Particle.news

Download on the App Store

Microsoft Unveils In‑Chip Microfluidic Cooling for AI GPUs, Claims 3x Gains Over Cold Plates

The prototype routes coolant through hair‑thin channels etched into silicon, and the company is now shifting to reliability testing and manufacturing integration.

Overview

  • Microsoft reports lab results showing up to a 65% reduction in peak silicon temperature rise and as much as threefold improvement versus advanced cold plates, depending on workload and configuration.
  • Coolant flows through micro‑channels etched on the back of the chip to target hotspots closer to the heat source than conventional cold plates.
  • Designs were shaped using AI and bio‑inspired geometries that resemble leaf veins, developed in partnership with Swiss startup Corintis.
  • Demonstrations included lab tests and a server running simulated Teams workloads, though the company has not disclosed details about the coolant used.
  • Engineers highlight the need for leak‑proof packaging and chip‑fab integration before wider use, with potential benefits including denser racks, safer overclocking, 3D chip architectures, and lower datacenter energy costs.