IN Brief:
- Siemens Digital Industries Software has introduced Simcenter PhysicsAI for Simcenter STAR-CCM+.
- The add-on uses geometric deep learning to create AI reduced-order models from CFD simulation data.
- AI-assisted simulation is moving into design exploration as electronics products become more thermally and mechanically constrained.
Siemens Digital Industries Software has expanded its Simcenter portfolio with Simcenter PhysicsAI, an add-on for Simcenter STAR-CCM+ designed to accelerate computational fluid dynamics design exploration.
The software uses geometric deep learning to build AI reduced-order models from existing CFD simulation data. Engineers can use those models to evaluate new geometry variations inside the simulation environment, while retaining access to high-fidelity CFD as the validation reference.
Siemens says the approach can reduce design exploration time from days to minutes, allowing teams to assess more design variants with lower computational demand than traditional CFD workflows. The system can be trained from historical simulation datasets, design-of-experiments studies, and newly generated results.
Functions include rapid design-space exploration, reuse of existing simulation data, validation metrics, GPU-accelerated inference, and optimisation-study support. The tool is available as an add-on to Simcenter STAR-CCM+ and forms part of Siemens’ wider simulation, high-performance computing, and AI engineering software work.
Electronics systems are increasingly shaped by airflow, heat spreading, enclosure design, board density, connector placement, power-conversion losses, battery behaviour, and acoustic or environmental constraints. A PCB can meet electrical requirements while still failing at enclosure level because heat cannot be removed, airflow is blocked, or the mechanical design leaves too little room for an effective thermal path.
Compact industrial controllers, power supplies, edge-AI computers, automotive electronics, optical modules, and data-centre hardware all face the same compression of design margins. Electrical, mechanical, and thermal decisions now overlap earlier in development, which places more demand on simulation before a physical prototype is built.
AI-assisted reduced-order modelling is a response to the cost of exploring those trade-offs. Full CFD remains necessary for trusted validation, but running every early-stage option through a complete solver workflow can be slow and expensive. A validated AI model can help screen weaker options, narrow the design space, and preserve high-fidelity simulation for the decisions that determine the finished product.
The same shift is visible in semiconductor design automation, where Siemens and TSMC have extended AI-enabled design work around advanced-node verification and implementation. PhysicsAI sits in engineering simulation rather than EDA sign-off, but both developments move AI into controlled workflows where validation, traceability, and design discipline remain essential.
Trust will determine adoption. Engineers need to know when an AI-reduced model is accurate enough for design exploration and when it must be checked against a deterministic solver. In safety-critical, high-power, or high-cost products, faster simulation only earns its place if the engineering basis remains defensible.
PhysicsAI therefore fits best as an acceleration layer for early-stage iteration, sensitivity analysis, and design-space reduction. As electronics designs become more constrained by heat, space, and system integration, the early screening phase is becoming a stronger determinant of product quality, cost, and development time.
