Ansys Tests Laser-Based Compute for Faster Physics Simulations and Optimization Problems

While the world remains captivated by artificial intelligence, high-performance computing (HPC) continues to play a crucial role in various industries, from aerospace to materials science. Ansys, a leading engineering simulation software company, is constantly pushing the boundaries of HPC to handle increasingly complex tasks with greater efficiency. One of the company's latest ventures involves testing a laser-based compute platform from Israeli startup LightSolver, which promises significant speedups in solving non-polynomial (NP-hard) equations—a common challenge in physics simulations. Ansys's Chief Technology Officer, Prith Banerjee, emphasized that while current HPC technologies, including shared memory, message passing, and GPUs, are advancing, the demands of modern engineering are growing exponentially. This necessitates the exploration of alternative computing methods. NP-hard problems, which are pervasive in physics simulations, often require approximations due to their computational complexity. Traditional methods involve breaking down these problems into a series of binary decisions, which can be time-consuming and resource-intensive. In contrast, LightSolver's Laser Processing Unit (LPU) utilizes massive parallelism to solve these problems much more quickly. How the LPU Works The LPU is essentially a coupled laser array where each laser's phase and amplitude can be modulated to encode a specific problem. The laser beams are then directed into a cavity with two mirrors facing each other. As the beams travel back and forth between the mirrors, they interfere with one another, creating patterns that represent potential solutions. A gain medium inside the cavity amplifies the laser light, and a detector camera captures the most amplified pattern, which corresponds to the best solution. This process is completed in microseconds, offering a significant advantage over traditional computing methods. LightSolver CEO Ruti Ben-Shlomi describes the LPU's operation as akin to tossing stones in a pond. Just as the ripples from multiple stones interact to form a single wave, the laser beams in the LPU create interference patterns that converge on the optimal solution. The company's current prototype features 100 lasers, but it aims to increase this to 200 by 2027 and 1,000 by 2029. To tackle large-scale problems, LightSolver envisions clusters of LPUs working together to distribute the workload. Practical Application and Results Ansys integrated LightSolver’s LPU emulator into its physics simulation suite, LS-Dyna, to test its performance on graph partitioning. Graph partitioning is a technique used to divide large simulations into smaller, manageable parts to enable parallel processing. The goal is to minimize the number of edges that cross the partition, known as the min-cut, to enhance efficiency. During trials, the LPU emulator achieved better solutions in 80% of cases, leading to simulations running approximately 15-20% faster. While this might seem modest, Banerjee highlighted that a 20% speedup in LS-Dyna is highly significant, especially given the compute-intensive nature of physics simulations. Banerjee believes that the LPU's efficiency will be even more pronounced in other workloads, such as cell placement in electronic design automation (EDA) software. This is particularly relevant as Ansys is currently in the process of merging with Synopsys, a major player in EDA solutions. Cell placement involves optimizing the layout of electronic components to minimize power consumption and improve performance, which is a classic NP-hard problem where the LPU's parallel processing capabilities could offer substantial benefits. Future Prospects and Broader Applications Beyond graph partitioning, LightSolver is exploring how its LPU can accelerate partial differential equations (PDEs), which are fundamental in computational fluid dynamics and finite element analysis. These applications often require solving equations over a large number of grid points, and the LPU's ability to handle more grid points per laser array could lead to even more dramatic improvements in simulation speed and accuracy. For instance, by 2027, LightSolver aims to develop an LPU capable of solving PDEs with 100,000 grid points, and plans to scale this to a million grid points by 2029. This scaling approach is different from the LPU's handling of NP-hard problems, as it focuses on increasing the resolution of the simulations rather than the number of variables. This makes the LPU particularly promising for simulations that require high levels of detail and precision. Industry Insights and Company Profiles Industry insiders view LightSolver's technology as a potential game-changer in the field of HPC, especially for tasks that are computationally intensive and traditionally slow. The LPU's ability to operate at room temperature and use conventional principles makes it more practical and accessible compared to quantum computing, which typically requires specialized, cryogenic environments. Synopsys, the company set to acquire Ansys, stands to benefit significantly from the integration of LPU technology. With a strong presence in EDA software, Synopsys could leverage the LPU to optimize processes that are currently limited by computational constraints, thereby enhancing its offerings and maintaining its competitive edge in the market. Overall, LightSolver’s LPU represents a promising step forward in alternative computing technologies, offering a balance between speed, efficiency, and practicality that could reshape the landscape of HPC and simulation software.