Take a sectioned liquid-cooled plate. The copper tubes are embedded in the aluminum channel; the middle layer of sealant is shiny, while a small section along the edge remains uncoated, resembling unevenly applied putty.
Most people will glance at it and move on. But when you mount this board beneath an 800 W chip, that empty space becomes the chip’s critical vulnerability.
This seam was not drawn; it was created through the process.
The design drawings for the liquid-cooled plate contain no specified gap. The outer diameter of the copper tube is clearly marked, as is the width of the aluminum slot; theoretically, subtracting one value from the other should result in a perfectly tight fit. In practice, however, this is not the case. In a tube-in-slot cooling plate configuration, the aluminum plate is milled with a groove, into which the copper tube is pressed. The tube is held in place by the elastic recovery force generated after compression; this force presses the tube wall against the groove wall. To ensure that this force remains effective, the outer diameter of the copper tube must be slightly larger than that of the aluminum slot; a common industry allowance is 0.1 mm for a tube diameter of 610 mm. If the interference is too small, the fit is not secure; if it is too large, the aluminum slot will expand and crack immediately. Within this 0.1–0.2 mm range lies the most elusive aspect of thermal resistance in the entire plate. That is not all: the copper tube becomes elliptical at bends; once elliptical, the gap becomes wider at one end and narrower at the other. After tube pressing, the aluminum undergoes slow creep, causing the positive pressure to diminish year by year. Following this comes surface treatment: the natural oxide film on the aluminum surface serves both as its protective layer and as a barrier to adhesive bonding. Consequently, the width of this gap is not a constant but rather a distribution. It cannot be specified in the drawings; it can only be observed when the assembly is sectioned open.
Whether the seam contains air or adhesive makes a difference of more than seventy times.
Let’s do the simplest calculation. Thermal resistance R equals thickness divided by the product of the thermal conductivity and the area: R = t / (k·A). Take a thickness of 0.1 mm over an area of 1 cm². If the gap contains air, whose thermal conductivity is approximately 0.026 W/(m·K), the resulting thermal resistance is 38.5 K/W. If the gap is filled with a thermal-conductivity material of 2.0 W/(m·K), the thermal resistance drops to 0.5 K/W—a difference of 77-fold. This is why those who design cold plates generally emphasize “filling.” The survival of the “filling-gel” category is also due to this very point: whether there is anything in the gap is far more important than what that material actually is. However, the following point deserves even greater attention: a thicker gel layer is not necessarily better. The industry consensus on the process window is 0.1–0.3 mm. Beyond 0.3 mm, the gel layer itself becomes a new source of thermal resistance. Some prototype components, in an attempt to fill an oversized slot, are filled with nearly 1 mm of gel, then try to remedy the situation with a material rated at 3.0 or even 4.0 W/(m·K). This combination is akin to installing a jet engine in a bicycle: the specifications look impressive, but the bike won’t go fast. If the gap width is not properly controlled, increasing the thermal conductivity will be futile.
Many people have never calculated exactly which segment of the adhesive has been improved.
The thermal resistance of a cold plate is the sum of the resistances in series. Between the chip and the cold plate there is an interface material layer; the cold plate base has a certain thickness, the copper tube has a wall thickness, and there is also convection between the tube wall and the coolant. The gap in question accounts only for the section from the “outer wall of the copper tube to the aluminum sink.” The question is: if this section accounts for only 10% of the total thermal resistance, how much can the total thermal resistance be reduced by increasing the thermal conductivity of the thermal paste from 1.5 to 3.0? The reduction would be roughly five points. When converted to junction temperature, this may result in only a one-or two-degree drop. Consequently, some users replace the paste with a more expensive formulation, find no improvement in measurements, and then suspect that the material supplier is exaggerating its claims. The supplier is also unfairly treated: their figures are accurate; they simply applied them to the wrong location. The method for determining this is not complicated: create a thermal resistance waterfall chart. Measure the total thermal resistance, subtract the values that can be calculated for each individual section, and the remainder represents the interface. Only when this section accounts for more than 30% of the total should you consider changing the paste; if it accounts for less than 15%, first examine the interface material and the flow path rather than rushing to change the material.
Four Ways Glue Can Fail in Seams
First: Air bubbles. If vacuum degassing is insufficient, or if the adhesive is applied too rapidly, air is entrained into the system. When the joint is only a few tenths of a millimeter wide, once an air bubble becomes trapped inside, it cannot escape. Upon slicing the part, one sees a row of tiny pores within the adhesive layer, resembling a sponge cake. Second: Pumping out. Under repeated thermal cycling, the adhesive layer expands and contracts; over time, this gradually forces the material out through the joint. The pumping out of interfacial materials is common knowledge; the same applies to filling adhesives, though it is less frequently discussed. Once the material is pumped out, the joint reverts to air, returning the system to that 38.5 K/W state. Third: Moisture absorption and delamination. When ethylene glycol is mixed with water and a small amount of ions, the interface first forms bubbles before separating. Aluminum surfaces are particularly demanding; if the oxide film is not thoroughly cleaned, the adhesive will simply sit on top of it. Anyone who works on aluminum components understands this, but those who work with adhesives often overlook it. Fourth: Galvanic corrosion. The standard electrode potential of aluminum is approximately-1.66 V, while that of copper is approximately +0.34 V—a difference of two volts. Once an electrolyte seeps into the joint, aluminum acts as the anode and corrodes first. For every 10°C increase in temperature, the corrosion rate roughly doubles. If the coolant lacks sufficient corrosion inhibitors, the ethylene glycol itself will decompose into glycolic acid, formic acid, oxalic acid, and similar compounds, with both copper and aluminum catalyzing this reaction. If the adhesive layer absorbs water, it effectively creates a bridge for this electrochemical cell. Among these four failure modes, only the first can be detected through first-piece inspection. The other three only become apparent after the full cycle has been completed.
Three paths, each with its own set of consequences.
If you want to resolve this gap completely, there are three approaches in the industry, but none guarantees a win-win outcome. Interference fit is the cheapest option; it relies on the elasticity of aluminum to provide support. In the short term, this works fine, but in the long term, the gamble is that the aluminum will not undergo creep. Thermal-conductivity gel filling can fill irregular gaps and buffer the small difference in thermal expansion between copper and aluminum. Copper has a linear expansion coefficient of approximately 17×10⁻⁶/℃, while aluminum’s is about 23×10⁻⁶/℃; for a one-meter-long plate with a temperature difference of 55℃, the resulting displacement is 0.33mm. This displacement must be absorbed by something, and the gel can absorb a portion of it. The trade-off is that the gel will age, pump out over time, and absorb moisture. Vacuum brazing creates a metallurgical bond between copper and aluminum, providing thermal conductivity one order of magnitude higher than that of gel. The trade-off is the temperature: aluminum-silicon brazing material must be heated to 580–620℃; at this temperature, the aluminum plate will deform and the copper tube will anneal, meaning the assembly will still require leveling after welding. Both equipment investment and yield rate incur costs. When choosing among these options, the deciding factor is not whether the technology is the most advanced, but rather your production volume, your service-life requirements, and how much you are willing to pay for a few degrees of junction temperature.
How should we conduct the acceptance inspection? Which points can be implemented?
The first batch of samples must be sectioned for metallographic examination to assess the adhesive layer thickness, void fraction, and wettability. Do not rely solely on the average thickness reported in the report; examine the thinnest region as well. The thermal resistance test rig must undergo cyclic testing. A common industry practice is to measure interfacial thermal resistance every 200 cycles over a temperature range of –40 °C to 85 °C. Publicly available engineering data indicate that an increase in interfacial thermal resistance of 50%–80% after 1,000 cycles is considered typical .
If the increase exceeds this range, it indicates that the interface has already begun to degrade; investigate the adhesive and surface treatment accordingly. Airtightness should be verified using helium leak testing; the typical requirement is that the leak rate does not exceed 1×10⁻⁹ Pa·m³/s. Among the failure modes of cold plates, leakage accounts for the majority of cases, so this parameter cannot be overlooked. Coolant pH, conductivity, and corrosion inhibitor concentration should be measured periodically. For aluminum systems, chloride ion levels must also be closely monitored. Eaton’s technical documentation states unequivocally: point corrosion in aluminum is most often initiated by chloride ions.
last
I placed the cut panel on the table as a reminder. In the liquid-cooling industry, everyone is racing to optimize flow channels, materials, pumps, and quick-connect fittings. Ultimately, what often becomes the performance bottleneck is that gap measuring less than 0.3 mm.