Pump-out is the slow, cyclic migration of thermal grease out of the interface until the centre of the joint is starved and junction temperature starts climbing. It is the most common reason a design that passed qualification begins failing in the field two years later.
What actually happens
The mechanism is straightforward once you see it as a mechanical process rather than a chemical one.
The device heats up
The die, the substrate and the heat sink all expand — but not by the same amount. Silicon is around 2.6 ppm/K, copper 17 ppm/K, aluminium 23 ppm/K.
The gap changes width
Under this CTE mismatch the joint does not just get thicker or thinner uniformly — it flexes, and the gap width changes across the die area.
Pressure squeezes the grease sideways
When the gap narrows, the viscous material has to go somewhere. It flows outward, like stepping on a wet sponge.
The gap reopens with less material
When it cools, the grease does not fully flow back. Air takes its place.
The cycle repeats
Each cycle moves a little more material outward. After hundreds to thousands of cycles, voids form at the centre — exactly where the heat is.
Which applications are most at risk
| Risk factor | Why it accelerates pump-out |
|---|---|
| Large CTE mismatch | Greater gap-width swing per cycle — e.g. silicon die on aluminium heat sink |
| Wide temperature swing | ΔT of 80 K pumps far faster than ΔT of 20 K |
| Frequent power cycling | EV inverters, traction drives, grid converters cycle constantly in service |
| Large die or substrate area | Longer migration distance, larger absolute volume change |
| Low-viscosity grease | Flows more readily under squeeze films |
| High mounting pressure | More squeeze force driving lateral flow |
| Sealed, unserviceable design | No opportunity to repaste — failure is permanent |
The classic high-risk combination is an automotive traction inverter: silicon carbide dies on a substrate, an aluminium cold plate, ΔT over 100 K, and power cycling on every trip. That is precisely where grease-based interfaces have the shortest life.
How the failure shows up
- Gradual then sudden. Resistance often stays flat for a long time, then climbs steeply once voids connect across the die centre.
- Worst at the hottest spot. Localised hot spots appear rather than a uniform temperature rise.
- Visible on teardown. You see a ring of displaced grease around the die and a dry, grey, cracked area in the middle.
- Correlates with cycling, not hours. Two units with the same operating hours can differ dramatically if one cycled more.
How to prevent it
Option 1 — Use a solid interface material
This is the direct answer. A solid pad has no carrier fluid and cannot flow. Pump-out is not a degradation mode for it, because there is nothing to pump. This is the primary reason industrial and automotive designs have moved away from grease.
Lenecold's graphene pad is a solid, die-cut part: it holds its position under cycling, and has been validated at 150 °C for 1,500 hours with thermal resistance drift under 10%.
Option 2 — Reduce the CTE mismatch
Use an intermediate spreader or baseplate material closer to silicon — copper-tungsten, AlSiC, or a copper-molybdenum laminate. This reduces the driving mechanism directly, but adds cost and a second interface.
Option 3 — Manage the mechanical design
- Mount the heat sink at the thermal centre so expansion is symmetric
- Use compliant mounting hardware that maintains pressure without over-constraint
- Avoid stiff, over-torqued corners that force the substrate to bow
Option 4 — Accept it and design for service
Valid for consumer electronics: specify a repaste interval. Not valid for sealed industrial, automotive or aerospace hardware.
How to test for it
Standard industry practice is power or thermal cycling with periodic resistance measurement:
- Deep thermal cycling: e.g. −40 °C to +125 °C, several hundred to a few thousand cycles, measuring thermal resistance at intervals.
- Power cycling: closer to real service — heat the device to full load, cool to ambient, repeat, recording ΔT each cycle.
- High-temperature storage: 125–150 °C soak, measuring resistance drift over 1,000+ hours. This catches dry-out and oil separation rather than pump-out.