A consumer electronics customer needed a 1.8 mm heat sink housing that dissipated 6.5 W while holding ±0.05 mm on the LED seating plane. We replaced unfilled PA66 with 40% aluminium-oxide-filled PA66, cut sink temperature by 11 °C, and held flatness to 0.08 mm.
Key takeaways
- Filled PA66 conducts heat roughly 3x better than unfilled PA66, but it also shrinks differently in flow and cross-flow directions.
- Thermal performance is decided by wall thickness and contact pressure, not by filler percentage alone.
- A filled grade needs a hot-oil or hot-water mold at 90–110 °C; a cold-tool trial will give you false warpage data.
- Aluminium-oxide filler costs about 2.4x unfilled PA66 per kilogram, which usually pays back in a smaller heat sink footprint.
- Design the seating plane as a machined-in insert pocket, and let the mold hold the flatness — not the screw torque.
The challenge: a heat sink that had to be a structural housing
The customer built a compact network-attached storage device. The original design fastened a die-cast aluminium heat sink to a moulded PA66 cover with four screws. Two problems came back from the field.
The first was thermal. Under sustained write load the housing skin reached 71 °C at an ambient of 32 °C, and the customer’s own specification capped touch temperature at 60 °C. The second was assembly. The die-cast sink cost 4.10 USD per unit and added 96 g, and the four screw bosses in the PA66 cover cracked in transit when the torque exceeded 0.45 N·m.
The brief was to remove the aluminium part entirely and give the thermal job to the moulded housing. The constraints were tight:
- Housing wall 1.8 mm, with a 0.6 mm gap over the processor and a thermal pad bridging it.
- LED seating plane flatness within 0.08 mm across a 42 mm span, and the LED aperture centred to ±0.05 mm.
- Annual volume 60,000 units, with a tooling budget of 18,000 USD and a first-article date 50 days out.
- Housing must still pass a 1.2 m drop onto concrete and a 0.45 N·m screw torque test without boss cracking.
- No metal insert allowed in the seating area, because the customer did not want a second supplier.
What we did
Step 1: picking the filler, not the grade
We modelled three candidate resins in Moldflow before touching steel: unfilled PA66, 30% glass-fibre PA66, and PA66 with 40% aluminium oxide. Glass fibre raises strength but does almost nothing for thermal conductivity, because glass is an insulator. Aluminium oxide is a ceramic filler with a thermal conductivity around 30 W/m·K, so a 40% loading lifts the compound from roughly 0.25 W/m·K to 0.9–1.1 W/m·K.
The moulded test plaques confirmed it. A 1.8 mm plaque of unfilled PA66 dropped 24 °C across a 6.5 W load; the same plaque in 40% aluminium oxide dropped 13 °C. That was enough headroom to hit the 60 °C touch limit with margin.
We deliberately did not chase the highest loading on the market. At 50% and above, the compound turns brittle and notched impact falls sharply, which would have killed the drop test. Forty percent was the highest loading that still passed Izod on our own test bars.
Step 2: a two-cavity tool with an oil circuit
The tool is a two-cavity design in 1.2738HH pre-hardened steel, with a 1.2344 (H13) core insert hardened to 48–52 HRC for the seating-plane area. Two cavities kept the tool inside the 18,000 USD budget and still delivered 60,000 units a year at a 38-second cycle.
The critical decision was temperature control. Filled PA66 needs a mold surface at 90–110 °C to develop full crystallinity; run it cold and the part stays amorphous, then shrinks unpredictably over the following 48 hours. We ran separate circuits for core and cavity fed by an oil unit at 105 °C, with 12 mm cooling lines on a 45 mm pitch, and a dedicated 8 mm bubbler directly under each LED seating boss.
Measured core-to-cavity temperature difference after stabilisation was 2.1 °C, down from 9 °C on the first cold-tool trial we ran as a comparison.
Step 3: gate position and shrinkage compensation
Aluminium-oxide filler does not create the fibre orientation problem that glass does, but the compound still shrinks 0.4% in flow and 0.7% across flow. We placed a single fan gate on the inner face, opposite the LED aperture, so the flow front arrives at the seating plane last and the weld line lands in a non-cosmetic area.
Moldflow predicted 0.34 mm of differential shrinkage across the 42 mm seating span. We compensated by adding 0.12 mm of steel on the cross-flow side of the cavity before the first trial, rather than cutting to nominal and chasing the error afterwards.
Step 4: trial, measure, adjust
Trial 1 ran at 105 °C mold temperature, 285 °C melt, 55 MPa hold pressure and 8 seconds of hold time. The parts looked right and measured wrong: seating-plane flatness came in at 0.19 mm and the aperture sat 0.07 mm off centre. We measured 30 parts from each cavity on the CMM and mapped the deviation.
The error was not random. Deviation grew linearly from the gate end to the far end, which told us the cavity was packing unevenly rather than shrinking unevenly. We raised hold pressure to 70 MPa and extended hold time to 11 seconds, and the flatness fell to 0.11 mm. A second steel adjustment of 0.05 mm on the far-end cavity face brought it to 0.08 mm, inside the 0.08 mm limit with the CMM uncertainty included.
Step 5: fixing the cracked bosses
Filled PA66 has lower elongation at break than unfilled, so the original 2.4 mm boss wall cracked at 0.45 N·m regardless of the resin. We increased boss outside diameter from 5.0 mm to 6.2 mm and added a 0.6 mm fillet at the root, then ran 20 torque cycles per cavity. No cracking.
The drop test drove one more change. At the original 1.8 mm uniform wall the housing passed, but the corner ribs sheared. We thickened the four corner ribs to 2.4 mm with a 1.2% draft and kept the outer wall at 1.8 mm, so the housing now passes 1.2 m drop with the PCB installed.
The result
The programme shipped in 47 days from mould design freeze, three days inside the 50-day target. The table below compares the original aluminium-sink assembly with the filled-PA66 housing as measured on production parts.
| Parameter | Original design (die-cast Al sink) | Filled PA66 housing |
|---|---|---|
| Housing skin temperature at 6.5 W, 32 °C ambient | 71 °C | 60 °C |
| Temperature rise across the wall | 24 °C | 13 °C |
| Seating-plane flatness (42 mm span) | 0.06 mm (machined) | 0.08 mm (as moulded) |
| LED aperture position, cavity to cavity | n/a | ±0.04 mm |
| Part weight | 96 g housing + 58 g sink | 71 g total |
| Part cost per unit | 4.10 USD sink + 2.35 USD housing | 3.05 USD |
| Assembly steps | 9 (4 screws, pad, sink) | 5 |
| Cycle time | n/a | 38 s |
| First-pass yield at T1 | n/a | 94.1% |
| First-pass yield at SOP | n/a | 99.2% |
The cost figure is the one the customer cared about most. Removing the die-cast aluminium sink and two of the four assembly operations cut 3.40 USD per unit, which against 60,000 units a year returned the 18,000 USD tool in under ten weeks of production.
We do not mould in-house — mold making in-house, molding through vetted partners — and that separation is what made the 47-day schedule possible. Because we do not compete with our customers for moulding capacity, the tool was prioritised in the partner’s press schedule, and the same team that cut the steel ran the two steel adjustments without a handover.
Our answer to the question we get asked most
Buyers usually ask whether the filled grade is worth the price. Our answer: at 2.4x the resin cost, a filled grade is worth it only when it removes a metal part or lets you keep the wall thin. On this housing the compound added 1.55 USD of resin per unit and removed 4.10 USD of die-cast aluminium, so the payback was immediate. If you are filling a 3 mm wall purely for thermal reasons and no metal part is coming out of the design, unfilled resin plus a small stamped sink is usually cheaper.
When filled PA66 is the wrong choice
Three situations sent this customer’s sister products back to unfilled resin or to metal, and they are worth naming because the resin choice is not automatically the answer.
- Very thin walls below 1.0 mm. Aluminium-oxide loading pushes melt viscosity up, and a 0.8 mm wall over a 90 mm flow length will short-shot before it fills.
- Cosmetic Class A surfaces. The filler shows as a faint mottle on high-gloss faces; a textured or matte finish hides it.
- Loads above roughly 15 W. At some point the wall has to grow to 3 mm or more, and a small stamped or die-cast sink becomes lighter and cheaper again.
What carries to your project
- Pick the filler for the physics, not the datasheet headline. Glass fibre buys stiffness; ceramic filler buys thermal conductivity. A 40% glass-filled PA66 would have failed this thermal brief while looking stronger on paper.
- Run the thermal trial at production mold temperature. A cold tool gives amorphous parts with the wrong shrinkage, and any flatness reading taken from them is worthless.
- Compensate steel before the first trial. Adding 0.12 mm where the simulation predicts shrinkage is cheaper than cutting twice, and it keeps the trial focused on packing rather than on geometry.
- Read the deviation map, not the single number. A flatness value alone told us nothing; the linear gradient from gate to far end told us the problem was hold pressure.
- Widen the boss, not the screw. Filled resin has less elongation, so the fix for a cracked boss is more wall and a root fillet, never a higher torque spec.
Related guides in this series
- Injection Molding Materials Guide — how to read filler loadings and what each one costs you.
- PA66 vs POM: Material Comparison — where PA66 wins and where it does not.
- Case Study: Warpage Control on Glass-Filled PA66 Brackets — the fibre-orientation problem this part avoided.
- Injection Molding Cooling System Design — circuit layout and the bubbler that held the seating boss.
- Injection Molding Defects and Quality Guide — how flatness and short-shot failures are traced.
- Injection Molding Tolerances Guide — what ±0.05 mm actually costs on a moulded part.
- Hot Runner vs Cold Runner — why this two-cavity tool stayed on a cold runner.
Send us your thermal brief
If you are trying to move a heat sink into the moulded housing, the useful starting points are the wattage, the ambient, the touch-temperature limit and the available wall thickness. Send those four numbers with your 3D model and we will run the material comparison and return a written DFM report within 3 working days, at no cost.
Send your thermal brief to RCH Plastic — our answer will name the filler loading we would use and the mold temperature it needs, before you commit to a tool.
