Overview
This research investigates a fully digital design-to-manufacturing workflow for TPMS heat sinks aimed at single-phase immersion cooling of high-power data center processors. The work focuses on gyroid TPMS geometry, CuCrZr material selection, LPBF manufacturability, SP5 socket constraints, and CFD-based thermal evaluation.
Problem
AI, cloud computing, and HPC workloads are pushing processor and rack-level heat loads beyond what conventional air-cooled heat sinks can efficiently handle. Single-phase immersion cooling offers a promising alternative, but most heat sinks are still designed for high-velocity air flow rather than low-velocity or pump-assisted dielectric fluid environments. This creates a need for heat sink geometries that promote 3D mixing, thermal uniformity, and manufacturability.
Approach
The study developed a digital workflow for a finless TPMS gyroid heat sink. The work included requirements definition, SP5 socket envelope constraints, lattice selection, gyroid parameterization, CuCrZr material evaluation, LPBF platform comparison, build-preparation considerations, and CFD simulation under representative immersion cooling conditions.
Workflow / Method
- Define processor/package constraints using SP5 socket envelope.
- Select candidate lattice structures: gyroid, diamond, Kelvin, and honeycomb.
- Generate TPMS geometry using nTopology implicit modeling.
- Evaluate material options and select CuCrZr for thermal performance and LPBF processability.
- Review Cu-capable LPBF platforms and build feasibility.
- Prepare digital manufacturing workflow using tools such as Netfabb/EOSPRINT.
- Run CFD/conjugate heat-transfer evaluation using immersion coolant assumptions.
- Compare flow behavior, thermal uniformity, maximum temperature, and manufacturability.
Design and Simulation Details
- Target application: single-phase immersion-cooled data center processors
- Socket constraint: AMD EPYC/SP5-style envelope
- Heat-load case: 2 kW digital simulation case
- Coolant: Novec-class representative dielectric fluid
- Inlet temperature: 40 °C
- Inlet velocity: 0.5 m/s
- Candidate lattices: gyroid, diamond, Kelvin, honeycomb
- Material: CuCrZr
- Manufacturing method: Laser Powder Bed Fusion
- Key geometry concept: TPMS gyroid with smooth, continuous flow paths
Engineering Value
The research shows that heat sink performance in immersion environments depends not only on surface area but also on flow topology, mixing, and thermal boundary-layer disruption. The gyroid structure is valuable because it supports smoother 3D flow paths, better mixing, and more uniform thermal behavior than more channelized structures.
Key Outcomes
- Compared gyroid, diamond, Kelvin, and honeycomb lattice structures.
- Identified gyroid as the most balanced option for thermal behavior and additive manufacturability.
- Showed that honeycomb can have high theoretical surface area but weaker effective cooling due to channelized flow and poorer mixing.
- In the 2 kW digital case, the gyroid heat sink reached approximately 68.1 °C maximum solid temperature.
- Estimated convection coefficient was approximately 946 W/m²K.
- CuCrZr was selected as a suitable material because it offers strong thermal conductivity, mechanical strength, and better LPBF processability than pure copper.
- The study created a digital-first workflow for virtual validation before physical prototyping.
Limitations / Future Scope
The work is digital-only and does not include physical prototype testing. Manufacturing effects such as surface roughness, internal porosity, dimensional variation, fluid aging, contact resistance, and long-term fouling were not fully represented. Future validation should include physical LPBF samples, experimental immersion testing, and rack-level thermal modeling.
- Physical prototype fabrication
- Experimental immersion-cooling validation
- CT inspection for printed TPMS geometry
- Surface roughness and porosity modeling
- Contact resistance inclusion
- Multi-heat-sink rack-level simulation
- Comparison with two-phase cooling models