Custom Liquid Cold Plates for High-Density Server Cooling
As High-Performance Computing (HPC), artificial intelligence, and edge networks continue to evolve, the Thermal Design Power (TDP) of processing chips is escalating rapidly. High-density silicon—such as ASIC accelerators, GPUs, and edge AI processors—routinely generate massive heat fluxes that exceed the physical limitations of traditional forced-air cooling.
For ASIC accelerators, edge servers, edge AI devices and high-density computing modules, thermal design is no longer only an airflow problem. Compact layouts, higher chip power and limited server space make it difficult to rely only on air cooling. A custom server cold plate helps transfer heat from the chip package to the circulating coolant while fitting the mechanical layout of the server board.
Ecotherm manufactures custom liquid cold plates and server cold plates based on customer drawings. We support CNC machined cold plates, vacuum brazed cold plates, friction stir welded cold plates, gun-drilled cold plates and tube embedded cold plates for high-power electronics and server thermal management projects.
Know more：Custom liquid cold plate
*We do not manufacture complete servers, ASIC chips, AI accelerator cards, CDUs, pumps, rack manifolds, cabinet piping systems or full data center liquid cooling systems. We also do not provide immersion cooling or microchannel products. Our focus is the custom thermal component: the cold plate, heat transfer structure and manufacturable metal part.
Ecotherm’s core manufacturing capabilities focus on drastically reducing the external Rcs and Rsa to provide quantifiable cooling improvements.
Why Server Cold Plates Are Used in ASIC Accelerators and Edge Servers
A custom server cold plate operates by circulating a liquid coolant (such as a water-glycol mixture) through internal flow channels directly above the heat-generating components. Liquid has a specific heat capacity orders of magnitude higher than air, allowing it to absorb and transport extreme thermal loads out of the server chassis.
Transitioning to a liquid cold plate architecture provides critical advantages for high-density computing:
Supports Chip-Level Cooling in Compact Layouts:
In an edge AI device or 1U/2U rack server, vertical space is highly restricted. A low-profile cold plate fits within tight Z-height boundaries while delivering superior heat transfer.
Targeted Heat Spreading:
ASIC arrays often feature localized hot spots. Custom internal flow channels can be routed specifically over these high-heat zones to ensure uniform temperature distribution.
Acoustic and Environmental Reliability:
Removing high-speed server fans minimizes acoustic noise and reduces the ingestion of airborne dust, which is particularly beneficial for an industrial edge gateway deployed in harsh environments.
Heat Sources in Servers, ASIC Accelerators and Edge AI Devices
A cold plate should be designed around the actual heat source layout. The same total wattage can create very different thermal results depending on chip size, contact area, TIM thickness, coolant path and mounting pressure.
| Heat Source / Device | Thermal Challenge | Server Cold Plate Design Focus |
|---|---|---|
| ASIC accelerator | High heat flux, compact chip footprint | Contact flatness, internal flow path, low thermal resistance |
| Edge server | Limited space and airflow path | Low-profile cold plate, fitting direction, compact coolant routing |
| Edge AI device | High workload in small enclosure | Local hot spot control, mechanical integration, reliability |
| AI inference module | Concentrated chip power and dense PCB layout | Mounting pressure, TIM selection, coolant inlet/outlet position |
| GPU / CPU module | Large heat load and direct-to-chip cooling requirement | Heat spreading, balanced flow, pressure drop control |
| VRM / power module | Local hot spots near board edge | Small auxiliary cold plate or integrated heat spreading area |
| Industrial edge gateway | Harsh environment and long operating time | Sealing, corrosion resistance, vibration and mounting stability |
A practical server cold plate design must balance thermal resistance and pressure drop. A denser flow channel may improve heat transfer, but it can also increase flow resistance and pump demand. The design should be reviewed as a thermal and mechanical component, not only as a metal plate.
Server Cold Plate Design Factors
Developing a reliable edge server cooling solution requires balancing thermodynamic performance with mechanical safety.
| Design Factor | Impact on Server Performance | Ecotherm Manufacturing Approach |
| Flatness Control | Microscopic air gaps between the chip package and the cold plate cause severe interfacial thermal resistance. | We utilize precision CNC face-milling to achieve a base flatness of $\le 0.03\text{ mm}$, optimizing TIM layer thickness. |
| Pressure Drop (Delta P) | High internal fluid resistance overworks the system pump and reduces overall coolant flow. | Internal channel layouts are CNC-machined and simulated to balance the heat transfer coefficient with strict pressure drop limits. |
| Leakage Prevention | Coolant leaks near high-value ASIC chips cause catastrophic server failure. | We employ advanced metallurgical bonding (FSW/Vacuum Brazing) and conduct 100% helium mass spectrometer leak testing. |
| Coolant Inlet/Outlet Design | In compact edge servers, tubing interference can disrupt the chassis layout. | We design custom inlet and outlet port orientations (top, side, or angled) to integrate seamlessly with your internal routing. |
Custom Manufacturing Capabilities
Ecotherm manufactures custom thermal components for server electronics, power electronics and high-density computing hardware. For server cold plate projects, we can support:
- Custom liquid cold plates
- Custom server cold plates
- CNC machined cold plates
- Vacuum brazed cold plates
- Friction stir welded cold plates
- Gun-drilled cold plates
- Tube embedded cold plates
- Copper and aluminum cold plate structures
- Heat pipe modules and vapor chamber heat sinks when the design requires additional heat spreading
Frequently Asked Questions (FAQ)
Why is a server cold plate better than a heat pipe module for ASIC accelerators?
While a heat pipe module is excellent for moving moderate heat to an air-cooled fin stack, it eventually relies on ambient air. High-power ASIC accelerators generate heat fluxes that exceed the capacity of local air. A direct-to-chip cold plate removes the heat from the server chassis entirely via the liquid loop, allowing for much higher compute density.
How do you prevent leaks in custom liquid cold plates?
Leakage prevention is our highest priority. Rather than relying solely on O-rings or gaskets, we utilize metallurgical bonding techniques like Friction Stir Welding (FSW) and flux-free vacuum brazing. Every cold plate undergoes stringent hydrostatic pressure testing and helium mass spectrometer leak detection before shipping.
Can you design a cold plate for a compact edge AI device with limited tubing space?
Yes. For highly constrained edge environments, we customize the internal flow path and position the coolant inlet and outlet fittings exactly where your chassis layout allows. We can also integrate low-profile, right-angle fluid connectors to save vertical space.
Is a server cold plate the same as a complete liquid cooling system?
No. A server cold plate is one thermal component inside a liquid cooling system. It does not include the CDU, pump, manifold, rack piping or external heat rejection system.
Can Ecotherm customize server cold plates based on drawings?
Yes. Ecotherm manufactures custom server cold plates based on 2D drawings, 3D files, heat load, coolant requirements, pressure drop targets and mounting interface details.
Does Ecotherm provide immersion cooling?
No. We do not provide immersion cooling tanks, immersion systems, CDUs, pumps or rack-level liquid cooling systems. We focus on custom thermal components such as liquid cold plates, heat sinks, heat pipe modules and vapor chamber heat sinks.