Most cold plate channel designs fail at the quoting stage for one of four reasons: the depth-to-width ratio exceeds what a cutter can hold, the internal corner radius is smaller than any tool that fits the channel, the ribs between channels are too thin to survive cutting forces, or the sealing method was chosen after the channel layout instead of before it.
None of these show up in a thermal simulation. All of them show up as a re-quote, a request for drawing changes, or a part that arrives out of flatness.
This page covers the manufacturing constraints that shape a machined liquid cold plate channel — the numbers a supplier will push back on, and why.
Quick reference: machined channel limits
These are typical values for CNC-machined aluminum (6061/6063) and copper (C11000/C10200) cold plates. If your design is specifically aluminum, our aluminum cold plate machining page covers the joining, post-machining and DFM issues that follow after the channel geometry is set. Treat them as indicative starting points, not specifications — every shop’s tooling and machine rigidity differ.
| Constraint | Comfortable | Adds cost | Usually refused |
|---|---|---|---|
| Channel depth : width | up to 3 : 1 | 3 : 1 to 5 : 1 | above 6 : 1 |
| Internal corner radius (plan view) | ≥ 0.5 × channel width | ≥ 0.35 × channel width | sharp corners |
| Rib thickness between channels | ≥ 0.5 × rib height | ≥ 0.35 × rib height | below ~0.8 mm |
| Minimum channel width | ≥ 3 mm | 1.5–3 mm | below 1 mm by milling |
| Floor thickness under channel | ≥ 2 mm | 1–2 mm | below 1 mm |
| Flatness on sealing face | 0.10 mm | 0.05 mm | below 0.02 mm |
Copper sits one column to the right of aluminum on almost every row. A geometry that is routine in 6061 may be an “adds cost” case in C11000.
1. Depth-to-width ratio is a tool stiffness problem
A milled channel is cut by an end mill whose diameter cannot exceed the channel width. The deeper the channel, the further that cutter has to stick out of the holder — and tool deflection rises with roughly the cube of stickout length.
What this looks like in practice:
- Up to 3:1 — routine. A 4 mm wide, 12 mm deep channel cut with a 4 mm end mill is unremarkable.
- 3:1 to 5:1 — the shop drops the depth of cut per pass, slows the feed, and takes a spring pass. Cycle time can double. Expect the quote to reflect it.
- Above 5:1 — needs reduced-neck or extended-reach tooling, which is more fragile and more expensive. Some shops will decline rather than risk breaking a tool inside a nearly finished plate.
- Above 8:1 — milling is the wrong process. Look at skiving, bonded fin, extrusion, or a tube-in-plate construction instead.
The failure modes when the ratio is pushed are tool deflection (the wall bows outward at the top and pinches in at the bottom, so the channel is not the width you drew), chatter (visible tool marks, poor surface, unpredictable roughness), and breakage mid-cut.
Those are aluminium failure modes. Copper is the stiffer material but yields at a much lower stress, so the wall holds shape under cutting load and then smears or rolls a burr instead — the same depth-to-width ratio behaves differently in copper.
The same arithmetic governs an air-cooled fin field: cutter diameter is capped by the gap, depth of cut is the fin height, and the ratio behaves the same way on a heatsink as it does on a channel.
Design lever: if you need more flow area, widen before you deepen. A 6 mm × 6 mm channel and a 3 mm × 12 mm channel have the same cross-section, but only the first is easy to cut.
2. Internal corner radius equals tool radius — this is not negotiable
This is the single most common DFM error on cold plate drawings, and it is worth stating plainly:
In a milled channel, the minimum internal corner radius in the plan view is exactly half the diameter of the smallest tool that cuts that corner.
A 4 mm wide channel cut with a 4 mm end mill turns a corner at R2.0 minimum. You cannot specify R1.0 in that corner. Nothing about the machine, the programming, or the operator changes this — it is the geometry of a round cutter.
There is one way around it, and it is expensive: use a smaller tool. A 2 mm end mill can produce an R1.0 corner in a 4 mm channel. But now:
- the 4 mm width needs multiple passes instead of one,
- the 2 mm tool has half the stiffness, so the depth-to-width limit from Section 1 gets much worse,
- cycle time goes up substantially.
For a serpentine channel with a dozen U-turns, that decision multiplies across every corner.
Why this matters more than it sounds. Sharp internal corners are almost always drawn out of CAD habit, not thermal necessity. In a flow channel, a generous corner radius is usually better — it reduces the separation zone at the turn, which lowers pressure drop and removes a low-velocity dead spot where the local heat transfer coefficient collapses.
So the constraint and the physics point the same way. Draw the corner at the tool radius on purpose, and state the tool diameter assumption on the drawing.
Two related radii people forget:
- Floor-to-wall radius. A flat end mill leaves a small radius where the channel wall meets the floor (typically 0.2–0.5 mm, set by the cutter’s corner condition). A truly sharp floor corner is not achievable by milling.
- Outside corners of the pocket can be sharp — that constraint only applies to internal corners.
3. Rib thickness and the walls between channels
The ribs separating adjacent channels are cantilevers being pushed sideways by the cutter. Thin, tall ribs deflect during machining, which produces channels that are on-dimension at the top and off-dimension at the bottom, and ribs that are not straight.
A workable rule of thumb for aluminum is rib thickness ≥ 0.5 × rib height for comfortable machining, dropping to roughly 0.35 × height with careful process planning. Below about 0.8 mm at any meaningful height, milling becomes unreliable regardless of ratio.
Copper is worse than aluminum here, not better. Despite being a soft metal, C11000 is gummy — it produces built-up edge on the cutter, higher cutting forces than its hardness suggests, and poor chip breaking. Thin copper ribs smear and deflect.
If your design needs fin ratios that violate this, the geometry is telling you to change process:
- Skiving produces very thin, tall parallel fins (down to a few tenths of a millimetre) in aluminum or copper — but only straight, parallel geometry with no turns.
- Bonded fin / brazed fin stock achieves high fin density with a separate fin insert.
- Extrusion gives fine parallel profiles at low unit cost, but requires a die, so the MOQ jumps.
None of these are drop-in substitutes for a serpentine milled channel. Choosing them changes the channel layout entirely, which is why process selection belongs at the start of the design, not the end.
4. Channel layout: what each one costs to make
| Layout | Thermal behaviour | Pressure drop | Manufacturability | Typical use |
|---|---|---|---|---|
| Serpentine | Good coverage, but coolant heats along the path — measurable temperature gradient inlet to outlet | Highest | Easiest to mill; every U-turn is a corner-radius decision | Single or few heat sources, moderate flow |
| Parallel straight | Uniform if flow distributes evenly; uneven distribution is the failure mode | Lowest | Easy to mill; needs a well-designed inlet/outlet header | Multiple similar heat sources in a row |
| Pin fin | High surface area, tolerant of uneven flow, good for spreading | Moderate to high | Slow to mill (many features); each pin has a minimum diameter set by tooling | High heat flux, small footprint |
| Microchannel (<1 mm) | Highest heat transfer coefficient | Very high; needs clean coolant and filtration | Not practical by milling — needs skiving, etching, bonded construction, or additive | High-power-density devices |
A practical note on serpentine vs parallel. Serpentine is easier to make and needs no flow-distribution design, which is why it dominates low-volume custom work. Its weakness is real though: the coolant leaving the last pass is hotter than the coolant entering the first, so the last device in the path runs hotter. If your heat sources have equal power and equal junction temperature limits, that gradient is your design margin being spent for free.
Parallel layouts fix the gradient but move the problem into the header: if the inlet manifold does not distribute evenly, some channels starve, and a starved channel is worse than a serpentine’s gradient.
5. Distortion, flatness, and why the sealing face is the hard part
A cold plate starts as rolled or extruded stock carrying residual stress. Machining a network of channels into one face removes material asymmetrically and releases that stress — the plate bows.
This matters because the face that has to be flat is usually the face that mates against the device, and the perimeter that has to be flat is the sealing land. A plate that measures 0.03 mm flat before the channels are cut can be several times that after.
How shops handle it, in rough order of cost:
- Rough machine → stress relieve → finish machine. Adds a thermal cycle and handling, but is the standard answer for anything with a tight flatness spec.
- Balanced material removal — machining relief on the back face so stress release is symmetric. Costs nothing extra if designed in from the start.
- Careful fixturing — the plate must not be clamped flat during machining and then spring when released. Vacuum fixturing and low-stress workholding help.
- Final grinding or fly-cutting of the mating face after all other operations.
What to specify. Give a flatness value only on the surfaces that need it — the device mounting face and the sealing land — and give the value over the relevant area, not the whole plate. A blanket “0.05 mm flatness” over a 400 × 250 mm plate is far more expensive than 0.05 mm over the 80 × 80 mm area that actually touches the module.
6. Chip evacuation in deep channels
In a deep, narrow slot, chips have nowhere to go. They get recut, which destroys surface finish, accelerates tool wear, and generates heat that feeds back into distortion.
This is why a 3 mm × 15 mm channel is not just “a 3 mm channel, but deeper” — it is a different machining problem. Shops manage it with through-spindle coolant, peck cycles, trochoidal toolpaths, and reduced step-downs, all of which add cycle time.
Design levers that help:
- Avoid closed pockets where chips cannot escape; open the channel to a face or a larger cavity where possible.
- Avoid dead-end channel terminations at full depth.
- Where a deep feature is unavoidable, tell the supplier it is a functional dimension so they plan the process rather than discovering it on the machine.
There is also a cleanliness consequence. A channel geometry that traps chips during machining traps them during deburring and washing too. Any cold plate that will run in a closed loop with a pump and quick-disconnects needs to leave the shop clean; a design that makes cleaning hard is a design that ships particles into the customer’s coolant loop.
7. The sealing method constrains the channel layout — decide it first
This is the sequencing error that causes the most rework. Channel geometry gets finalised, then the sealing method gets chosen, and the channel layout has to change anyway.
O-ring groove with a bolted cover. Needs a continuous groove around the channel network, a bolt pattern outside the groove, and enough land width to carry both. That perimeter allowance is typically 15–25 mm of plate width you cannot use for channels. In exchange, it is fully machined, needs no furnace or welding equipment, is serviceable, and has effectively no minimum order quantity.
Friction stir welding a cover plate. Aluminum in practice. Needs a weld land wide enough for the tool shoulder — usually well over 6 mm, and the shoulder must not run over a channel. The weld path also has to be reachable by the FSW machine’s travel. Gives a permanent, leak-tight joint with no gasket.
Vacuum brazing. Works for both copper and aluminum, gives excellent joint strength and thermal continuity, and allows more channel area because the sealing land can be narrow. Constrained by furnace chamber size and by the need for good flatness on both mating surfaces before brazing. Because a furnace run has a fixed cost regardless of how full it is, small quantities carry high per-part amortisation.
The practical point: the sealing method sets how much perimeter you must reserve, whether the cover can cross channels, and what your realistic minimum quantity is. Choose it before you lay out the channels — the trade-offs between vacuum brazing and friction stir welding are covered in detail separately.
8. A DFM checklist before you send the drawing
Run these before the RFQ goes out. Each one catches a re-quote. If you are also deciding which shop to send it to, that is covered separately.
- Is any channel deeper than 3× its width? If so, is that depth actually required?
- Is every internal corner radius at least half the width of the channel it sits in?
- Have you stated the tool diameter your corner radii assume?
- Is every rib at least half as thick as it is tall?
- Is the floor under the channels thick enough for the working pressure, plus margin for machining tolerance?
- Is flatness called out only where it is needed, and over the area where it is needed?
- Is the sealing method chosen, and is there enough perimeter land for it?
- Does the design allow chips and cleaning fluid to escape from every channel?
- Is the material grade specified (6061 vs 6063; C11000 vs C10200), not just “aluminum” or “copper”?
- Are the working parameters stated — coolant type, flow rate, working pressure, and the leak or burst test you require?
If a supplier quotes this drawing without asking about any of the above, that is information about the supplier.
What to take away
Cold plate channel design is not primarily a thermal problem. The thermal side is solvable — more area, more turbulence, more flow. The constraints that decide whether a design can be made, at what price, and in what quantity are geometric and process-based: how far a cutter can reach, how tight a corner a round tool can produce, how thin a rib can stand up to cutting forces, and what the sealing method demands of the perimeter.
Designs that respect those four constraints get quoted quickly and made accurately. Once the geometry is ready, use our cold plate quote checklist to make sure the drawing also includes the material, quantity, joining, port, surface and test requirements needed for pricing. Designs that ignore them come back with questions — or worse, get quoted by someone who did not notice, and arrive out of tolerance.
Have a channel design you want checked before it goes out for quote? Send the drawing and we will flag the manufacturability issues before pricing — we machine these in copper and aluminium, prototype quantities through production.