# Research paper: Enhancing Heatsink Design for Single-Phase Immersion Cooling Using Deep Neural Networks and Data-Driven Engineering — retained record

> Retained semantic-scholar record from semantic-scholar.

Canonical: https://highsignal.app/data/records/0101b6b3be366a24

[← sources](https://highsignal.app/data)


source: [semantic-scholar](https://highsignal.app/data/semantic-scholar) published 2025-10-28 00:00 ingested 2026-07-21 08:13

[original source ↗](https://www.semanticscholar.org/paper/d97505210f6ae6901de7975a695d79e374a7ace0)

## retained text



```
Query: ai datacenter power cooling efficiency
Authors: Braxton J. Smith, S. Pundla, D. Agonafer
Citations: 0
The increased demand for computational power in modern computing components, coupled with the advancements in packaging technologies used to achieve this improvement, has driven a continuous and exponential rise in the thermal design power (TDP) across CPUs, GPUs, and AI accelerators. The heat fluxes produced by many of these components has surpassed the limits of air-cooling methods. To address these challenges, a movement from air-cooling to single-phase immersion cooling has emerged as alternative. Single-phase immersion cooling offers far superior thermal performance at the component level, and improvements in energy efficiency and size for the whole datacenter. To fully leverage the benefits offered by immersion cooling, server components must be designed and optimized for the immersion environment with a primary focus on the processor heatsinks. Traditionally, the design and optimization process relies on computational fluid dynamics (CFD), which is highly time intensive. This study investigates the application of deep neural networks (DNN) in developing surrogate models for heatsink performance in immersion cooling, allowing for far faster result turnaround compared to CFD. The DNN model accounts for variations in processor power, and heatsink geometry, as well as the immersion fluid temperature and flowrate for both EC-110 and PAO-6. Based on these inputs, the DNN predicts various thermal-hydraulic performance metrics such as the pressure drop, maximum case temperature, and thermal resistance.
```



document fetched 2026-07-23 08:17 · canonical https://semanticscholar.org/paper/d97505210f6ae6901de7975a695d79e374a7ace0
