A Review of Non-Uniform Load Distribution and Solutions in Data Centers: Micro-Scale Liquid Cooling and Large-Scale Air Cooling

Overview

The number of transistors on electronic components is increasing exponentially, leading to ultra-high heat flux (10^6-10^7 W/m²), and non-uniform load distribution on the chip and local hot spots in the thermal environment are key issues in a data centre. The cooling system's share of data-centre power consumption is about 38%, second only to IT equipment. Microchannel liquid cooling is an effective method to inhibit heat accumulation on the chip, and optimising air distribution is crucial for energy savings. The review covers microchannel heat sink (MCHS) structure, manifold and header layout, flow boiling and nanofluids, plus room-scale air distribution and rack-based cooling.

Findings: a cited double-layer cooling plate design with optimised headers and cross-flow reduced fluid maldistribution and lowered thermal resistance by 11.8%. A cited graphene nanofluid in a heat pipe increased thermal performance by 49.4%, though the review stresses that nanofluid stability limits practical use. A cited deep reinforcement learning controller for zonal thermal parameters reduced fan power by 12% to 50%. Complex MCHS structures usually increase liquid-cooling pump power, and optimised configurations suit only specific conditions with limited transferability.

Flow boiling stability and nanofluid stability are unresolved, and hot-spot prediction still lacks accuracy under dynamic loads. The authors propose 'multi-scale synergy' hybrid cooling that couples micro-scale liquid cooling with large-scale air management under real-time monitoring and control, expected to suppress local hot spots and heighten temperature uniformity.

Limits: as a review it has no new measurements and the cited gains were reported under different conditions; multi-scale synergy is a proposed direction, not a proven design.

Key points

1. Chip heat flux is described as reaching ultra-high levels of 10^6-10^7 W/m².
2. The cooling system's share of data-centre power consumption is about 38%, second only to IT equipment.
3. A cited double-layer cooling plate design with optimised headers and cross-flow reduced fluid maldistribution and lowered thermal resistance by 11.8%.
4. A cited graphene nanofluid in a heat pipe increased thermal performance by 49.4%, though nanofluid stability limits practical use.
5. A cited deep reinforcement learning controller for zonal thermal parameters reduced fan power by 12% to 50%.
6. Complex MCHS structures usually increase liquid-cooling pump power, and optimised configurations suit only specific conditions with limited transferability.

Using it in design

• For high-heat-flux chips, weigh the pump-power increase of complex microchannel cold-plate geometry against the thermal-resistance gain (for example 11.8%).
• At room scale, suppress hot spots through air-distribution optimisation and verify the accuracy of prediction-based control under dynamic load.
• Consider the hybrid strategy (multi-scale synergy) coupling chip-scale liquid cooling with room-scale air management as an option.

Who should read it

Cooling designers and researchers combining chip-level microchannel cooling and room-level air distribution.

Document contents

1. 1. Introduction
2. 2. Non-Uniform Load (MCHS: sidewall, embedded pin-fins, flow configurations; flow distribution optimization; flow boiling and novel cooling medium: flow boiling characteristics, nanofluid)
3. 3. Local Hot Spots in Indoor Environment for Large-Scale (3.1 Air Management for Solving Local Hot Spots, 3.3 Prediction Control for Solving Local Hot Spots)
4. 4. Discussion and Outlook (4.1 Issues and Solutions for Microchannel Phase Change Liquid Cooling, 4.2 Issues and Solutions for Novel Cooling Medium, 4.3 Issues and Solutions for Multi-Scale Thermal Management)
5. 5. Conclusions

Read the document

Publication details

Authors

Yifan Li; Congzhe Zhu; Xiuming Li; Bin Yang

Publisher

Energies (MDPI AG)

Year

2025

Volume

18

Issue

1

Pages

149

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.3390/en18010149

Licence

CC BY 4.0 — share with attribution

Licence evidence

www.mdpi.com

Citation