Siemens, NVIDIA Expand Agentic AI Collaboration for 3D IC Power and Thermal Design

Estimated reading time: 2 minutes

Siemens announced an expansion of its EDA agentic AI portfolio, developed in collaboration with NVIDIA. The latest updates combine the self-verifying agentic AI capabilities introduced at DAC 2026 with NVIDIA accelerated computing to address one of advanced packaging's most demanding hurdles: electrical, power, thermal and codesign for 3D ICs.

The new capability orchestrates Siemens’ electrothermal analysis portfolio into a single autonomous workflow. By leveraging Siemens’ Design Space Exploration and Optimization (DSE/DSO) technology, the workflow allows engineers to evaluate and validate hundreds of design alternatives against signoff-grade, physics-based solvers. In benchmark studies, this approach reduced engineering effort by up to 95 percent (reducing a multi-week process to a matter of hours), cut workflow setup time by up to 99 percent (typically, less than five minutes versus several days across seven tools) and enabled up to 50x greater design exploration coverage compared to conventional manual flows.

"Electrothermal codesign has traditionally required experts to coordinate multiple tools and manually iterate through a limited set of design alternatives," said Ankur Gupta, executive vice president, EDA IC Portfolio, Siemens. "By combining agentic AI with Siemens’ proven physics-based analysis, engineers can now evaluate up to 50 times more design scenarios, enabling faster and more confident decisions for the next generation of AI infrastructure."

“Power delivery and heat dissipation increasingly determine what’s possible in 3D chip design,” said Tim Costa, vice president and general manager of industrial and computational engineering at NVIDIA. “With NVIDIA Agent Toolkit and Nemotron-3.5-Lightning, Siemens’ agents coordinate electrical and thermal solvers to deliver up to 50x greater coverage in design space exploration with 95% less engineering effort. By diagnosing how power delivery, die placement and package materials contribute to missed targets, these agents make electrothermal analysis an integral part of chip development.”

AI factories are redefining the chip

AI factories are redefining what the silicon inside them must deliver: more compute, more memory bandwidth and dramatically improved thermal performance - all within an ever-smaller physical footprint. NVIDIA’s Rubin GPU illustrates this scale, pairing up to 50 petaflops of NVFP4 inference with 288 gigabytes of HBM4 per GPU that scales up into larger co-designed systems such as NVIDIA Vera Rubin NVL72. As compute and memory converge in these tightly integrated packages, power density rises and heat becomes concentrated. Electrothermal codesign can no longer be a late-stage check; it must become a first-order design requirement.

The Challenge: Manual complexity

Today, solving these challenges requires that specialists manually coordinate workflows across multiple tools, transfer data between domains and repeatedly iterate between electrical, power, thermal and extraction analyses. As complexity increases, engineers are often forced to limit the number of scenarios they evaluate given time and compute constraints, increasing the risk that critical bottlenecks remain undiscovered until late in the design cycle.

The Solution: Agentic 3D IC engineering workflow

Siemens is introducing a new model. An engineer defines the design objective and the Fuse EDA AI Agent - integrated with Intelligence Center X - orchestrates multiple Siemens tools. The agent runs coupled electrothermal analyses, identifies missed targets and diagnoses whether the root cause lies in power delivery, placement, materials or a combination of all three.

The agent uses analysis results to guide design-space exploration, evaluate tradeoffs and rerun promising configurations against the engineer’s targets. This iterative cycle helps teams assess more alternatives and identify power and thermal issues earlier in development.