SkillsProto machines heat sink prototypes to ±0.005 mm on the interface face in 3-7 days at MOQ 1, but the size has to be decided before anything is cut. Sizing a heat sink means subtracting every other thermal resistance from the junction budget, then converting the residue into fin area at the airflow you actually have, then checking that the fin geometry is one a process can produce. Most oversizing happens at the second step, where a spreadsheet assumes airflow the enclosure will never deliver.
The one equation that sizes a heat sink
Everything else is bookkeeping around this:
R_total = (T_junction_max − T_ambient) / P
R_total is the whole path from die to air, in degrees Celsius per watt. If a device allows a junction of 100 °C in a 40 °C enclosure and dissipates 25 W, the entire path must be below (100 − 40) / 25 = 2.4 °C/W. That number is not the heat sink's rating. It is the budget that every element in the stack has to fit inside.
| Element in the path | Typical share of the budget | What sets it |
|---|---|---|
| Junction to case | From the device datasheet | Package, die attach, internal spreader |
| Case to heat sink (TIM) | 0.1-0.5 °C/W for a thin, well-applied interface | Interface material, thickness, flatness, mounting pressure |
| Spreading resistance in the base | Often 15-40% of the sink's own resistance | Source footprint relative to base footprint, base thickness |
| Fin field to air (convection) | The largest single term | Fin area, airflow, fin spacing, fin efficiency |
The practical rule: take the datasheet junction-to-case figure, subtract it, subtract the interface, and what remains is the heat sink's allowance — including spreading. Engineers who skip spreading routinely specify a sink that meets the number on paper and misses it on the bench, because spreading resistance is inside the sink, not beside it.
Step 1: budget the resistance before choosing any geometry
Write the stack down in order and keep a margin. A margin of 20-30% on the sink term is normal for a first prototype, because the airflow assumption is the least reliable input in the whole calculation.
- Junction to case: from the datasheet at the actual power dissipation, not at nominal.
- Case to sink: 0.1-0.5 °C/W depending on TIM thickness. A 0.1 mm bond line of a typical thermal compound sits near the low end; a thick pad sits near the high end.
- Spreading: not a constant; it depends on the ratio of source area to base area and on base thickness.
- Convection: what is left.
If the residual after the first two terms is small — say under 1 °C/W on a 25 W part — the design problem is not fin area at all. It is getting heat out of a small source into a large base, and a thicker base or a pedestal under the die does more than another twenty fins.
Step 2: check spreading before adding fins
Spreading resistance appears when heat enters a base through a small footprint and has to travel sideways to reach the fin field. Two variables control it: base thickness and the ratio between source and base area.
A 10 x 10 mm source on a 60 x 60 mm base with a 6 mm base thickness is a spreading-limited design. The same sink with a 12 mm base, or with a raised pedestal machined under the source, carries the heat into the fin field with a materially lower penalty. This is the cheapest change available, because extra base thickness costs material and a little roughing time, while extra fins cost finishing time on every part.
Two practical tests:
- Footprint ratio. If the source covers less than about a quarter of the base area, spreading is a first-order term and should be calculated, not assumed.
- Base thickness against spread distance. As a rule of thumb, a base thickness of roughly a tenth of the spread distance keeps spreading resistance from dominating. That usually means 8-12 mm on a 60-100 mm base, not the 4-5 mm that looks sufficient in CAD.
Step 3: convert the budget into fin area at real airflow
Convection resistance is the term the whole design is built around: R_conv ≈ 1 / (h × A_eff), where h is the convective coefficient and A_eff is the effective wetted area after fin efficiency is applied.
The trap is h. Natural convection in a closed enclosure sits in the low single digits, roughly 3-10 W/m²K. Forced air at 2-3 m/s reaches roughly 25-50 W/m²K, and higher velocities push further. A design calculated at 50 W/m²K and installed in still air is wrong by roughly a factor of five, and no amount of fin area recovers that error.
| Cooling situation | Order of magnitude for h | What it means for fin spacing |
|---|---|---|
| Sealed enclosure, no fan | 3-10 W/m²K | Wide spacing, 6-10 mm, to keep buoyant flow alive |
| Open frame, natural convection | 5-12 W/m²K | Wide spacing, taller fins help |
| Fan-assisted, 1-2 m/s | 20-40 W/m²K | 2-4 mm spacing, denser fields pay |
| Ducted, 3-5 m/s | 40-80 W/m²K | Dense, thin fins; pressure drop becomes the limit |
Fin efficiency then reduces the geometric area to an effective area. A tall, thin aluminium fin is not isothermal: the tip runs cooler than the root, and the efficiency of a 25 mm tall, 1 mm thick aluminium fin in natural convection is well below 1. Treating geometric area as effective area is the second most common sizing error.
Step 4: check the geometry against the process
A design that cannot be made at the quantity you need is not a design. Machining, extrusion and casting each impose different fin limits, and the limit usually binds on aspect ratio — fin height divided by gap width, because the cutter has to reach the bottom of a deep, narrow slot.
| Route | Fin geometry it supports | Tooling | Sensible quantity | Lead time |
|---|---|---|---|---|
| CNC from plate | Pin fins, straight fins, skived-style thin fins, variable density, integral pedestals | None | 1-500 | 3-7 days |
| Machined from stock extrusion | Whatever the profile offers, plus cross-cuts | None if the profile is standard | 1-500 | 3-7 days |
| New extrusion die | Fixed profile, uniform fins | Die | Thousands | Die lead time, then days per part |
| Aluminium tooling for molded polymer sinks | Molded fin fields, limited by draft and flow | Aluminium mold | 100-10,000 | 7-10 days to T1 |
| Die cast metal sink | Drafted fins, thicker sections | Steel mold | 500+ | Mold in 15-25 days |
The engineering point: for a prototype, machining is the only route that supports a fin field designed purely for thermal reasons, including non-uniform density and a machined pedestal under the source. Deep narrow gaps raise cost quickly because the cutter has to be long and thin, but 3-7 days with no tooling is still the fastest way to test whether the calculation was right.
Step 5: verify on a bench, not in a spreadsheet
Size the part, cut it, and measure it. Two measurements settle most of the argument: the source temperature at steady state at the design power, and the air temperature at the sink inlet rather than at the other end of the enclosure. The difference between those two numbers, divided by the power, gives the real sink-plus-interface resistance.
Almost every miss traces to one of three inputs: airflow lower than assumed, spreading not accounted for, or the TIM bond line thicker than intended because the interface was not flat. Flatness under the source footprint is worth calling out explicitly, which is why the machining spec on a heat sink normally names the interface face rather than the whole base.
A worked example
A 20 W device with a 95 °C junction limit in a 35 °C enclosure, natural convection inside a sealed box.
| Step | Value | Note |
|---|---|---|
| Total budget | (95 − 35) / 20 = 3.0 °C/W | Whole path, die to air |
| Junction to case | 0.6 °C/W | From the datasheet at 20 W |
| Case to sink, TIM | 0.3 °C/W | Thin compound, flat interface |
| Remaining for the sink | 2.1 °C/W | Includes spreading |
| Spreading allowance | 0.5 °C/W | 12 mm base, pedestal under source |
| Convection target | 1.6 °C/W | At h ≈ 8 W/m²K, natural convection |
| Implied effective area | Roughly 0.08 m² | 1 / (8 × 1.6) |
| Practical interpretation | A 100 x 100 mm sink with 20-25 mm fins at 8 mm pitch, 10 mm base | Wide spacing, tall fins, machined |
The number that decides the design is the 1.6 °C/W convection target, and it is set by h = 8, not by any property of aluminium. Change the assumption to a fan at 2 m/s and the same budget is met by a sink less than half the size. That is why the first question in heat sink sizing is not "how big" but "what airflow is guaranteed".
Getting a heat sink sized and cut
Bring the heat load, the junction limit, the ambient temperature inside the enclosure and the airflow you can guarantee to /quote/. SkillsProto machines prototype heat sinks in 6061, 7075, 6082 and copper to ±0.005 mm on critical features, returns DFM notes and a price within 24 hours, holds MOQ 1 across seven processes, ships in 3-7 working days with a 48-hour expedite where the geometry allows, and delivers DDP in 3-5 days. If the volume has reached the point where an extrusion or a die casting is the better route, the quote will say so.
Frequently Asked Questions
What thermal resistance do I need from a heat sink?
Subtract everything that is not the heat sink from the total budget. Take (junction limit − ambient) divided by power, then subtract junction-to-case from the datasheet and the case-to-sink interface term. What remains is the sink's allowance, and it has to cover spreading resistance inside the base as well as convection from the fins.
Why does my heat sink perform worse than the calculation?
Usually because the convective coefficient was assumed for forced air and the part sits in still air inside an enclosure, where h is closer to 3-10 W/m²K than 50. The second cause is spreading resistance: a small source on a thin base cannot push heat sideways into the fin field, so a thicker base or a machined pedestal does more than additional fins.
How thick should the base of a heat sink be?
As a working rule, about a tenth of the spread distance, which typically lands at 8-12 mm on a 60-100 mm base. The exact value depends on how small the source is relative to the base. If the source covers less than a quarter of the base area, spreading is a first-order term and a 4-5 mm base will underperform regardless of fin area.
Can a CNC machined heat sink have pin fins?
Yes. Pin fin and cross-cut fields are machined from a solid block with no tooling, which is why they are practical at prototype quantity. The limit is aspect ratio: a deep, narrow gap forces a long thin cutter and slow feeds, so the cost rises quickly with fin height divided by gap width. Machining suits 1-500 pieces; casting and extrusion take over above that.
Does a denser fin field always cool better?
No. Fin spacing has to match the airflow. In natural convection, gaps below roughly 4 mm choke the buoyant boundary layer and performance falls as fins are added. Dense 2-4 mm spacing pays only with fan-driven air at 1-2 m/s or more, and in ducted designs the limit becomes pressure drop rather than fin area.
What finish matters on a heat sink interface?
Flatness and roughness under the source footprint matter, because they set the thermal interface bond line thickness. A machined face at Ra 1.6 micrometres or better with a controlled flatness callout over the contact area is what the thermal path needs. Cosmetic finishing elsewhere on the sink does not change performance and should not be specified as if it did.
Related capabilities
Related guides
Have parts to quote?
Upload your files — DFM-checked quotation within 24 hours.
