Suppressing high-frequency temperature oscillations in microchannels with surface structures

Thermal management of high performance electronic devices such as gallium nitride (GaN) power amplifiers and solid state lasers is critical for their efficient and reliable operation. Dissipating high heat fluxes with conventional single phase liquid cooling usually requires significant flow rates which can result in high pressure drops and pumping power.¹ On the other hand, flow boiling in microchannels is promising because it harnesses the latent heat of vaporization and therefore has the potential to achieve high heat flux dissipation with reduced flow rates and pumping power. However, a major challenge of two-phase cooling is flow instability which can cause significant temperature and pressure drop oscillations that lead to thermal and mechanical fatigue.^2,3 A few types of flow instabilities include Ledinegg instability,^4,5 density wave oscillation,⁶ and pressure drop oscillation.⁷ The pressure drop type oscillation occurs when the flow loop has a compressible volume upstream (e.g., a surge tank), usually much larger than the volume of the channel, and therefore is associated with a long-timescale oscillation period (1–50 s). Experimentally, long-timescale flow oscillations have been commonly observed, and they are known to lead to severe temperature and pressure drop fluctuations,⁸ and often trigger premature critical heat flux (CHF).^9,10 For microchannels where the channel diameter is much smaller than the capillary length, another type of flow instability with orders-of-magnitude shorter timescale (∼tens of milliseconds) has also been observed.^11–15 It is usually associated with the rapid growth of vapor bubbles expanding along the channel direction in the nucleate boiling regime¹¹ and the fast switching between dry-out and liquid rewetting in the annular flow regime.^12–14 While this high-frequency oscillation has been visualized with high-speed images,^11–15 typical measurement sampling rates in flow boiling studies are not high enough to capture the corresponding temperature and pressure drop oscillations on this timescale. Recent transient measurement studies^16,17 have shown fast (<20 ms) and large-amplitude wall temperature changes as a vapor slug passes through the microchannel. These results suggest that there is a need to further characterize the short-timescale temperature oscillations at various heat flux conditions and provide strategies to mitigate this instability.

Recent efforts to suppress flow instability mainly include inlet restrictors or valves,^9,18 pin fins,^19,20 and surface structures.^13,14,21,22 While inlet restrictors and pin fins are effective in reducing reverse flow, they typically introduce additional pressure drop. The use of surface micro- and nanostructures to mitigate flow boiling instability²³ and enhance the heat transfer performance^{13,14,21,22,24} has been attractive because of their ability to modify wetting and liquid spreading behavior,^25–27 without adding a significant flow resistance. In fact in pool boiling, structured surfaces have been shown to enhance the heat transfer coefficient and CHF by increasing the nucleation density^28,29 and promoting wicking and wetting.^29,30 Recently, surface structures have been incorporated into microchannels to reduce temperature oscillation by promoting an early onset of nucleate boiling,^13,23 and enhance the CHF.^14,21 While the structures have been shown to reduce the long-timescale dry-out event,²¹ their effect on high-frequency flow oscillation has not been investigated in detail.

In this work, we investigated the role of surface microstructures in suppressing surface temperature oscillations during high-frequency flow instability events. We designed and fabricated microchannels with smooth surface and microstructured surface. We measured the high-frequency temperature oscillations of the microchannels under different mass fluxes (100–300 kg/m²s) and heat fluxes with high sampling rates and compared the results with high-speed imaging. We analyzed the frequency and the magnitude of the temperature oscillations, and discuss the effect of surface structures under low and high heat flux conditions, respectively.

We fabricated microchannels (500 μm in width and height, 10 mm in length) on a silicon substrate and bonded the silicon channel walls to a transparent Pyrex glass layer (Figure 1(a)). The microchannel sidewalls have tailored roughness resulting from the etching fabrication process to serve as nucleation sites. Cylindrical micropillars (Figure 1(b)) were etched onto the bottom surface of the channel and their geometries (diameter d = 10 μm, pitch l = 30 μm, and height h = 25 μm) were chosen to have both large permeability and capillary pressure.³¹ Similarly, microstructure shapes such as pyramids or cones may also serve as effective wick geometries³² with dimensions optimized to promote capillary flow, which can be investigated in future work. A 1 μm thick thermal oxide layer was grown to promote hydrophilicity of the structured surface and to serve as an electrical insulation layer for a Pt thin-film heater and four resistance temperature detectors (RTDs) on the backside of the microchannel. The heater and RTDs were incorporated on the backside of the microchannel to emulate the heat generation and measure the temperature of an electronic component (e.g., transistors). Specifically, the distances of RTD1 (T₁) through RTD4 (T₄) from the inlet of the microchannel are 0 mm (inlet), 1.4 mm, 5.7 mm, and 10 mm (outlet), respectively. More details can be found in the supplementary material.

We tested our devices using a customized flow loop (see supplementary material). Three different flow rates (G = 100, 200, and 300 kg/m²s) were investigated, and degassed deionized water (100 °C at 1 atm) was used as the test fluid. The inlet subcooling was 10 °C to minimize sensible heating of the fluid while avoiding boiling upstream of the microchannel. A peristaltic pump was used to avoid contamination of the test fluid. Although the peristaltic pump intrinsically causes oscillations in the liquid mass flow rate, the frequency of the pump head (1.6–5.1 Hz, see supplementary material) is an order of magnitude smaller than that of the flow oscillations observed in the microchannel (20–60 Hz). Therefore, during a single oscillation cycle in the microchannel, the flow supplied by the pump is almost constant. The fluid tubing connecting the outlet of the microchannel and the liquid reservoir was covered with thermal insulation to minimize condensation of the two-phase flow, which may impose an additional mode of instability from flow condensation. Calibration of RTDs and the heat flux calculation (based on the heater input power) are discussed in detail in the supplementary material. During the experiments, the heat fluxes were incremented every 10 min, the temperature and pressure were measured with a sampling rate of 1000 Hz, and high speed videos were recorded at 2000 frames per second.

The heat flux q″ is shown as a function of the time-averaged temperature rise ΔT at 100 kg/m²s in Figure 2(a). T₃ (measured by RTD3), which is the highest temperature measured among the RTDs, was used as a conservative way of estimating the temperature rise (ΔT = T₃−T_sat). Note that ΔT is not the wall superheat and includes the temperature drop across the substrate silicon (500 μm thick). The experiments were terminated when ΔT approached 60 °C to prevent mechanical failure of the devices caused by thermal gradient and high temperature, and the highest heat fluxes in Figure 2(a) are not CHF based on the conventional definition (i.e., no temperature excursion was observed). Although the maximum ΔT (50 °C at 500 W/cm²) for the structured surface microchannel is high, it was achieved with a small mass flux (100 kg/m²s) and can be reduced by increasing the mass flux (see supplementary material, ΔT reduced to 42 °C at 500 W/cm² and 300 kg/m²s).²¹ The experiments were repeated, and similar boiling curves were observed (supplementary material). At a low mass flux (100 kg/m²s), CHF values typically reported in literature³³ for a smooth surface microchannel are lower (<400 W/cm²) and are associated with long-timescale temperature excursions.³⁴ We designed our loop to eliminate any components with substantial compressible volume (surge tank and condenser) which may trigger these long-timescale instabilities. This enabled us to better isolate and capture the short-timescale instability. The boiling curves and pressure drop curves for all of the three mass fluxes investigated can be found in supplementary material. The pressure drops of the smooth surface and the structured surface microchannels are very similar, which is a result of the small dimensions of the microstructures (25 μm tall) compared to the height of the microchannel (500 μm).

We also show the range of the temperature oscillation at each heat flux as the shaded regions in Figure 2(a). The lower and upper bounds were calculated based on 5th and 95th percentile of all the measured temperature (5% of the measured temperature is below or above the bounds), to exclude occasional outlier data. Initially, single phase heat transfer at heat fluxes below 40 W/cm² led to curves with low slopes and the temperature oscillations were small. The onset of nucleate boiling (ONB) is indicated by the drop in ΔT due to an increase of the heat transfer coefficient. Large temperature oscillations (>10 °C) were observed following the ONB. High speed images show that the temperature oscillations at such low vapor quality were mainly caused by oscillation between single phase and two-phase flow (see supplementary material video). Specifically, the temperature decreased during bubble growth and departure (two-phase flow) and increased during the wait time (single phase flow, see supplementary material). As the heat flux was increased, the temperature oscillations reduced due to more frequent nucleation events. No apparent difference was observed for the smooth and structured surface samples because the bubble nucleation occurred from the sidewalls which were the same for both samples. These observations suggest that reducing flow instability at low heat flux requires promoting nucleation and this is supported by studies that incorporated dedicated/designed nucleation sites such as nanowire bundles and reentrant cavities.^23,34

As the heat flux was further increased and the flow transitioned to the annular flow regime (>250 W/cm²), temperature oscillations increased for the smooth surface microchannel and the time-averaged temperature rise also exceeded the structured surface microchannel (Figure 2(a)). High speed visualizations show that the flow switched between complete dry-out and rewetting annular two-phase flow with a frequency approaching 40 Hz (see supplementary material video). This result indicates that although no CHF can be identified from the boiling curve, complete dry-out has already occurred periodically and can only be captured by temperature measurement with a high sampling rate. The time-resolved temperature at a relatively high heat flux (≈400 W/cm²) is shown in Figure 2(b) (smooth surface) and Figure 2(c) (structured surface). The smooth surface microchannel showed fast temperature oscillations with a peak-to-peak value (magnitude) of approximately 10 °C (T₃) and a period of 26 ms. T₂ and T₄ showed a similar oscillation frequency but lower amplitudes due to heat spreading away from the heater. In comparison, the temperature of the structured surface remained relatively constant, and no cyclic oscillations were observed (Figure 2(c)).

We compared the high speed images with the measured temperatures on the backside of the microchannels. At a heat flux of ≈400 W/cm², time-lapse images (Figure 3(a)) show that in one cycle, the annular flow dried out on the smooth surface and remained dry for longer than half of the period of the cycle, until the rewetting flow flushed the channel. The temperature decreased during two-phase flow and increased during dry-out (single-phase vapor). For the structured surface, dry-out was significantly suppressed (Figure 3(b)), which led to a relatively stable and lower temperature. The wicking capability of the structures stabilized the temperature under the short-timescale flow oscillation. The fluctuation of the liquid thickness which can promote (liquid thickness ≤ micropillar height) or eliminate (liquid thickness > micropillar height) thin-film evaporation within the microstructures may have caused the small temperature variation of T₂ and T₄ as observed in Figure 2(c).

We calculated the frequency of the temperature oscillations using the fast Fourier transform (FFT) method (see supplementary material) as shown in Figure 4(a). The frequencies corresponded to oscillation periods from 17 to 50 ms for heat fluxes higher than 150 W/cm². We also obtained the flow oscillation frequency from high speed images by counting the number of cycles within the duration of the video. The two frequencies show good agreement (Figure 4(a)), indicating that the temperature oscillations reflect the flow behavior in the microchannel. The frequency increased with both the applied heat flux and the mass flux. A higher heat flux increased the rate of vapor generation, which accelerated the accumulation of the incoming liquid upstream of the microchannel. A higher mass flux resisted the flow reversal and aided the liquid rewetting process. FFT analysis of the temperature of the structured surface did not show apparent peak frequencies, because the measured temperature was not periodic and was more stable.

We further calculated the magnitude of the temperature oscillations (based on 5 and 95 percentile as described previously) as shown in Figures 4(b)–4(d). The peaks at low heat flux are due to flow instability (wait time approximately 0.1–1 s) following the ONB. As the heat flux increases, bubbles are generated more frequently and the temperature (T) oscillation reduces. For the smooth surface microchannel, the T oscillations increased again at high heat fluxes (annular regime) due to dry-out (as discussed in Figures 2(b) and 3(a)), and the gradual increase in the dry-out duration. In comparison, the structured surface delayed and suppressed dry-out due to capillary wicking, and the magnitude of the T oscillations remained small. This allowed the structured surface microchannel to achieve high heat fluxes (700 W/cm², Figures 4(c) and 4(d)) with stable temperatures.

Figures 4(b)–4(d) also suggest that the short-timescale T oscillations are most severe at a low mass flux (100 kg/m²s). As the mass flux increased to 200 and 300 kg/m²s, the magnitude of the short-timescale T oscillations (without the presence of long-timescale instability) decreased. For example, at the same heat flux of 400 W/cm², the T oscillations decreased from 9 °C to 5 °C as the mass flux increased from 100 to 300 kg/m²s. This is because the higher mass flux conditions had higher oscillation frequencies that reduced the duration of complete dry-out in each oscillation cycle.

At very high heat fluxes, we also observed long-timescale flow instability which led to large-magnitude T spikes. This behavior is indicated by the sharp increase in the T oscillation magnitude of the last two data points in Figures 4(c) and 4(d) (labeled by the arrows). The long-timescale flow instability is usually associated with premature CHF conditions.²¹ Figure 4(e) shows the measured temperature during long-timescale oscillations (q″ = 495 W/cm², G = 300 kg/m²s) for the smooth surface microchannel. Within the long-timescale temperature oscillations, short-timescale oscillations coexisted. Possible causes for these modes include density-wave type oscillations (short-timescale) and pressure-drop type oscillations (long-timescale).³ The additional short-timescale instability during the long-timescale instability can also increase the risk of system failure, which has been previously overlooked. The results in Figure 4 indicate that surface structures are effective to suppress both modes of instabilities.

In conclusion, we investigated temperature oscillations as a result of the high-frequency flow oscillation in microchannels and the effect of surface microstructures on the resulting temperature oscillations. Our major observations are as follows:

1. Both short timescale (ms) and long-timescale (s) instabilities can exist in microchannels. The long-timescale instability (usually associated with system compressibility) can cause temperature excursion (∼ a few seconds). The short-timescale instability (rapid dry-out and rewetting within the microchannel) can result in fast rate of temperature change with high oscillational frequency and is a reliability concern.

2. Short-timescale flow oscillations can lead to fast and large temperature oscillations due to periodic dry-out and rewetting. The wicking surface structures can suppress dry-out and reduce the temperature oscillations. The structured surfaces are especially beneficial at high heat fluxes and during the annular flow regime. At low heat fluxes where vapor quality is low, T oscillation is mainly due to switching between single-phase and two phase flow. The structured surfaces do not help in this regime where promoting nucleation is the key. An ideal microchannel would promote nucleation at low heat fluxes and promote wetting at high heat fluxes.

3. The frequency of the short-timescale oscillations increases with increasing mass flux. The duration of dry-out in each cycle reduces, and the resulting temperature oscillations also reduce. Increasing the mass flux can reduce the magnitude of the short-timescale T oscillations.

Our study suggests that using hydrophilic structures that promote capillary wicking is promising for reducing high-frequency temperature oscillation (thermal fatigue) in high heat flux two-phase microchannel thermal management devices. Future work will include incorporating high-precision pressure sensors locally at the microchannel to investigate the role of surface structures on the pressure drop instability.

See supplementary material for details on device fabrication, experiment, additional experimental results, and data processing method.

The authors would like to acknowledge Dan Preston and Kevin Bagnall at MIT for valuable discussion, and the MIT Microsystems Technology Lab for fabrication staff support, help, and use of equipment. This work was partially funded by the Office of Naval Research (ONR) with Dr. Mark Spector as program manager under Award No. N00014-15-1-2483, and the Cooperative Agreement between the Masdar Institute of Science and Technology (Masdar Institute), Abu Dhabi, UAE, and the Massachusetts Institute of Technology (MIT), Cambridge, MA, USA—Reference 02/MI/MI/CP/11/07633/GEN/G/00.