The optimized copper fin designs are produced via ECAM, enabling micro-scale features as fine as 30 micrometers that standard manufacturing cannot achieve, and offer superior thermal conductivity compared to aluminum alloys.

The work was published in Cell Reports Physical Science on May 7, 2026, framing a pathway to bridge computational design with manufacturing to advance energy-efficient cooling for AI and data centers.

The innovation employs electrochemical additive manufacturing (ECAM) to produce high-resolution copper fins with a topology-optimized, tree-like design that maximizes heat transfer and minimizes flow resistance.

A team from the University of Illinois Urbana-Champaign used topology optimization and electrochemical additive manufacturing to create pure copper cold plates for direct-to-chip liquid cooling in data centers.

Researchers have developed an enhanced direct-to-chip cooling technology using optimized copper cold plates to improve data center cooling efficiency.

If scalable, this technology could lessen data center energy demand and support grid stability as computing power grows, contributing to a more sustainable tech infrastructure.

The next phase involves testing the plates on real server chips and partnering with hyperscale cloud providers to evaluate performance in large-scale deployments.

Finite design optimization involved topology optimization over 1,000 iterations to identify a fin geometry that enhances heat flow and efficiency.

Laboratory tests show the new cold plates deliver up to 32% better cooling and reduce pumping energy by 68%, potentially lowering data center cooling energy from about 30% of total consumption to as low as 1.1% of total usage.

The study highlights that current air cooling is insufficient for modern high-power chips, and direct-to-chip liquid cooling with optimized copper plates could be scalable beyond servers to other electronics and industrial cooling challenges.

Preliminary estimates suggest a 1 GW data center using the optimized cold plates would require roughly 11 MW for cooling, a substantial reduction from current cooling energy needs.

The researchers estimate that with the new plates, a 1 GW data center could transition from 550 MW used for cooling to around 11 MW, representing a substantial energy saving and enabling higher chip performance without overheating.