🤖 AI Summary
Microsoft announced a lab-scale breakthrough in microfluidic cooling that etches tiny liquid channels directly into the back of silicon chips, removing heat up to three times more effectively than today’s cold-plate systems and cutting peak silicon temperature rise by about 65% in tests (varies by chip and workload). The team demonstrated the system on a server running a simulated Teams workload, used AI to map chip hot spots and steer coolant precisely, and collaborated with Swiss startup Corintis to evolve a bio-inspired channel topology (leaf/wing vein patterns) that outperformed straight channels. Practical engineering work included leak-proof packaging, coolant formulation, etching methods and multiple design iterations to balance channel depth against clogging and silicon strength.
For AI/ML infrastructure the implications are substantial: bringing coolant into the silicon removes insulating layers that limit cold-plate performance, enabling higher power density, safer overclocking for spiky workloads, tighter rack/server packing, better PUE and potentially lower operational costs and carbon footprint. Microfluidics also opens paths to new chip architectures (e.g., stacked 3D designs with fluid-flow “pillars”) that today’s cooling limits make impractical. Microsoft is now exploring integration with its own Cobalt/Maia chips and fabrication partners to move the tech toward production, emphasizing a systems-level approach across silicon, coolant, server and datacenter.
Loading comments...
login to comment
loading comments...
no comments yet