Microchannel cooling
Theory: narrower channels carry heat better
Overview
I need to cool 100 W from a 1 cm² chip, and I'm told an ordinary cold plate won't do. What's different about microchannels?
Channel size. Flow in narrow channels is laminar and the distance from wall to centre is short, so heat moves into the fluid quickly. With a nearly constant Nusselt number, the heat transfer coefficient grows inversely with channel size. 100 W/cm² is 1 MW per m², so a coefficient of a few kW/(m²·K) would mean hundreds of kelvin between wall and water. Reaching tens of kW/(m²·K) needs channels around 100 μm — at the cost of rapidly rising pressure drop.
Heat transfer coefficient
$D_h$ is hydraulic diameter (4 × area ÷ perimeter) and $k$ fluid conductivity. The value is for fully developed laminar flow in square channels. Near the inlet thin boundary layers raise it further, which matters in short channels; aspect ratio also changes the Nusselt number.
Pressure drop
$f$ is the friction factor, $Re$ Reynolds number, $L$ channel length and $u$ mean velocity. In laminar flow $f\,Re$ is constant, so at fixed velocity pressure drop scales with the inverse square of hydraulic diameter — outpacing heat transfer, which improves only linearly.
1981: the 790 W/cm² experiment
Microchannel cooling leapt to attention with Tuckerman and Pease's 1981 Stanford paper: they etched channels about 50 μm wide into the back of a silicon chip, ran water through them and removed 790 W/cm² — tens of times what air cooling managed then. Years of debate followed over whether conventional correlations fail at the microscale, but careful treatment of entrance effects, wall conduction and measurement error showed ordinary laminar theory largely holds. Today the approach extends to direct liquid cooling of high-performance chips in data centres.
Worked examples
Example 1: channel size, heat transfer and pressure drop
Water, square channels, 10 mm long, mean velocity 1 m/s:
| Hydraulic diameter | Heat transfer coefficient | Pressure drop | Reynolds number |
|---|---|---|---|
| 1 mm | 2.2 kW/(m²·K) | 0.3 kPa | 998 |
| 200 μm | 10.8 kW/(m²·K) | 7.1 kPa | 200 |
| 50 μm | 43.3 kW/(m²·K) | 114 kPa | 50 |
Example 2: wall-to-water difference at 100 W/cm²
Assuming the wetted area is twice the chip footprint (square channels with walls as wide as the channels, fin efficiency 1 — a simple assumption), the wall-to-water difference is about 231 K at 1 mm, 46 K at 200 μm and 12 K at 50 μm.
At 50 μm the temperature difference is small, but the pressure drop is 114 kPa.
Right — 20 times the heat transfer for 400 times the pressure drop. In practice designs balance around 100–300 μm and shorten the flow path; feeding coolant at the chip centre and flowing outward both ways halves the length and the pressure drop. Lower velocity cuts pressure drop but raises coolant temperature rise and heats the outlet side. Flow rate, channel size, length and allowable pressure are decided together — that's the core of the design.
Example 3: required flow rate
| Heat load | 10 K coolant rise | 20 K |
|---|---|---|
| 100 W | 144 mL/min | 72 mL/min |
| 300 W | 431 mL/min | 216 mL/min |
How to approach the analysis
Analysis cautions
- Conjugate heat transfer: heat spreading in the walls (silicon or copper) is significant, so solve solid and fluid together.
- Entrance effects: in short channels high inlet heat transfer raises the average; fully developed values alone are conservative.
- Flow distribution: inlet/outlet manifolds cause uneven flow across parallel channels, leaving some channels hot.
- Property variation: water viscosity roughly halves from 20 °C to 60 °C, affecting pressure drop.
“Performance dropped after a few months”
A microchannel cold plate met its target in testing, but after months of operation chip temperatures rose over 10 K. The cause was clogging: particles, corrosion products and biofilms are not negligible in channels tens of micrometres wide. When some channels clog, flow rises in the rest, but the area over the blocked channels runs hot. Filters, water-quality control and margin in channel size are essential. It is worth analysing the temperature with some channels blocked.
Common mistakes
Mistakes and fixes
| Mistake | Effect | Fix |
|---|---|---|
| Narrowing channels for heat transfer alone | Insufficient pump | Evaluate pressure drop together |
| Using turbulent correlations | Wrong heat transfer | Laminar relations (check Re) |
| Fluid-only analysis | Wall spreading ignored | Conjugate heat transfer |
| Assuming uniform flow | Some channels overheat | Model manifolds |
| Ignoring clogging | Long-term degradation | Margin and water-quality control |
I'd like to try forced convection and channel calculations myself.
Try the forced convection simulator for heat transfer, the pipe flow calculator for pressure drop and the heat exchanger NTU tool for exchanger effectiveness. Related cooling methods are covered in entry length, jet impingement, pin-fin arrays and compact heat exchangers.
Related Topics
Experience the theory firsthand with the interactive simulator for this field
All Simulators