PowerLattice Voltage Regulator Boosts AI Energy Efficiency (spectrum.ieee.org)

PowerLattice, a startup specializing in power solutions for AI data centers, has announced the development of miniaturized voltage regulators that could significantly improve energy efficiency. Traditional GPU power delivery systems often lead to inefficiencies, requiring up to 1,700 watts to operate a GPU for large language models instead of the ideal 700 watts. PowerLattice claims its new chiplets can reduce power consumption by up to 50% and double performance per watt by moving the voltage conversion process closer to processors—within millimeters rather than centimeters—thereby minimizing energy loss due to resistance in low-voltage states. The innovation involves creating tiny chiplets using specialized magnetic alloys for inductors, enabling operation at frequencies up to 100 times higher than conventional solutions. This design not only conserves space but also enhances efficiency, allowing companies to mix and match the chiplets based on their specific needs. As demand for energy-efficient AI infrastructure grows, PowerLattice's technology presents a competitive edge amid larger incumbents like Intel. However, the company's ambitious claims regarding power savings are met with skepticism, necessitating real-time load management capabilities to realize those efficiencies fully. Despite challenges, the trend towards heterogeneous chiplet integration offers a promising market opportunity for startups focusing on advanced power solutions.