1

Thermal Management of Hotspots Using DiamondHeat Spreader on Si Microcooler for GaN Devices

Yong Han, Boon Long Lau, Gongyue Tang, and Xiaowu Zhang, Senior Member, IEEE

Abstract— A diamond heat spreader has been applied on thehybrid Si microcooler for the improvement of the hotspots coolingcapability for GaN devices. The microwave chemical vapordeposition diamond heat spreader under tests is of thickness400 µm and thermal conductivity as high as 1500 ∼ 2000 W/mK,and is bonded through the thermal compression bonding processat chip level. Eight hotspots, each of size 450 × 300 µm2, werefabricated on a Si thermal test chip to mimic the heating areasof eight GaN units. Heat dissipation capabilities were studiedand compared through experimental tests and thermal/fluidsimulations, and consistent results have been obtained. Usingthe diamond heat spreader, to dissipate 70-W heating power, themaximum chip temperature can be reduced by 40.4% and 27.3%,compared with the structure without a heat spreader and theone with a copper heat spreader, respectively. While maintainingthe maximum hotspot temperature under 160 °C, 10-kW/cm2

hotspot heat flux can be dissipated. The thermal effects of the heatspreader thickness, the diamond thermal conductivity, and thebonding layer are investigated. Based on the simulation results,the higher power density of the GaN device can be dissipated,while maintaining the peak gate temperature under 200 °C.The concentrated heat flux has been effectively reduced usinga diamond heat spreader, and much better cooling capability ofthe Si microcooler has been achieved for high-power GaN devices.

Index Terms— Diamond heat spreader, GaN device, hotspotthermal management, microjet array impingement.

I. INTRODUCTION

H IGH heat flux removal is a major consideration in thedesign of a number of microelectronic devices, such as

power amplifiers [1], [2]. The operation of a GaN high electronmobility transistor on a silicon chip posts huge challenge to thethermal management, as the heat dissipation concentrates ontiny areas, which could lead to highly nonuniform temperaturedistribution, thus diminishing the device performance andadversely impacting reliability. The effective thermal man-agement will be a key to enable the technology to reach itsfull potential [3]–[5]. Both microchannel and microjet heatsinks can dissipate the high heat fluxes anticipated in highpower electronic devices [6]–[8]. Liquid jet impingement can

Manuscript received March 22, 2015; revised July 2, 2015; acceptedSeptember 7, 2015. Date of publication October 7, 2015; date of current ver-sion December 11, 2015. This work was supported in part by the Silicon MicroCooler Consortium Project and in part by Honeywell Aerospace and ElementSix Technologies.

The authors are with the Institute of Microelectronics, Agencyfor Science, Technology and Research, Singapore 117685 (e-mail:hany@ime.a-star.edu.sg; laubl@ime.a-star.edu.sg; tangg@ime.a-star.edu.sg;xiaowu@ime.a-star.edu.sg).

provide a high heat transfer coefficient, when arranged inarrays. Brunschwiler et al. [9] created a series of microjetarrays with branched hierarchical parallel fluid delivery andreturn architectures. The peak heat transfer coefficient mea-sured was 8.7 × 104 W/m2K. Compared with the impingingmicrojet, the microchannel has a lower averaged heat transfercoefficient. However, the coolant can exchange energy witha larger effective surface area with multiple walls withineach channel. Colgan et al. [10] presented a Si microchannelcooler and optimized the cooler fin for cooling a very highpower chip, and 300-W/cm2 uniform heat flux was dissipated.The hybrid microcooler, combining the merits of both themicrojet array impingement and the microchannel flow, hasbeen applied for cooling a high-power device [11], [12].To dissipate the high concentrated heat flux, the effective heatspreading capability is quite significant as well. A chemicalvapor deposition (CVD) diamond heat spreader of high ther-mal conductivity turns out to be a great candidate for thisapplication [13]–[17]. Rogacs and Rhee [18] conducted thenumerical simulation to evaluate the thermal effect of the dia-mond heat spreader for a small heat source. Calame et al. [19]conducted the experimental and simulation analyses on themicrochannel cooler capped with a layer of diamond, andbetter cooling performances were exhibited than SiC alone.According to Liu’s numerical calculations, a thin diamond heatspreader could enable ∼10% to 20% decrease in the wholethermal resistance of the heat sink [20].

In this paper, the enhancement of heat dissipation capabilityusing a diamond heat spreader on a Si hybrid microcooler hasbeen investigated through experiments and simulations. Thedesigned cooling system is illustrated in Fig. 1. A customizedthermal test chip of eight tiny hotspots is fabricated, andbonded with a Si cooler and a diamond heat spreaderthrough the thermal compression bonding (TCB) process. TheSi hybrid microcooler, combining the microjet impingementarray and the microchannel flow, can achieve a high heattransfer coefficient with a low pumping power requirement.The jet flow from the microjet plate impinges on the top wall,and is constrained to flow along the microchannel to the outlettrench. A diamond heat spreader is directly attached betweenthe chip and the cooler. Cooling performance comparisonshave been made among the structures with a diamond heatspreader, with a copper heat spreader, and without a heatspreader. The improvement of hotspot cooling capabilityusing a diamond heat spreader is more pronounced for thethin chip of thickness 100 μm. Two types of diamond ofdifferent thermal conductivities are tested and compared.

2

Fig. 1. Exploded view of the cooling solution with directly attached diamondbetween the chip and the microcooler.

TABLE I

THERMAL CONDUCTIVITIES OF DIFFERENT DIAMOND HEAT SPREADERS

The thermal effects of the heat spreader thickness and thebonding layer are studied. For a GaN device, the dissipatedpower densities with various cooling solutions are evaluated,while maintaining the peak gate temperature under 200 °C.The consistent results of experiments and simulations havedemonstrated high cooling capability improvement using thediamond heat spreader on the microfluid cooler.

II. FABRICATION AND EXPERIMENTS

The experimental tests have been implemented on thecustomized Si thermal test chip of size 7 × 7 mm2, witheight hotspots evenly located in line, as illustrated in Fig. 1.The size of each hotspot is 450 × 300 μm2, which is a goodapproximation of one GaN unit area composed of ten gatefingers of 300-μm gate width, 0.3-μm gate length, and 45-μmgate pitch. The space between each hotspot is set to be thesame as that of the transistor banks of the GaN device, whichis 690 μm. The highly doped n-type resistors are built on theSi test chip as hotspot heaters. The resistivity measurementof the finished wafers shows good consistency [21].

The diamond heat spreaders supplied by Element Six areprepared through microwave CVD, and metalized with a thinTi/Pt/Au layer (total thickness around 1 μm) for tight bondingwith a Si chip and a microcooler. The size of the heat spreaderis 9 × 9 mm2, and its thickness is 400 μm. Several types ofdiamond heat spreader of different thermal conductivities areconsidered in this paper, as listed in Table I.

Fig. 2. Images of the fabricated hybrid Si microcooler. (a) X-ray top view.(b) Microscopy cross-sectional view.

The hybrid microcooler is fabricated by bonding twoSi plates together, one is with 18 microchannels and the otheris with a 14×18 microjet array. The internal dimensions of themicrocooler have been optimized, considering the combinedeffect of heat convection and conduction. The nozzle diameteris 100 μm, depth is 400 μm, and pitch is 350 μm. Thechannel width, fin width, and depth are 250, 100, and 250 μm,respectively. The channel length is 5 mm. The size of theoutlet trench is 6.5 × 0.5 mm2, and the distance between thenozzle array and the trench is 4.5 mm. The microchannels ormicrojets are etched on different wafers with the deep reactiveion etching process. After etching, the wafer is metalized with4-μm Au/Sn solder by evaporation. Then, these wafers arediced into a single plate, and chip-level bonding is conductedto bond the microchannel plate and the microjet plate together,as shown in Fig. 2. Plate bonding is carried out through theTCB process, in which the 280 °C chuck holding time is 2 min,and the compressive force is 49 N.

After the assembly of the microcooler, the outside surfacesof the bonded microcooler are then metalized for the followingbonding process. All the prepared components, including thechip, the heat spreader, and the microcooler, are bonded inone step through the TCB process under the same conditionas above. The soldering quality at the bonding interface waschecked using both X-ray and scanning acoustic microscopy.No void was detected in the bonding layers. To prepare thetest vehicle, the bottom side of the microcooler is attachedon the Printed Circuit Board using epoxy, and the test chipon the top is wire bonded for electrical connection, as shownin Fig. 3. Copper of measured thermal conductivity 386 W/mK(uncertainty 3.5%) is normally used as the heat spreadermaterial. Another test vehicle with a copper heat spreader ofthe same size was fabricated for performance comparison. Toachieve reliable assembly, the reflow process with a 20-μm-thick Au/Sn preform solder is used for components bonding,in which the peak temperature is ∼230 °C.

The dc power is supplied to the hotspot heaters on thethermal test chip. A water coolant is driven through the flowloop using a microgear pump. This pump forces the waterthrough a 15-μm filter and a flow meter before entering themicrocooler. The heated water from the microcooler is cooleddown by a water bath, and then flows back to the coolingsystem. A differential pressure transmitter is attached to themanifold to measure the pressure drop. The inlet water ambienttemperature is ∼25 °C. The test chip temperature at steady

3

Fig. 3. Photo images of the fabricated and assembled test vehicles with andwithout a diamond heat spreader.

Fig. 4. Image of the simulation model with symmetry boundaries and meshof the fluid part. Blue region: fluid.

state is measured and recorded using an infrared (IR) camera.In the test, the steady state (temperature variation ±0.1 °C) isusually reached within 30 min. Prior to the measurement, thechip is coated with a thin layer of matt black paint, and theemissivity of the coating is estimated at 0.9. The temperaturemeasurement uncertainty is around ±3.5%.

III. THERMAL/FLUID SIMULATION

The simulation models are constructed using COMSOLMultiphysics that runs the finite-element analysis together withadaptive meshing and error control. The built-in fluid flowand heat transfer interfaces are used in the 3-D model thatcouples both solid and fluid parts. Due to the symmetries in thesystem, a quarter of the structure with symmetrical boundaryconditions is constructed, as illustrated in Fig. 4.

The solution was tested for mesh independence by refiningthe mesh size. The velocities and the temperatures matchedwithin 0.1% for both the mesh sizes. The convergence criterionof the solutions is 10−6. The viscous heating feature isconsidered in the heat transfer interface. No slip boundarycondition is applied for stationary walls. Natural convectionfrom the topside of the test vehicle is also included. Thetemperature-dependent thermal conductivity of Si is consid-ered, the expression of which is kSi = 152 × (298/T)1.334.The thermal conductivity of a Au/Sn material is assumed to beconstant, which is 57 W/mK. A Au/Sn bonding layer of 5-μmthickness is considered between the Si and diamond contactinterfaces. The models without and with a heat spreaderconsist of >2.06 and 2.25 million tetrahedral elements, respec-tively. The element size of the fluid part is calibrated forfluid dynamics, while that of the solid part is calibrated for

Fig. 5. Maximum hotspot temperature as a function of total heating powerfor the structure without a heat spreader, with a copper heat spreader, andwith a Type 1 diamond heat spreader from the experimental tests and thesimulation analysis.

general physics. High heat fluxes are loaded only on the tinyareas of the hotspot heaters.

IV. RESULTS AND DISCUSSION

First, the experimental tests are performed on the thermaltest chip of 200-μm thickness. The volume flow rate inthe microcooler is set at 400 mL/min, and the pumpingpower is ∼0.2 W. The heating power from 10 to 100 Whas been loaded into eight hotspot heaters (each of size450 × 300 μm2). The steady-state simulation is performedwith the loading and environment conditions set according tothe tests. Based on the estimated low Reynolds number in themicrocooler, it is considered to be operated in laminar regime.The experimental and simulation results are shown in Fig. 5.A Type 1 diamond heat spreader of thickness 400 μm is usedin this test.

Without a heat spreader, the structure can dissipate 54-Wheating power (hotspot heat flux of 5 kW/cm2), and themaximum hotspot temperature is 136 °C. With the copperheat spreader, the structure can dissipate 54 W, and themaximum hotspot temperature is 124 °C. By doubling thecopper thickness, the improvement is <1%. The directlyattached diamond heat spreader can enable highly improvedheat dissipation capability for the cooling system. For 54-Wpower dissipation, the maximum hotspot temperature canbe reduced by ∼25.7% compared with the one without aspreader, and 17.1% compared with the one with the copperheat spreader. To maintain the chip temperature under 180 °C,the structure with the diamond heat spreader can dissipate∼100 W (9.2 kW/cm2), while that without a spreader candissipate ∼70 W, and the one with the copper heat spreadercan only dissipate ∼80 W.

As shown in Fig. 5, excellent agreement has been obtainedbetween the experimental and simulation results, suggestingthat the thermal performance is accurately simulated. In thehybrid microcooler, the spatially averaged heat transfer coef-ficient at the top impingement wall in the microchannel is∼11.3 × 104 W/m2K, and the local value in the stagnationzone is much larger. The temperature distribution on the topsurface of the chip from the test is illustrated in Fig. 6.

4

Fig. 6. IR camera image of the heating areas in the thermal testchip (a) without and (b) with a diamond heat spreader for 54-W heating.

Fig. 7. Thermal effect comparison of two types of diamond heat spreaderfor different heating powers.

The maximum temperature occurs at the hotspot locatednear the Si chip center, as shown in Fig. 6. For 54-W heating,the maximum temperature of the hotspot located near the chipedge is 8 °C lower than the center one in the structure without aheat spreader. The diamond heat spreader can maintain similarpeak hotspot temperature, and enables 2 °C peak temperaturedifferences between the hotspots located near the edge andthe center. The peak hotspot temperature variation rate (αT )among all the hotspots is calculated to evaluate the coolingperformance. This rate is represented as αT = (TE − TC )/TA,where TC and TE are the peak temperatures of the hotspotslocated near the center and the edge, respectively, and TA isthe average value of the peak temperatures of all the eighthotspots. In the current case, for 54-W power dissipation, therate αT of the structure without a heat spreader is ∼6.2%,while that with the copper heat spreader is ∼5.9%, and theone with the diamond heat spreader is as small as 2.5%.

Another diamond of different thermal conductivities hasbeen used in the thermal test for comparison, as is shownin Fig.7. Type 1 diamond was used in the above tests. Thethermal conductivity of Type 2 diamond drops slightly fromroom temperature to higher temperature. The results showthat similar cooling capability has been achieved using thesetwo types of diamond. To dissipate the same heating power,the maximum hotspot temperature rise using Type 2 diamondinstead of Type 1 diamond is <2%, and the increase in therate αT is <0.4%.

Fig. 8. Maximum hotspot temperature as a function of heating power for a100-μm chip without a heat spreader, with a copper heat spreader, and with aType 1 diamond heat spreader from the experimental tests and the simulationanalysis.

Fig. 9. Simulation heat flux distribution on the top surface of theSi microcooler in the structure (a) without a heat spreader and (b) with adiamond heat spreader for 5-kW/cm2 heat flux loading.

More experiments and simulations have been conducted onthe thin test chip of 100-μm thickness, which means shorterheat conduction path from the heat source to the coolingsolution. The comparison results are shown in Fig. 8.

More pronounced improvement of the hotspot cooling capa-bility has been achieved using the Type 1 diamond heatspreader in this case. Compared with the results shownin Fig. 5, only slight temperature decrease is enabled bymerely reducing the chip thickness from 200 to 100 μm. For70-W heating, the maximum temperature decreases by <2%.However, with smaller thermal resistance between the heatsource and the heat spreader, much better heat dissipationcapability has been achieved. With the diamond heat spreader,110-W heating power (10.2 kW/cm2) can be dissipated whilemaintaining the maximum hotspot temperature under 160 °C.To dissipate 70-W heating power, the maximum hotspottemperature can be reduced by 27.3% with the copper heatspreader, and 40.4% with the diamond heat spreader. Theconcentrated high heat flux from the tiny hotspot is effec-tively reduced for the bottom jet-based microcooler to handle.Fig. 9 illustrates the heat flux distribution on the top surfaceof the Si microcooler for the structure with and without

5

Fig. 10. Effect of the Type 1 diamond heat spreader thickness on the thermalperformance of the cooling structure for 110-W power heating.

a diamond heat spreader. The enhancement effect is quiteobvious.

Smaller and more uniform heat flux has been enabledusing the diamond heat spreader. The maximum heat fluxin Fig. 9(a) is ∼2.66 kW/cm2, while that in Fig. 9(b) is only∼0.39 kW/cm2, suggesting that the concentrated heat flux hasbeen reduced to 14.7% using the diamond heat spreader. Themaximum thermal resistance of the whole cooling structure,which is related to the total heating power and the maximumtemperature of the cooling structure, can be reduced by 72.5%with diamond for the hotspot thermal management.

V. THERMAL EFFECT INVESTIGATION

More simulations have been performed to investigate theeffect of the diamond heat spreader thickness on the ther-mal performance of the structure with a 100-μm-thick testchip. The thickness of the heat spreader is changed from100 to 700 μm. The variations in the maximum hotspottemperature and the maximum heat flux at the top surfaceof the Si microcooler are analyzed, as illustrated in Fig. 10.

A Type 1 diamond heat spreader is used in this simulation.By increasing the heat spreader thickness, reduced temperatureand heat flux can be achieved. The heat spreader used inthe tests is 400-μm thick, and the thermal performance canbe slightly improved by increasing the thickness to 700 μm.The effect is more sensitive by increasing the thickness from100 to 400 μm, where the maximum hotspot temperatureand the heat flux can be reduced by ∼12.1% and 62.3%,respectively. The temperature distribution is also affected. For110-W power heating, the variation rates αT of 6.9%, 3.2%,and 2.6% are achieved with the diamond heat spreader ofthickness 100, 400, and 700 μm, respectively.

Four types of diamond heat spreader of different thermalconductivities, which are listed in Table I, are analyzed andcompared, as shown in Fig. 11. The cooling capabilities ofthe structure with Types 1–3 are quite similar. The maximumhotspot temperature rise by changing the heat spreader fromType 1 to Type 4 is ∼8%. With the Type 4 heat spreader,for 110-W power dissipation, the peak temperature variationrates αT is ∼4.9%, which is higher than Type 1 (3.2%)and Type 2 (3.8%). Even so, with the Type 4 diamond

Fig. 11. Effect of the heat spreader thermal conductivity on the thermalperformance of the cooling structure for different heating powers.

Fig. 12. Simulation temperature profile vertically across the structure withthe diamond heat spreader.

heat spreader of relatively low thermal conductivity, 110-Wheating power can be dissipated, and the maximum hotspottemperature is ∼175 °C.

There are two bonding layers in the structure with thediamond heat spreader, which are quite critical to assurereliable components assembly and low thermal resistance.In the experimental tests, the thickness of both the layersis ∼5 μm. Fig. 12 shows the temperature profile verticallyfrom the Si chip to the microcooler. To dissipate 110-W power,the temperature rise caused at the chip–diamond bonding layeris ∼8.1 °C, while that at the diamond–cooler bonding layeris negligible. The bonding layer on the topside of the heatspreader, which is more close to the heat source, has strongereffect on the hotspot cooling. Further simulation has shownthat the thermal performance is quite sensitive to the thicknessof the bonding layer at chip–diamond interface. By increasingthe thickness from 5 to 10 μm, the temperature rise willincrease by 8.8%, and if from 5 to 20 μm, the temperaturerise will increase by 12.9%. No obvious temperature increaseis observed by doubling or quadrupling the current thicknessof the bonding layer at the diamond–cooler interface.

VI. HEAT DISSIPATION CAPABILITY OF GaN DEVICE

In the above experiments and simulation, one hotspot heateris used to mimic the heating area of one GaN unit consisting

6

Fig. 13. Image of the gate finger simulation model with symmetry boundariesand mesh of the heating regions. Blue region: fluid.

Fig. 14. Temperature profile in the longitudinal direction across ten gatefingers of a 100-μm-thick chip with the diamond heat spreader and thetemperature distribution image for the heating power density of 4.02 W/mm.

of ten gate fingers, and consistent results have been obtained.In this paper, using the same simulation scheme for steady-state study, a model considering the heat producing regions ofgate fingers has been built to evaluate the thermal performanceof the GaN transistor with the designed cooling solution.

A number of assumptions were made to limit the scopeof the investigation. Based on the one-quarter modelshown in Fig. 4, the cooling structure under one hotspot isconsidered in the model illustrated in Fig. 13. Instead of onehotspot, in this case, the heating regions of ten gate fingers(each of size 300 × 0.3 μm2) are constructed. This modelconsiders a 2-μm-thick GaN layer of the thermal conductivitykGaN = 141 × (298/T)1.211 on the top of the Si substrate.The thermal boundary resistance is included between theGaN layer and the Si substrate, and the value is assumed tobe 3.3 × 10−8 m2K/W [22]. The models without and with aheat spreader consist of >1.96 and 2.07 million tetrahedralelements to be mesh-independent. The load and conditionsettings for the fluid inside the microcooler are the same asthe model shown in Fig. 4.

The results displayed in Fig. 14 show that, with the Type 1diamond heat spreader and 0.2-W pumping power for themicrocooler, to maintain the peak gate temperature under200 °C, the heat power density of 4.02 W/mm (total heatingpower of 96.5 W) can be dissipated, which is ∼45.6% more

TABLE II

SIMULATION COMPARISON OF HEAT DISSIPATION CAPABILITY

than that without a heat spreader, and around 28.4% morethan the one with the copper heat spreader. As shown inTable II, both Type 1 and Type 4 can enable much bettercooling capability for the concentrated high heat flux of theGaN device. For the chip of 200-μm thickness, compared withthe structure without a spreader, the dissipated power densitycan be increased by 22% and 18% using Type 1 and Type 4diamonds, respectively. For a thinner chip, the enhancementof the heat dissipation will be more pronounced.

VII. CONCLUSION

The improvement of heat dissipation capability using thediamond heat spreader on the Si hybrid microcooler has beeninvestigated for GaN devices. The experimental tests have beenconducted on the customized thermal test chip of eight tinyhotspots to mimic eight GaN units. A test vehicle with acopper heat spreader was tested for comparison. Two typesof diamond heat spreader of different thermal conductivitieshave been used in the test vehicles. High improvement ofhotspot cooling capability has been demonstrated using thediamond heat spreader, especially for the case with a thinnerthermal test chip. Using the Type 1 diamond heat spreader,110-W heating power can be dissipated while maintaining themaximum hotspot temperature under 160 °C. To keep the max-imum temperature under 160 °C, the structure without a heatspreader can only dissipate 65 W, and the one with the copperheat spreader can dissipate 80 W. The thermal effects of theheat spreader thickness, the diamond thermal conductivity, andthe bonding layer have been analyzed and compared. In thetest vehicle, the bonding layer of thickness around 5 μm hasachieved high bonding quality and low thermal resistance. Thebonding layer at the chip–diamond interface is quite criticalto the thermal performance. The heat dissipation capabilityof the GaN transistors has been evaluated by constructing agate finger model. Higher power density can be dissipatedusing the diamond heat spreader while maintaining the peakgate temperature under 200 °C. By effectively spreading theconcentrated heat flux, the diamond heat spreader is verified tohighly improve the hotspot cooling capability of the Si hybridmicrocooler for GaN devices.

REFERENCES

[1] Y. Mizuno, I. Soga, S. Hirose, O. Tsuboi, and T. Iwai, “Si microchannelcooler integrated with high power amplifiers for base station of mobilecommunication systems,” in Proc. IEEE 61st ECTC, May/Jun. 2011,pp. 1541–1546.

6

[2] A. J. Robinson, “A thermal–hydraulic comparison of liquid microchan-nel and impinging liquid jet array heat sinks for high-power electronicscooling,” IEEE Trans. Compon. Packag. Technol., vol. 32, no. 2,pp. 347–357, Jun. 2009.

[3] X. Chen, F. N. Donmezer, S. Kumar, and S. Graham, “A numericalstudy on comparing the active and passive cooling of AlGaN/GaNHEMTs,” IEEE Trans. Electron Devices, vol. 61, no. 12, pp. 4056–4061,Dec. 2014.

[4] Y. Han, Y. L. Lee, and X. Zhang, “Trapezoidal microchannel heatsink with pressure-driven and electro-osmotic flows for microelectroniccooling,” IEEE Trans. Compon., Packag., Manuf. Technol., vol. 3, no. 11,pp. 1851–1858, Nov. 2013.

[5] K. R. Bagnall, Y. S. Muzychka, and E. N. Wang, “Application of theKirchhoff transform to thermal spreading problems with convectionboundary conditions,” IEEE Trans. Compon., Packag., Manuf. Technol.,vol. 4, no. 3, pp. 408–420, Mar. 2014.

[6] M. K. Sung and I. Mudawar, “Single-phase hybrid micro-channel/micro-jet impingement cooling,” Int. J. Heat Mass Transf., vol. 51, nos. 17–18,pp. 4342–4352, Aug. 2008.

[7] X. Song et al., “Surrogate-based analysis and optimization for the designof heat sinks with jet impingement,” IEEE Trans. Compon., Packag.,Manuf. Technol., vol. 4, no. 3, pp. 429–437, Mar. 2014.

[8] M. Fabbri and V. K. Dhir, “Optimized heat transfer for high powerelectronic cooling using arrays of microjets,” J. Heat Transf., vol. 127,no. 7, pp. 760–769, Nov. 2004.

[9] S. Brunschwiler et al., “Direct liquid jet-impingment cooling withmicron-sized nozzle array and distributed return architecture,” inProc. 10th Intersoc. Conf. Thermal Thermomech. Phenomena Electron.Syst. (ITHERM), San Diego, CA, USA, 2006, pp. 196–203.

[10] E. G. Colgan et al., “A practical implementation of silicon microchannelcoolers for high power chips,” IEEE Trans. Compon. Packag. Technol.,vol. 30, no. 2, pp. 218–225, Jun. 2007.

[11] Y. Han, B. L. Lau, H. Zhang, and X. Zhang, “Package-level Si-basedmicro-jet impingement cooling solution with multiple drainage micro-trenches,” in Proc. IEEE 16th EPTC, Dec. 2014, pp. 330–334.

[12] M. K. Sung and I. Mudawar, “Effects of jet pattern on single-phase cooling performance of hybrid micro-channel/micro-circular-jet-impingement thermal management scheme,” Int. J. Heat Mass Transf.,vol. 51, nos. 19–20, pp. 4614–4627, 2008.

[13] P. Hui and H. S. Tan, “Three-dimensional analysis of a thermal dissipa-tion system with a rectangular diamond heat spreader on a semi-infinitecopper heat sink,” Jpn. J. Appl. Phys., vol. 35, no. 9A, pp. 4852–4861,1996.

[14] P. Hui and H. S. Tan, “Temperature distributions in a heat dissipationsystem using a cylindrical diamond heat spreader on a copper heat sink,”Jpn. J. Appl. Phys., vol. 75, no. 2, pp. 748–757, 1994.

[15] F. Ejeckam et al., “Diamond for enhanced GaN device performance,”in Proc. 14th IEEE ITHERM Conf., May 2014, pp. 1206–1209.

[16] Y. Han, B. L. Lau, X. Zhang, Y. C. Leong, and K. F. Choo, “Enhance-ment of hotspot cooling with diamond heat spreader on Cu microchannelheat sink for GaN-on-Si device,” IEEE Trans. Compon., Packag., Manuf.Technol., vol. 4, no. 6, pp. 983–990, Jun. 2014.

[17] K. Jagannadham, “Multilayer diamond heat spreaders for electronicpower devices,” Solid-State Electron., vol. 42, no. 12, pp. 2199–2208,1998.

[18] A. Rogacs and J. Rhee, “Performance-cost optimization of a diamondheat spreader,” in Proc. 12th Int. Symp. APM, 2007, pp. 65–72.

[19] J. P. Calame, R. E. Myers, S. C. Binari, F. N. Wood, and M. Garven,“Experimental investigation of microchannel coolers for the high heatflux thermal management of GaN-on-SiC semiconductor devices,” Int.J. Heat Mass Transf., vol. 50, nos. 23–24, pp. 4767–4779, 2007.

[20] G. Liu et al., “Design of micro-channel heat sink with dia-mond heat spreader for high power LD,” Proc. SPIE, vol. 8419,pp. 841910-1–841910-8, Oct. 2012.

[21] B. L. Lau, Y. J. Lee, Y. C. Leong, K. F. Choo, X. Zhang, andP. K. Chan, “Development of thermal test chip for GaN-on-Si devicehotspot characterization,” in Proc. IEEE 14th EPTC, Dec. 2012,pp. 746–751.

[22] A. Sarua et al., “Thermal boundary resistance between GaN and sub-strate in AlGaN/GaN electronic devices,” IEEE Trans. Electron Devices,vol. 54, no. 12, pp. 3152–3158, Dec. 2007.

Yong Han received the B.S. degree in electronicand information engineering from Shanxi University,Taiyuan, China, in 2004, and the Ph.D. degreein electronic science and technology from theInstitute of Electronics, Chinese Academy ofSciences (IECAS), Beijing, China, in 2009.

He was an Assistant Researcher with the KeyLaboratory of High Power Microwave Sources andTechnologies, IECAS. From 2011 to 2012, he wasa Research Associate with the Institute for Researchin Electronics and Applied Physics, University of

Maryland, College Park, MD, USA. Since 2012, he has been a Research Sci-entist with the Institute of Microelectronics, Agency for Science, Technologyand Research, Singapore. His current research interests include the thermalmanagement of high power electronic devices, electromagnetic design of themicrowave source, computational modeling and thermal analysis, advancedmicroelectric system packaging development, and electric cooling solutions.

Boon Long Lau received the B.S. degree fromthe National University of Singapore, Singapore,in 2005.

He was with ST Microelectronics, Geneva,Switzerland, where he was involved in process engi-neering. Since 2012, he has been with the Instituteof Microelectronics, Agency for Science, Technol-ogy and Research, Singapore. His current researchinterests include microchannel and through-silicon-via process integration.

Gongyue Tang received the B.E. degree inmechanical engineering from the Beijing Universityof Aeronautics and Astronautics, Beijing, China, theM.E. degree in spacecraft design from the ChineseAcademy of Space Technology, Beijing, and thePh.D. degree in mechanical engineering fromNanyang Technological University, Singapore.

He served as a Mechanical Analyst Team Leaderwith Dyson Operations Pte Ltd., Singapore, anda Principal Engineer with the United Test andAssembly Center, Ltd., Singapore. He is currently a

Research Scientist with the Institute of Microelectronics, Agency for Science,Technology and Research, Singapore, a research institute of the Scienceand Engineering Research Council. His current research interests includeheat transfer, microfluidics, thermal management, and advanced coolingtechnologies for integrated circuit packages and microelectronics systems.

Xiaowu Zhang (SM’10) received the B.S. degreein physics from the National University ofDefense Technology, Changsha, China, in 1986, theM.E. degree in mechanics from the University ofScience and Technology of China, Hefei, China,in 1989, and the Ph.D. degree in mechanical engi-neering from the Hong Kong University of Scienceand Technology, Hong Kong, in 1999.

He was a Lecturer with the Ballistic Research Lab-oratory of China, East China Institute of Technology,Nanjing, China, from 1989 to 1995. He has been

with the Institute of Microelectronics, Agency for Science, Technology andResearch, Singapore, since 1999, where he is currently a Principal Investigatorwith the Interconnection and Advanced Packaging Program. He has authoredor co-authored over 120 technical papers in refereed journals and conferenceproceedings. His current research interests include computational modelingand stress analysis, design for reliability, stress sensors, impact dynamics,advanced integrated circuit (IC) and microsystems packaging development,and 3-D IC integration with through-silicon-via technology.