アイクリスタル List of Products and Services

In power semiconductors like GaN, even the slightest roughness on the nanometer scale (one-millionth of a millimeter) can affect performance, making crystal processing after creating the cylinder essential in traditional methods. To find appropriate experimental conditions, at least two sets of conditions must be tested for each factor just to observe trends. Therefore, for five factors, a minimum of 32 experiments is required. Based on those trends, dozens of experimental conditions are tested to find the combination of factors that leads to the desired results. As a result, traditional optimization methods centered on experiments required a significant number of trials. In contrast, the approach taken by Aicrystal, which learns from results, explores, and suggests conditions, was able to reduce the number of experiments to just 19. Moreover, this approach achieved a level of processing precision that allowed for the omission of subsequent processes, and the benefits of not needing additional capital investment are significant. The conditions deemed appropriate in this case were combinations that had never been tried by engineers, illustrating the value of collaboration between AI and humans. *For more details, please refer to the PDF document or feel free to contact us.*

• Scientific Calculation and Simulation Software
• others

"GaN (gallium nitride)" is a power semiconductor that is expected to be adopted as a power source for 5G base stations. It is commonly used as a thin sliced disk wafer, but the initial manufacturing state is cylindrical. The GaN cylinder emits gas, which serves as the raw material for GaN, from holes at the bottom, allowing the crystals to grow. In reality, there are over 100 holes, leading to infinite combinations. When simulating this, it takes an astonishing 6 hours for just one condition. If you run 1000 variations, it consumes 6000 hours, which equals 250 days, taking a significant amount of time. This is where "Process Informatics" from AICrystal comes into play. By training AI with the results of 1000 randomly generated simulations, the calculation time is drastically reduced. The time required is just 1 second per condition. Thanks to the significant reduction in calculation time enabled by AI, it is now possible to compute 10,000 conditions in just about 3 hours. This optimization allows for the gas conditions and hole arrangements to be fine-tuned, successfully tripling the growth rate. *For more details, please refer to the PDF document or feel free to contact us.*

• Scientific Calculation and Simulation Software
• others