Single-phase, miniaturized convective cooling

Cooling with water is beneficial for a miniaturized cold plate because, to cool the same power density, 4000 times less volume is needed than with air. Miniaturization improves the maximal coolable power density, albeit at the expense of an increased pressure drop. For this reason, approaches that increase the power density while reducing the pressure drop are important.

In jet impingement, the coolant is in direct contact with the backside of the electronic component. This method incurs neither the thermal resistance overhead of a base plate nor the overhead and reliability problem of thermal interface materials. We focus on the design and architecture of a micron-sized round nozzle array with interspersed drainage holes on a face-centered quadratic lattice. The latter has a novel parallel manifold that reduces the pressure drop for large numbers of nozzles. This structure is similar to branched hierarchical networks in animals and plants. These features are fabricated using microfabrication of silicon wafers followed by bonding processes.

A single-phase distributed-return submerged jet-impingement cold plate with a nozzle array having a nozzle pitch of 100 µm was designed and optimized. Using a hierarchical branching concept, the pressure drop of the manifold feeding 50,000 inlet/outlet nozzles was reduced to < 0.1 bar at a flow rate of 2.5 l/min for a chip area of 4 cm².

Four main flow patterns that affect the heat transfer coefficient could be identified: the pinch-off regime, the impingement regime, the transition regime, and the separation regime. In the pinch-off regime at gaps < 25 µm, the heat-transfer coefficient and the pressure drop are inversely proportional to the gap. The most practical range for a cold plate design is the impingement regime with H ≥ 1.2 , where the heat-transfer coefficient is constant even for a changing gap. The impingement regime increases with jet velocity and the pitch-to-nozzle diameter ratio.

We show that the optimal jet plate structure has a nozzle diameter of 25 µm at a cell pitch of 100 µm and a gap of 30 µm.This results in a thermal resistance of 0.15 Kcm²/W. The measured results of the device nearest to the optimum with = 43 µm, a gap of 50 µm and a pitch of 150 µm has a thermal resistance from junction to inlet of 0.17 Kcm²/W, which is sufficient to cool power densities up to 370 W/cm² for a 63 °C temperature increase at the junction. One of the advantages of a small pitch of distributed return jet cold plates is the high cooling-performance uniformity  0.5 °C. Another advantage is the elimination of both the thermal interface material and the base plate of a cold plate with its thermal resistance and reliability issues.

In summary we have demonstrated direct jet impingement chip cooling technology capable of cooling power densities of 400 W/cm² – six times more than the current limit of air cooling and two times the current processor design limitation of 200 W/cm². Direct single-phase liquid impingement on the chip backside eliminates thermal interfaces, which are a major reliability concern and performance bottleneck for all current cooling solutions. Two parallel manifolds with a dual branching concept distribute and drain coolants to and from 50,000 nozzles microfabricated by deep reactive ion etching of silicon. The minor dependence on coolant viscosity in thin impingement gaps and ultrashort ducts allows excellent performance for the cooling of low-temperature CMOS. We are currently focusing on further increases of cooler performance with surface-enhancement structures. We are also testing approaches to form sealed modules after low-temperature bonding to processors or other surfaces.

Coolers with branched hierarchical feed and drain networks are being explored in collaboration with ETH Zurich.

Although the emphasis is on single-phase cooling for short-term applications, two-phase liquid cooling is also being explored as part of a longer-term effort.

 References

[1]	T. Brunschwiler, B. Smith, E. Ruetsche, and B. Michel, "Datacenters with Direct Energy Re-use," Proc. Intl. Supercomputing Conference, June 17-19, Dresden, Germany (2008).
[2]	B. Agostini, R. Revellin, J.R. Thome, M. Fabbri, B. Michel, D. Caimi, and U. Kloter, "High Heat Flux Boiling in Silicon Multi-Microchannels: Part III - Saturated Critical Heat Flux of R236fa and Two Phase Pressure Drops," J. Heat and Mass Transfer, in press (2008).
[3]	B. Agostini, J.R. Thome, M. Fabbri, B. Michel, D. Caimi, and U. Kloter, "High Heat Flux Boiling in Silicon Multi-Microchannels: Part I - Heat Transfer Characterstics of Refrigerant R236fa," J. Heat and Mass Transfer, in press (2008).
[4]	B. Agostini, J.R. Thome, M. Fabbri, B. Michel, D. Caimi, and U. Kloter, "High Heat Flux Boiling in Silicon Multi-Microchannels: Part II - Heat Transfer Characterstics of Refrigerant R2245fa," J. Heat and Mass Transfer, in press (2008).
[5]	B. Agostini, M. Fabbri, J.E. Park, L. Wojtan, and J.R. Thome, "State of the Art of High Heat Flux Cooling Technologies," Heat Transfer Engineering 28(4), 258-281 (2007).
[6]	B. Agostini, J.R. Thome, M. Fabbri, and B. Michel, "High Heat Flux Two-Phase Cooling in Silicon Multi-Microchannels," IEEE Trans. on Components and Packaging Technologies, in press (2007).
[7]	M. Fabbri, A. Wetter, B. Mayer, T. Brunschwiler, B. Michel, H. Rothuizen, R. Linderman, and U. Kloter, "Micorchip Cooling Module Based on FC72 Slot Jet Arrays without Cross-Flow," 22nd IEEE SEMI-THERM Symposium, San Jose, CA, (2006).
[8]	T. Brunschwiler, H. Rothuizen, M. Fabbri, U. Kloter, B. Michel, R.J. Bezama, and G. Natarajan, "Direct Liquid Jet-Impingement Cooling with Micron-Sized Nozzle Array and Distributed Return Architecture," 20th ITHERM Conference, San Diego (2006).