Core Viewpoint - InnoSwitch (英诺赛科) announces collaboration with NVIDIA to support the 800 VDC power architecture, which is expected to enhance efficiency, power density, and reduce energy consumption and CO2 emissions in AI data centers [1] Group 1: 800 VDC Power Architecture - The transition from 48V to 800V in rack power systems can reduce current by 16 times, significantly decreasing I²R losses and minimizing copper demand [1] - Traditional AI systems operating at 48V face challenges such as inefficiency and high copper consumption, with over 45% of total power used for cooling [1] - The 800 VDC architecture is positioned as a solution to support the transition from kilowatt to megawatt-level AI clusters [1] Group 2: GaN Technology Advantages - InnoSwitch's third-generation GaN technology offers a decisive advantage, reducing driving losses by 80% and switching losses by 50% compared to SiC at 800V input, leading to an overall power reduction of 10% [3] - Only 16 GaN devices are needed to achieve the same conduction losses as 32 silicon MOSFETs at 54V output, doubling power density and reducing driving losses by 90% [3] - The use of GaN materials in the low-voltage conversion stage of 800 VDC can lower switching losses by 70% and increase power output by 40% within the same volume compared to existing silicon MOSFET architectures [3] Group 3: Full-Stack GaN Solutions - As the only full-stack GaN supplier, InnoSwitch is capable of mass-producing GaN solutions from 1200V to 15V, providing a complete solution from 800V to 1V [5] - InnoSwitch's GaN technology has demonstrated superior reliability, passing rigorous stress tests and ensuring a high-performance lifespan of over 20 years for data center products [5] - The integration of the 800 VDC power architecture with InnoSwitch's GaN technology is set to revolutionize AI data centers, enabling a leap from kilowatt to megawatt racks, fostering a new era of efficient, high-performance, and environmentally friendly AI computing [5]
英诺赛科为800 VDC电源架构提供全GaN电源解决方案，赋能新一代AI Factories