EPC releases white papers on GaN power conversion for AI data centers and motor drives

Efficient Power Conversion (EPC) has published two new technical white papers addressing power conversion challenges in two fast-growing markets: AI data centers and high-performance multi-axis motor drives.

The first paper, titled "GaN-Based Power Delivery in AI Data Centers: High-Density Power Conversion Architectures from 800 VDC to Point-of-Load," examines the industry's shift toward 800VDC power distribution as AI server racks scale from tens of kilowatts up to hundreds of kilowatts and even megawatt-scale compute clusters. At 800VDC, a 200 kW rack requires only one-sixteenth of the current needed at a traditional 48V distribution bus, which reduces the required conductor size, resistive power losses (known as I²R losses), and cooling needs.

The paper explores three architectures for converting 800VDC down to lower 50V, 12.5V, and 6.25V power rails, focusing particularly on modular Input-Series, Output-Parallel (ISOP) LLC converters. By splitting the high-voltage input across multiple modules, this architecture allows lower-voltage GaN transistors, which generally offer a better performance-to-resistance tradeoff (referred to as figure of merit) than higher-voltage devices, to handle high-ratio voltage conversion at high switching frequencies.

Practical test results demonstrate strong performance potential: an 800V-to-50V implementation reaches 98.6% peak efficiency, while the EPC91123 800V-to-12.5V converter achieves 98.2% peak efficiency in a compact 8mm profile. A 6kW 800V-to-6.25V architecture reaches 98% peak efficiency and 97% efficiency at full load.

Jason Zhang, vice president of DC-DC marketing and system engineering at EPC, said that as GPU power requirements continue rising, power conversion needs to move closer to the processor while minimizing the number of conversion stages involved. He said EPC's seventh-generation GaN technology, combined with modular ISOP architectures, provides a practical path from an 800VDC power backbone down to the very low bus voltages and extremely high currents required by next-generation AI processors.

The second white paper, titled "From Device to Platform: EPC23108/09/10/11 100 V Integrated GaN Power Stages Simplify Multi-Axis Motor Drive Design," addresses the growing complexity of motor drive systems used in humanoid robots, drones, surgical robotics, and precision motion control systems.

EPC's EPC231xx chip family integrates high-side and low-side GaN transistors, gate drive circuitry, level-shifting components, a control interface, and protection features into a single monolithic 100V ePower Stage IC. The accompanying EPC91128 through EPC91131 three-phase BLDC/PMSM motor inverter evaluation platforms give engineers a complete environment to test current handling, sensing, protection, thermal performance, and different pulse-width modulation (PWM) control interfaces.

These platforms operate across a recommended voltage range of 14-80V, with 48V being typical, and support switching frequencies from 20 to 150 kHz. Complementary PWM control versions offer maximum flexibility, while single-PWM-input-pin versions reduce the number of control lines and simplify wiring, an increasingly valuable benefit as robotic systems scale up to include dozens of individual motors.

Marco Palma, vice president of motor drive marketing and system engineering at EPC, said that in multi-axis robotics, power density is only part of the overall challenge. He said integration also needs to simplify control, protection, sensing, thermal management, and wiring, and that the EPC231xx family and its evaluation platforms give engineers a validated starting point that can be efficiently replicated across many motor axes.

Together, the two white papers illustrate how GaN technology is moving beyond simple component-level performance improvements toward enabling complete power architectures, spanning everything from multi-kilowatt AI infrastructure down to compact, highly integrated motor drive systems. Both white papers, WP019 covering the motor drive platform and WP020 covering AI data center power delivery, are available for download in both English and Chinese.