How Synopsys Is Transforming Semiconductor Engineering

Demand for AI and high-performance computing is driving the need for increasingly complex chip designs. Yet the simulations required to design and manufacture those chips have become some of the industry’s most demanding computing workloads. Computational lithography—the process of modeling how circuit patterns will be transferred onto silicon—is a striking example: A typical mask set can require 30 million or more hours of CPU compute time.

At advanced process nodes, shrinking features and increasingly intricate mask patterns require more computation to preserve accuracy and manufacturing yield. Scaling traditional CPU infrastructure to meet that demand increases cost, power consumption, and data center space while leaving engineering teams waiting longer for results.

For Synopsys, the challenge extended beyond making individual applications run faster. Its customers need a computing foundation that can support more complex simulations, faster design iterations, and tighter connections between design and manufacturing without sacrificing accuracy.

Synopsys responded by expanding its collaboration with NVIDIA and integrating accelerated computing and AI more deeply across its engineering software. NVIDIA CUDA-X™ libraries and NVIDIA Blackwell platforms accelerate existing solvers and simulation workloads, while NVIDIA AI technologies help engineers retrieve information, automate multistep work, and initialize complex simulations more efficiently.

The collaboration spans four complementary capabilities:

• Accelerated computing for computational lithography, materials discovery, circuit simulation, and power integrity analysis
• AI physics for thermal simulation and design exploration
• Agentic engineering for chip design, verification, and manufacturing workflows
• Digital twins for virtual development and system-level validation

Together, these capabilities allow Synopsys to address the same underlying constraint across different disciplines—the time and computing resources required to answer difficult engineering questions.

The computational lithography results point to a broader shift. Synopsys and NVIDIA are applying GPU computing across semiconductor engineering workflows, from atomic-scale materials research to circuit, power, and device simulation.

Materials Discovery

At the atomic scale, Applied Materials is working with Synopsys and NVIDIA to accelerate materials modeling. Early results using Synopsys QuantumATK optimized with NVIDIA cuEST show a potential 30x speedup for complex quantum chemistry workloads compared with open source models running on CPUs. That additional computing capacity can help researchers examine material behavior and reaction pathways that influence next-generation semiconductor processes.

Beyond quantum chemistry workloads, Synopsys is delivering significant advancements in computational materials simulation, leveraging NVIDIA accelerated computing to drive faster, more efficient research in materials and semiconductor design. By integrating Synopsys QuantumATK atomistic simulation with NVIDIA CUDA-X libraries and NVIDIA Blackwell architecture, Synopsys is enabling engineering teams to achieve up to a 15x improvement in time to results for density functional theory (DFT) and nonequilibrium Green’s function (NEGF) methods.

Northrop Grumman Microelectronics Center (NGMC) uses Synopsys QuantumATK with NVIDIA GPUs to accelerate the development of next-generation materials for semiconductor applications in both defense and commercial systems. By cutting development time from weeks to hours, this approach speeds up the innovation process by 6–10x and makes it possible to solve challenges that were once considered impossible.

Semiconductor Device Simulation

Synopsys Sentaurus Device (S-Device) uses NVIDIA cuDSS to accelerate sparse linear solving in physics-based device simulations. Published benchmarks on NVIDIA B200 GPUs demonstrated a 12x speedup versus comparable CPU infrastructure. Faster simulation helps engineers evaluate device behavior, explore more design options, and shorten technology development cycles.

Circuit Simulation

The same accelerated foundation is beginning to change other stages of semiconductor engineering. Synopsys PrimeSim simulations used to validate circuit performance are projected to run up to 18x faster using the NVIDIA GB200 Grace Blackwell Superchip compared with CPU-based models. Separately, Astera Labs achieved a 3.5x speedup running PrimeSim on NVIDIA B200 GPU-accelerated Amazon EC2 instances compared with CPU-only instances, shortening validation cycles for advanced connectivity designs.

Photonic Simulation

Ansys Lumerical, powered by accelerated computing, enables engineers to design and optimize photonic components and systems, including photonic integrated circuits (PICs), optical devices, metasurfaces, and advanced nanophotonic structures up to 100x faster. iPronics, a pioneer in optical connectivity solutions for AI data centers, achieved up to 13x faster solver acceleration leveraging NVIDIA H100 GPUs on AWS running Lumerical simulations.

Power Integrity Analysis

As chip designs grow more complex, ensuring reliable power delivery requires increasingly demanding analysis. Synopsys RedHawk-SC is using NVIDIA cuDSS to accelerate sparse-matrix calculations in power integrity workflows. Faster analysis helps engineers evaluate voltage drop, identify power delivery issues, and assess corrective changes earlier, supporting faster design closure.

Beyond accelerating existing solvers, Synopsys is applying AI physics to help engineers explore design alternatives faster. Synopsys is integrating NVIDIA PhysicsNeMo into its SeaScape platform, enabling teams to train AI surrogate models using high-fidelity simulation data. These models can predict thermal behavior for new design conditions, helping engineers evaluate more options and identify potential thermal challenges earlier. Synopsys and NVIDIA have demonstrated over 100x faster thermal simulation with this approach.

The Synopsys Autopilot Platform brings together domain-specific AgentEngineer solutions, trusted engineering context intelligence, reusable skills, and persistent memory to power long-horizon agents that can reason, plan, and execute across silicon-to-systems workflows—while its open, secure architecture gives customers flexibility across Synopsys, partner, and third-party models, infrastructure, tools, and agents. Integrated with NVIDIA NeMo Agent Toolkit, including NVIDIA Nemotron open models and data, Synopsys Agent Engineer supports more autonomous chip design flows.

In formal verification, Synopsys agents in development have boosted signoff depth and efficiency up to 72% while identifying critical bugs missed by manual review. NVIDIA is piloting AgentEngineer for AI-enabled formal verification.

The collaboration is also creating a bridge from semiconductor design to broader system engineering. Analog Devices (ADI) is augmenting an NVIDIA Isaac Sim™ environment, built on NVIDIA Omniverse™ libraries, with Synopsys physics to develop high-fidelity sensing simulations and robotic benchmarks. Synopsys and Analog Devices showcased a physical AI and robotics demonstration focused on robotic hand dexterity and sim-to-real development, using the Synopsys Motion solver. The solution combined ADI tactile and time-of-flight sensors, NVIDIA Isaac Sim, NVIDIA Isaac™ Lab, and Omniverse libraries, and Ansys Motion to create a high-fidelity digital twin used for training and validating robotic manipulation tasks.

Workloads that once required large CPU clusters or impractically long runtimes can be handled with fewer systems and faster iteration cycles. The result is not simply a faster version of the existing process. Engineers can evaluate more alternatives, use higher-fidelity models, and bring simulation insights into decisions earlier.

For Synopsys’ customers, accelerated computing and AI create more room to explore ambitious designs—from advanced chips and manufacturing processes to intelligent products and physical AI systems. For software makers building physical AI applications, NVIDIA libraries and models provide tools that frontier AI agents can use to prepare 3D scenes, connect simulation tools, and run tests—with physical accuracy, speed, and compute efficiency. Integrating these capabilities into their products helps developers and specialists spend less time assembling workflows and more time exploring scenarios and evaluating results.