An adhesive bond or gasketed cover can seal a liquid cold plate, but the bond line adds thermal resistance exactly where you are trying to remove heat, polymers creep and age in contact with coolant, and a bolted cover spends height and part count a dense assembly does not have. Friction stir welding closes the same joint with a continuous band of parent metal: full metal-to-metal conduction, nothing organic in the coolant loop, burst-pressure capability, and a sealed-for-life part in a single operation. Adhesives and bolts keep the advantage for dissimilar or heat-sensitive assemblies and for parts that must come apart again.
The quiet costs of polymer joints in thermal hardware
Liquid cooling is moving into servers, power electronics and EV platforms because air has run out of headroom. That makes the lid-to-body joint of every cold plate a critical item, and the traditional low-capital answers — epoxy, dispensed gaskets, bolted covers — carry costs that only show up later:
- Thermal resistance at the joint. An adhesive layer is an insulator sitting in the heat path. In a plate designed around millikelvin-per-watt budgets, the bond line is a permanent tax.
- Creep, ageing and coolant compatibility. Polymers under sustained pressure and temperature creep; glycol mixtures and dielectric fluids interact with them over years. Qualifying that behaviour takes long-duration testing, and requalifying after any coolant change.
- Height and part count. A bolted, gasketed cover needs flange lands, fastener bosses and access — millimetres and components that a thin plate envelope does not have to give.
- A leak is a field failure. In electronics, the consequence of a seal that lets go is the hardware the plate was protecting.
| Criterion | Epoxy / adhesive bond | Gasketed / bolted cover | Friction stir weld |
|---|---|---|---|
| Heat path across joint | Polymer layer (insulating) | Contact plus gasket interruptions | Continuous parent metal |
| Ageing mechanism | Creep, coolant attack, cure variability | Gasket relaxation, re-torque | None — metallurgical joint |
| Pressure capability | Limited by bond strength | Limited by clamping and gasket | Burst-test capable |
| Height / part count | Bond line plus surface prep | Flanges, bolts, gasket | No added parts or height |
| Process steps | Dispense, fixture, cure (hours) | Machine flanges, assemble, torque | One weld pass |
| Serviceability | No (destructive) | Yes — opens for service | No (sealed for life) |
| Dissimilar / heat-sensitive parts | Strong | Strong | Limited by tool access and clamping |
| Distortion risk | None | None | Low, but managed |
Pressure: what a typical automotive inverter loop actually asks for
These are typical OEM component specifications for an automotive inverter coolant loop on 50/50 water-glycol, not a StirLight guarantee. They are the numbers a bonded or welded plate is usually asked to meet:
- Normal operating: roughly 1-2.5 bar(g) — pump head plus expansion-tank pressure. The low-temperature loop that cools inverters typically sits lower than the engine loop.
- Max operating / design pressure: most OEM specs land at 2.5-3.5 bar(g), at coolant temperatures up to about 65-85 C for an LT loop (some specs go to 105 C).
- Proof / over-pressure: typically 1.5-2 times max operating, so around 3.5-5 bar(g), held 5-15 minutes with no leak and no permanent deformation. Watch lid ballooning between joints — measure deflection, not just leakage.
- Burst: typically at least 3 times max operating; common minimum requirements are in the 6-10 bar(g) range. A sound aluminium plate usually bursts far above that, so burst testing is a screen for joint quality and lid thickness margin.
A friction stir welded seam is parent metal, so the practical pressure limit is usually the plate design, not the seam. A polymer bond is limited by the adhesive. That is the comparison that matters at proof and burst, not at idle.
Thickness and architecture
StirLight’s cold-plate envelope, from the thin end up:
- Thin sheet, including cooling ribbons: micro friction stir welding down to 0.3 mm.
- Typical cold plates: 2-4 mm, about 0.3 m by 1 m, feasible on the robotic system.
- Battery-tray panels: typically around 1.5 m by 3 m.
- Very high pressure applications: 5-12 mm, feasible on the gantry system.
The one caveat welding carries is heat input. Even this process’s low, localised heat must be managed on long thin plates — which is why stationary-shoulder variants exist, and why parameters are developed on your actual geometry with distortion measured, not assumed. A related StirLight measurement, on an automotive battery-tray panel rather than a cold plate, achieved 0.5 mm flatness per square metre over 4 metres of weld with no post-weld straightening.
Where this substitution has already happened
Cold plates and battery cooling hardware are among the most established friction stir welding applications in industry. EV battery trays and cooling plates are welded in volume production across automotive supply chains because a welded seam gave leak-tightness and flatness that bonded and fastened builds could not hold at rate. Apple’s 2012 iMac replaced bonded and fastened enclosure joints with the process.
When bonding or bolting is still the right answer
- Dissimilar or heat-sensitive stacks — metal to plastic, or assemblies containing components that cannot see any welding heat.
- Parts that must open for service. A welded plate is sealed for life; if the design intent is periodic access, bolt it.
- Ultra-low-volume prototypes where any fixturing is overhead — though a feasibility weld needs no furnace and no cure cycle.
- Geometries a tool cannot reach or that cannot be clamped against the process forces.
Frequently asked questions
What pressure can a friction stir welded cold plate take?
The weld is parent metal. Against a typical automotive inverter loop (design 2.5-3.5 bar(g), proof 3.5-5 bar(g), burst commonly 6-10 bar(g)), the seam is not the limiting term; lid thickness and span between supports are.
Does welding distort the machined channels?
Heat input is low and local, and distortion is measured during parameter development on your geometry. Stationary-shoulder variants reduce heat input further.
Is there any contamination risk to the coolant loop?
No filler, flux or adhesive is introduced. The wetted joint is the same alloy as the plate.
Next step A structured feasibility trial answers the question on your geometry, with measurements rather than argument.