensive numerical computation. When paired with conventional simulation tools, the approach may allow teams to screen more candidates, focus detailed analysis where it matters and shorten the loop between design hypotheses and measured results.
NVIDIA said its Agent Toolkit could eventually link simulation tools and AI models, allowing engineering agents to run analyses and compare design options. That capability is framed as a future possibility rather than an available production workflow. Any enterprise use would require controls over data, model validation, tool access and review of engineering conclusions before they influence a product decision.
For technology leaders, the significance lies in the way compute infrastructure is moving upstream into engineering operations. Legacy workflows often treated simulation as a specialized, sequential step that competed for finite CPU resources. GPU acceleration and AI physics make it more feasible to treat simulation as an iterative, data-driven service—a shift that could favor platform vendors able to combine hardware, accelerated libraries and validated scientific workflows.