Multi-die chips, referred to as three-dimensional built-in circuits, or 3D-ICs, signify a revolutionary step in semiconductor design. The chips are vertically stacked to create a compact construction that enhances efficiency with out growing energy consumption.
Nevertheless, as chips change into denser, they current extra complicated challenges in managing electromagnetic and thermal stresses. To know and handle this, superior 3D multiphysics visualizations change into important to design and diagnostic processes.
At this week’s Design Automation Convention, a world occasion showcasing the newest developments in chips and methods, Ansys — an organization that develops engineering simulation and 3D design software program — will share the way it’s utilizing NVIDIA expertise to beat these challenges to construct the following era of semiconductor methods.
To allow 3D visualizations of simulation outcomes for his or her customers, Ansys makes use of NVIDIA Omniverse, a platform of utility programming interfaces, software program improvement kits, and providers that allows builders to simply combine Common Scene Description (OpenUSD) and NVIDIA RTX rendering applied sciences into present software program instruments and simulation workflows.
The platform powers visualizations of 3D-IC outcomes from Ansys solvers so engineers can consider phenomena like electromagnetic fields and temperature variations to optimize chips for sooner processing, elevated performance and improved reliability.
With Ansys Icepak on the NVIDIA Omniverse platform, engineers can simulate temperatures throughout a chip based on totally different energy profiles and ground plans. Discovering chip hot-spots can result in higher design of the chips themselves, in addition to auxiliary cooling units. Nevertheless, these 3D-IC simulations are computationally intensive, limiting the variety of simulations and design factors customers can discover.
Utilizing NVIDIA Modulus, mixed with novel methods for dealing with arbitrary energy patterns within the Ansys RedHawk-SC electrothermal information pipeline and mannequin coaching framework, the Ansys R&D group is exploring the acceleration of simulation workflows with AI-based surrogate fashions. Modulus is an open-source AI framework for constructing, coaching and fine-tuning physics-ML fashions at scale with a easy Python interface.
With the NVIDIA Modulus Fourier neural operator (FNO) structure, which may parameterize options for a distribution of partial differential equations, Ansys researchers created an AI surrogate mannequin that effectively predicts temperature profiles for any given energy profile and a given ground plan outlined by system parameters like warmth switch coefficient, thickness and materials properties. This mannequin gives close to real-time outcomes at considerably lowered computational prices, permitting Ansys customers to discover a wider design area for brand spanking new chips.
Following a profitable proof of idea, the Ansys group will discover integration of such AI surrogate fashions for its next-generation RedHawk-SC platform utilizing NVIDIA Modulus.
As extra surrogate fashions are developed, the group can even look to reinforce mannequin generality and accuracy by way of in-situ fine-tuning. This may allow RedHawk-SC customers to learn from sooner simulation workflows, entry to a broader design area and the power to refine fashions with their very own information to foster innovation and security in product improvement.
To see the joint demonstration of 3D-IC multiphysics visualization utilizing NVIDIA Omniverse APIs, go to Ansys on the Design Automation Convention, operating June 23-27, in San Francisco at sales space 1308 or watch the presentation on the Exhibitor Discussion board.
