This new chip could slash data center energy waste

· Source: Artificial Intelligence News -- ScienceDaily · Field: Technology & Digital — Cloud Computing & IT Infrastructure, Emerging Technologies & Innovation, Artificial Intelligence & Machine Learning · Depth: Expert, short

Summary

Engineers at UC San Diego have developed a new chip design for DC-DC step-down converters that could significantly improve energy efficiency in data centers, particularly for powering GPUs. The innovation addresses the challenge of converting high input voltages (e.g., 48 volts) to much lower voltages (1-5 volts) required by processors. Unlike traditional designs that rely on magnetic inductors, this new approach utilizes piezoelectric resonators in a hybrid configuration with commercially available capacitors. A prototype chip successfully converted 48 volts to 4.8 volts with a peak efficiency of 96.2 percent and delivered four times more output current than previous piezoelectric designs. While still in early development, this technology offers a promising alternative to current power conversion systems, though challenges remain in materials, circuit refinement, and packaging for real-world integration.

Key takeaway

For AI Scientists designing next-generation data center infrastructure, this research indicates a viable path to significantly reduce energy waste. Your focus should shift towards exploring piezoelectric-based power conversion as a long-term solution, even as current integration challenges like packaging and material refinement are addressed. Consider how these high-efficiency converters could impact overall system thermal management and power delivery architectures in future GPU clusters.

Key insights

A hybrid piezoelectric converter design offers superior efficiency and power density for data center voltage conversion.

Principles

Method

Combine a piezoelectric resonator with small, commercially available capacitors in a specific configuration to create multiple energy pathways and reduce strain.

In practice

Topics

Best for: AI Scientist, AI Hardware Engineer, AI Architect, Research Scientist

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence News -- ScienceDaily.