With NVIDIA accelerated computing running on Azure AI infrastructure powering the virtual machines, the collaborators said that the machines executed the simulations and identified optimal resources that balance cost with performance.
The overall result, they said, is seamless deployment, graphical interface access, scaling of distributed simulations, and post processing for large datasets in cloud environments.
Ansys’s Lumerical FDTD software simulates microLED acceleration performance across different GPU models on Microsoft Azure. Courtesy of Ansys.
According to Stefan Rusu, head of silicon photonics system design at TSMC, the size and complexity of the company’s multiphysics silicon solutions makes it a challenge to simulate all possible parameter combinations. The collaboration, he said, delivers accurate solutions in a fraction of the time.
The companies see the pilot as a solution to the design and fabrication of silicon photonic integrated circuits (PICs). The silicon PIC workflow is susceptible to minor missteps that may cause continuity challenges within chips, which can result in added cost and timeline setbacks up to several months.
According to the collaborators, deploying Lumerical FDTD on the cloud enables designers to identify optimal chip designs to account for the multiphysics challenges related to combining photonic circuits with electronic circuits.
