Energy-efficient thermal management of 11 kW multi-chip arrays via biomimetic hierarchical embedded liquid cooling

The rapidly increasing power demands of artificial intelligence (AI) and high-performance computing (HPC) systems have driven thermal design power (TDP) toward the 10kW scale, rendering high heat flux dissipation and thermal non-uniformity critical bottlenecks for system reliability. In this work, a biomimetic hierarchical embedded cooling (BHEC) architecture, inspired by coral morphology, is proposed to address the coupled challenges of ultra-high heat flux removal and temperature gradient mitigation in multi-chip module (MCM) packages. The proposed design integrates a cross-scale coolant distribution network with chip-level embedded microchannels, enabling coordinated flow regulation across chip, unit, and functional-cluster levels.

Experimental validation using a realistic thermal test vehicle demonstrates that the proposed BHEC module can dissipate a total package power exceeding 11kW while maintaining peak junction temperatures below 90℃. Under extreme operating conditions, the system accommodates a maximum heat flux of 377W·cm⁻² and restricts the maximum inter-chip temperature variation to within 15℃, representing a substantial improvement in thermal uniformity compared with conventional single-phase manifold cooling. Moreover, the proposed architecture achieves a coefficient of performance (COP) of 4114.29, highlighting its outstanding energy efficiency and suitability for sustainable data-center cooling.

These results establish a scalable and energy-efficient thermal management framework for next-generation high-density computing infrastructures.