Cold Plate Design: Rectangular-Channel Heat Transfer and Pressure Drop
A plain-language guide to sizing a liquid cold plate: why a rectangular channel isn't a round pipe, how the Shah-London correlations set the heat-transfer coefficient and friction, how bends add pressure drop, and how it all becomes a thermal resistance and base temperature.
A cold plate is a pipe you can machine into any shape
A liquid cold plate is a metal block with a coolant channel milled into it: the heat-generating parts bolt to one face, coolant runs through the channel, and the two never touch. Design comes down to two competing numbers. The thermal resistance tells you how hot the mounting face runs for a given heat load — you want it low. The pressure droptells you how hard the pump has to push to keep the coolant moving — you want that low too. Push the channel smaller and faster to cut the thermal resistance and the pressure drop shoots up; the whole job is finding the balance.
Why a rectangular channel isn't a round pipe
Milled channels are rectangular, not round, and that changes the physics in a way that's easy to get wrong. For a round pipe in slow (laminar) flow, the friction factor is exactly f = 64/Re. For a rectangular channel it isn't 64 at all — the Poiseuille number f·Re ranges from about 57 for a square channel up to 96 for a very wide, thin one (the parallel-plate limit). The heat transfer shifts the same way: the laminar Nusselt number (which sets the heat-transfer coefficient h = Nu·k/Dh) is about 3.6 for a square channel and climbs toward 8.2 as the channel gets thin. This calculator uses the Shah & London polynomials that capture that aspect-ratio dependence exactly, keyed off the channel's width-to-height ratio, rather than borrowing the round-pipe numbers.
The bends are not free
A serpentine cold plate turns the flow around many times, and every turn costs pressure. Each 45°, 90° or 180° bend adds a minor loss K·(ρv²/2) on top of the straight-section friction — and on a compact plate with tight U-turns those bends can rival the straight-run losses. The K values depend heavily on how sharp the corner is: a square-milled 180° U-turn loses far more than a generously radiused one. The values here (roughly 0.3, 1.1 and 2.0 for 45°, 90° and 180°) are representative of sharp milled bends; radiused or vaned turns are lower.
From heat-transfer coefficient to a base temperature, and the fin-efficiency caveat
How the convective, caloric and conduction resistances stack up to a base temperature, why tall channels don't deliver all the area they seem to, and a cold-plate design checklist.