introduction

while no generic relationship exists to relate component and printed circuit board (pcb) temperature with reliability [1], it has been shown that die circuit performance can be highly sensitive to operating temperature [2], and therefore temperature must be controlled. coupling this need with rising component power dissipation levels [3], shortening product design cycle times, and increased demand for more compact and reliable electronic systems with greater functionality, has heightened the need for thermal design tools that enable accurate solutions to be generated and quickly assessed.

the application of computational fluid dynamics (cfd) predictive tools, which simultaneously solve the appropriate governing equations for the solid and fluid domains, has the potential to achieve these goals [4,5]. however, depending solely on numerical predictions without supporting experimental analysis still remains an unreliable design strategy [6]. to minimise the need to qualify thermal designs produced by cfd analysis, the accuracy and validity of both the modelling methodologies applied and solver capability need to be carefully evaluated. this can only be achieved by comparing cfd predictions to accurate benchmark data for a range of different thermofluids systems that span from component through system level.

this categorisation reflects the complete heat transfer chain of an electronic system, with the heat dissipating components acting as thermal source, and the environment external to the system enclosure, as sink. thus effective thermal analysis needs to be focused at four distinct hierarchical levels [7]; namely: die, component, pcb, and system, with each step ultimately increasing the complexity of the problem posed for cfd analysis.

to date, the predictive accuracy of cfd codes dedicated to the thermal analysis of electronic equipment has not been comprehensively validated. though a number of fundamental benchmark test cases exist against which commercial cfd codes have been evaluated [8,9], it is difficult to translate the findings of such studies to the analysis of the more complex, discrete, three-dimensional, non-isothermal components in air-cooled electronic systems, where the aim is to predict the internal operating junction temperature.

furthermore, typically only simplified and limited studies have been undertaken by vendors prior to market release, and the premise that this work is sufficient to merit accurate analysis of complex thermofluid problems, such as encountered in electronics cooling applications, without the need for extensive experimental validation, is weak. this fundamental limitation has been well recognised in the technical literature, with the result that numerous benchmark studies have been published [10]. however despite the importance of this combined effort, these studies collectively do not constitute a comprehensive assessment of cfd analysis to the whole field of electronics cooling.

consequently, cfd based thermal designs still require experimental verification during the design phase, thereby diminishing the potential of this design method to reduce product development cycle times.

though current cfd softwares are capable of solving with reasonable accuracy the navier-stokes equations for laminar steady flows [11], for turbulent conditions, analysis is computationally constrained to solving these equations as time averaged, with the reynolds stresses modelled by the boussinesq hypothesis.

closure of the reynolds stresses is facilitated by the application of eddy viscosity turbulence models, and it is likely that this approach will continue to be utilised by cfd codes dedicated to the thermal analysis of electronic systems in the foreseeable future. unfortunately, due to the inherent limitations of low order turbulence models, the performance of such codes is not entirely satisfactory.

a detailed discussion on the potential shortcomings of the turbulence modelling approaches employed by these codes is given by rodgers [12], both for the analysis of electronics cooling in general, and more specifically for the analysis of component-pcb heat transfer.

to benchmark predictive accuracy and highlight potential limitations of a cfd code, the discrepancy between prediction and measurement needs to be distinguished from the experimental error. thus, good benchmark tests should exhibit complex thermofluid phenomena that can be accurately characterised by experimental analysis.

for system level analysis, considerable uncertainties may exist in both the physical quantities being measured and the numerical boundary conditions applied. in such instances of uncertainty, coupled with computational discretization constraints prohibiting the explicit modelling of length scales from micron to meter at component to cabinet level respectively, system level analysis can only realistically permit a pragmatic assessment of the applicability of the cfd technique to be used as a design tool in an industrial environment.

in contrast, by analysing in isolation cabinet sub-systems (units), such as populated pcbs and heat sinks, it may be possible both to experimentally characterise the associated thermofluids accurately and construct numerical models having well defined boundary conditions. it is important to stress that even though unit level test cases are less complex than system level, they still pose a significant challenge to a cfd code.

therefore, unit level analysis has the potential to permit both accurate, quantitative assessment of prediction accuracy and the identification of sources of numerical error. such an approach was applied in the work presented in this article, where numerical predictive accuracy for component-pcb heat transfer is assessed.

previous studies relating to benchmarking cfd for component-pcb heat transfer have focused on either single-board mounted components [13-20] or approximations of multi-component pcbs [21-23]. a review of this literature is given in rodgers et al. [12,24,25]. as these studies have not included the analysis of real components on populated or multi-component pcbs, which constitute the majority of pcb applications, rodgers et al. [12,24-27] and eveloy et al. [28] sought to address this weakness by analysing for the first time a multi-component board incorporating four real electronic components.

the cfd code evaluated was flotherm, from flomerics. this study is summarised in this article, where the benchmark strategy, numerical modelling methodology and a selection of results for the forced convection analyses are presented.

benchmark methodology and test cases

three pcb-mounted thermal test components, so16 , tsop 482 and pqfp 2083, were analysed in both natural and 2 m/s forced air flow, individually on single-component pcbs, and collectively on a multi-component pcb. test case complexity was incremented in controlled steps from the least challenging single-component pcbs, to the components individually powered on the multi-component pcb, to the fully powered multi-component pcb.

by comparing the predictions for the individually powered configurations on the populated board; firstly, with those obtained on the single-component pcbs, the effect of changed aerodynamic boundary conditions on predictive accuracy could be assessed; secondly, with those for the fully powered multi-component pcb, the ability of the code to capture component thermal interaction could be investigated. the modelling methodology was assessed by the analyses of the single-component pcbs.

this methodical approach permitted sources of prediction error, associated with either the modelling methodology or limitations of the cfd code, to be isolated. this would otherwise be difficult to achieve due to the complexity of multi-component pcb conjugate heat transfer.

the respective test cases, component architectures and pcb designs, are described in detail in rodgers et al. [12,24,25]. while the single-component pcbs conformed to jedec eia/jesd51-3, a non-standard design was used for the multi-component pcb, which had both the same copper tracking and component layouts on both sides. this design, shown in figure 1, permitted numerical modelling to be confined to one pcb side, as an adiabatic plane could be generated along the pcb central in-plane when the same components are powered at the same power dissipation level on both sides.

figure 1 - multi-component test pcb, showing component locations,
copper tracking distribution and air flow direction.

(note: planes f and g are defined for surface temperature profile
analyses in figure 2(b).)

code accuracy was assessed by comparison between predicted and measured component junction and component-pcb surface temperature distributions using thermal test chips and infra-red thermography respectively. surface temperature measurement permitted junction temperature prediction errors to be related to either component-pcb modelling, or inaccurate prediction of the flow field.

the measurements were performed with the test vehicle mounted in a still-air enclosure and wind tunnel for natural and forced convection respectively. the multi-component pcb forced convection test case was analysed in two opposite flow directions parallel to the pcb surface, figure 1, to assess the impact of the flow phenomena on both component operating temperature and predictive accuracy. perspex4 block obstructions were mounted onto the pcb to introduce different degrees of aerodynamic disturbance in either flow direction.

the forced air flow over this board was experimentally visualised using both a smoke-wire method and a novel paint film evaporation technique developed in this study. these visualisations were used to help assess flow field predictive accuracy, thereby permitting sources of prediction error in component junction temperature to be further investigated.

high measurement accuracy and reproducibility, with minimal thermal resistance variation between samples, established confidence in the experimental benchmark data. further confidence was gained from component structural analysis, undertaken using both destructive and non-destructive testing techniques, which permitted both package structural integrity to be assessed and component geometry to be verified for numerical modelling.

numerical analysis

the numerical software tool used was flotherm, from flomerics, a cfd software dedicated for the thermal analysis of electronic equipment. both flotherm versions 1.4 and 2.1 were employed to permit prediction discrepancies between software versions to be highlighted. this is of importance to the literature as the majority of previously published studies are based on the use of version 1.4 [10]. the differences between these software versions are highlighted by rodgers [12] in terms of predictive accuracy. the results for version 2.1 hold for the current software version, version 3.1, as the upgrades introduced would not impact on predictive accuracy for this study.

rosten's et al. [13] approach was employed for the modelling of component and pcb geometries, with some modifications [12]. reflecting the constraints imposed on a thermal designer in an industrial environment, a pragmatic approach was adopted for component modelling, whereby all component models were based on nominal package dimensions and material thermal properties. acknowledging the difficulties in defining a characteristic dimension [18,29], hence transition reynolds number, that adequately describes the heat transfer characteristic over the pcb, the fluid domain was solved as both laminar and turbulent for all forced convection test cases.

turbulence was modelled using both a zero- (lvel) and two-equation (k-e) eddy-viscosity models. while the k-e data constitute the à priori predictions for turbulent flow analysis, the application of this flow model to system level analysis may be prohibitive due to both grid and computational speed constraints. the lvel model, which is an adequate turbulence model for coarse grid computations of global flow and temperature distributions, was therefore also evaluated due to its greater applicability for system level analysis.

the numerical models for all test cases are described in detail by rodgers [12], where prediction sensitivity to both component geometry and potential uncertainties in material thermal properties is also assessed.

selection of results

due to the space constraints imposed on this article, discussion is confined to the forced convection analyses for which a selection of results is presented. as the majority of real pcb applications are multi-component boards, where more than one component has significant power dissipation, emphasis is placed on the fully powered multi-component pcb test case. numerical predictions are confined to the version 2.1 data for both the laminar and k-e turbulence flow models.

in table 1, prediction error for component junction temperature is presented as both an absolute (°c) and percentage value. in both flow directions for the fully powered multi-component pcb case, using either the laminar or k-e flow model, predictive accuracy is overall within ±10% for both the so16 and pqfp 208 components when accounting for experimental uncertainty. this accuracy would be acceptable at an intermediate product design phase [30], but not sufficient for temperature predictions to be used as boundary conditions for subsequent reliability and electrical performance analyses.

for such analyses, predictive accuracy would need to be within ±3°c [31] or ±5% of measurement. based on structural analysis and numerical parametric studies, the lower accuracy for the tsop 48 component was primarily attributed to uncertainties in the encapsulant thermal conductivity value. while the tsop model predictions must therefore be considered with some scepticism, the good agreement between measured and predicted pcb surface temperature in vicinity of this component [12,24,26,27], indicated that the component-pcb thermal interaction was correctly captured, both on the single- and multi-component pcbs.

while in the forward flow direction, the laminar and k-e flow models yield comparable predictions for the tsop and pqfp components, significant differences exist between flow model predictions for the so16 devices. this has nothing to do with the component type, which is only used to define the location on the board.

thus, the greatest deviation between flow model predictions occurs in the so16-m region, which measurements showed to be the most sensitive to aerodynamic disturbance [12,26]. neither flow model is found to be accurate for all components, indicating that the rules governing the application of a laminar or turbulence model are not clear.

in contrast, junction temperature predictions on the single-component pcbs are flow model insensitive, with their accuracy (±3°c or ±5%) qualifying for use in reliability and electrical performance analyses. this demonstrates the applicability of the component-pcb modelling strategy adopted, with the greater prediction error on the populated pcb being therefore attributable to a weakness of the flow models to predict these more complex flows. thus for the multi-component pcb case, aerodynamic factors significantly influence flow model predictive accuracy in both flow directions. differences between flow model predictions increase with distance from the pcb leading edge.

these differences are more pronounced in the forward flow direction, indicating proportionality to the amount of flow disturbance introduced in the flow field upstream of the component. note that the leading edge obstruction in the forward flow direction is significantly wider and taller than in the reversed flow direction. in the reversed flow direction, the laminar model produces the best predictive accuracy for all components, possibly reflecting the effect of milder flow disturbance being generated upstream.

because of this contradictory evidence, no conclusion can be drawn as to whether laminar or turbulent flow models should be used for this type of flow modelling. the results suggest that ultimately a transitional flow model may be required to predict the complete flow field over populated pcbs, hence yield best predictive accuracy for all components.

however, this need is constrained by the fact that such models are not available in cfd softwares dedicated for the analysis of electronics cooling. this is due to the code vendors' opinion, that the use of more sophisticated turbulence models is generally not justified for the majority of industrial analyses undertaken with their software.

ignoring computational constraints, advanced models may only offer a small improvement in predictive accuracy, providing that both the exact geometry of the problem and all boundary conditions are known to a high degree of accuracy. as such detailed information is generally not available during the design phase, approximations are required which only enable global flow field and heat transfer predictions to be obtained.

in contrast, standard turbulence models can provide efficient analysis and solution stability on simple grids. nevertheless, this line of thought may become outdated as future computational advances will facilitate the application of more advanced calculation strategies and turbulence flow models to industrial product development.

consequently, a need exists to evaluate more advanced codes so as to assess the ultimate capability of cfd analysis, hence the potential accuracy of temperature prediction, hence component life prediction. advanced turbulence modelling approaches for the analysis of component-pcb heat transfer are discussed by rodgers [12].

comparison of component junction temperature predictions between the individually- and fully powered configurations in table 1, reveals that prediction error is in part associated with component thermal interaction not being fully captured. this is most striking for the so16-e component, for which prediction error increases by 3°c between the individually- and fully powered configurations using the laminar flow model.

as its measured temperature rise between the two powering configurations was of only 6.8°c, possible variation of temperature dependent material or fluid property arising from the change of powering configuration does not explain the so16 discrepancy. it therefore must be related to inherent limitations of the cfd code to predict downstream component powered off temperature rise, that is its temperature rise due solely to component thermal interaction.

it is anticipated that this limitation could lead to significant prediction errors on densely packed pcbs, where a higher degree of component thermal interaction may exist, as previously found by anderson [22].

measured and predicted surface temperature profiles over the so16 component on both the single- and multi-component boards are shown in figure 2, to highlight sources of prediction error and differences between flow model predictions. for the single-component pcb case, figure 2(a), while good prediction accuracy for pcb heat spread is obtained upstream of the component and though not shown here, in the span-wise direction [12,26], accuracy decays downstream, indicating a weakness of the cfd code to fully capture downstream flow phenomena and hence predict the board temperature rise.

though for turbulent flow analysis, the surface heat transfer coefficient is calculated using wall functions, interestingly the laminar and turbulent k-e flow models predict similar board surface temperature. this therefore indicates that underprediction of the board surface temperature is more likely to be attributable to an underprediction of the air temperature adjacent to the board downstream of the component, which suggests a possible source of underprediction of junction temperature.

the effect of this limitation is more pronounced for the multi-component pcb, figure 2(b), where greater errors and differences between flow model predictions occur. therefore, prediction error is more likely to be related to inaccurate prediction of the flow field rather than to the modelling methodology.

figure 2 - comparison of measured and predicted component-pcb surface
temperature profiles for the so16 component on the single- and multi-
component pcbs [12,26,27].

this is illustrated by the differences in flow field predictions between the laminar and k-e models in figure 4 for the forward flow. as evident from the vector plots, the k-e turbulent model predicts significantly higher flow velocity, hence mass flow rate between the so16-m and pqfp components, compared to the laminar model.

this therefore results in a higher prediction of convective heat transfer. similar differences between flow model predictions are observed both upstream and on the span-wise right hand side of so16-e. the sensitivity of the so16 components junction temperature predictions to flow model, table 1, therefore results from different predictions of the flow field in these regions.

figure 3 - experimentally visualised flow fields on the multi-
component pcb in the forward flow direction at 2m/s, with the smoke
introduced 25mm upstream and flush with the pcb surface [12,27].

though differences in flow field predictions were not quantified by flow field measurements, numerical energy balance analysis of component heat transfer provided the link between junction temperature- and flow field prediction errors [12,28]. thus, while the laminar and k-e junction temperature predictions differed by 10°c and 8°c for so16-m and so16-e respectively, the predicted energy balances were very similar for each component.

this clearly indicates that the component internal conductive domain is only weakly sensitive to flow model. therefore for these components, prediction error is related to the representation of the convective domain. thus the complexity of the flow phenomena and hence the challenge posed to the cfd code, is illustrated by a smoke flow visualisation of the forward flow above the multi-component pcb in figure 3.

summary

this study has given an insight into the current capabilities of cfd codes dedicated to the thermal analysis of electronic equipment, for the prediction of component-pcb heat transfer. prediction accuracy for a multi-component pcb test case was shown to be acceptable at an intermediate thermal design phase, but not sufficient for temperature predictions to be used as boundary conditions for subsequent reliability and electrical performance analyses.

this was attributable to a weakness of the flow models to predict the complex forced air flows. the study has also provided experimental data that could be used either for assessing the predictive accuracy of other cfd codes and future software up-grades, or other component-pcb modelling methodologies than employed in this study.

though not discussed in this article, the test cases permitted an improved understanding of the physics of component-pcb heat transfer to be gained. the findings also highlight new areas that need to be addressed in future benchmark studies, and how the benchmark strategy could be further developed.

acknowledgements

the research was undertaken at the nokia research center, helsinki, finland. the authors gratefully acknowledge the co-operation of st microelectronics and amd for supplying the thermal test components, and flomerics ltd., uk, for their technical support.

the authors wish to thank; mr. carl-magnus fager, nokia research center, for his assistance in the experimental measurements, dr. john lohan, galway-mayo institute of technology, ireland, for his contribution to the publication work associated with this study, and prof. mark davies, university of limerick, ireland, for his constructive comments on the research.

references

[1] lall, p., pecht, m., and harkim, e., 1997, influence of temperature on microelectronics and system reliability, crc press, new york.

[2] kirschman, r. k., 1990, "low temperature electronics", ieee circuits and devices, vol. 6, part 2, pp.12-24.

[3] azar, k., 2000, "the history of power dissipation," electronics cooling, vol. 6, no. 1, pp. 42-50. [4] prakash, c., 1990, "application of computational fluid dynamics for analyzing practical electronic cooling problems," in heat transfer in electronic and microelectronic equipment, editor, bergles, a. e., hemisphere, new york.

[5] lee, t., chambers, b., and mahalingam, m., 1994, "application of cfd technology to electronic thermal management", proceedings of the 44th electronics components and technology conference, pp. 411-420.

[6] lasance, c. j. m., 1998, "status and challenges in thermal design of electronic parts and systems," in thermal management of electronic systems iii, editors, bardon, j. p., beyne, e., saulnier, j. b., elsevier, paris, pp. 91-107.

[7] nakayama, w., and bergles, a. e., 1990, "cooling electronic equipment: past, present and future," in: heat transfer in electronic and microelectronic equipment, editor, bergles, a. e., hemisphere, new york, pp. 3-40.

[8] freitas, c. j., 1995, "selected benchmarks from commercial cfd codes," transactions of the asme, journal of fluids engineering, vol. 117, pp. 208-218.

[9] fokkelman, w. f., meinders, e.r., and hanjalic, k., 1998, "the backward-facing step flow in the transitional regime," in thermal management of electronic systems iii, editors, bardon, j. p., beyne, e., saulnier, j. b., elsevier, paris, pp. 37-43.

[10] tucker, p. g., 1997, "cfd applied to electronic systems: a review", ieee transactions on components, packaging and manufacturing technology - part a, vol. 20, no. 4, pp. 518-529.

[11] hutchings, b., and iannuzzeli, r., 1987, "taking the measure of fluid dynamics software," mechanical engineering, may edition, pp. 72-76.

[12] rodgers, p., 2000, "an experimental assessment of numerical predictive accuracy for electronic component heat transfer," ph.d. thesis, university of limerick, limerick, ireland.

[13] rosten, h., parry, j., addison, j., viswanath, r., davies, m., and fitzgerald e., 1995, "development, validation and application of a thermal model of a pqfp", proceedings of the 45th electronics components and technology conference, pp. 1140.

[14] adams, v., blackburn, d., joshi, y., and berning, d., 1997, "issues in validating package compact thermal models for natural convection cooled electronics systems", ieee transactions on components, packaging and manufacturing technology - part a, vol. 20, no. 4, pp. 420-431.

[15] zahn, b. a., 1998, "evaluating thermal characterisation accuracy using cfd codes - a package level benchmark study of icepak and flotherm", proceedings of the sixth intersociety conference on thermal and thermomechanical phenomena in electronic systems, pp. 322-329.

[16] chiriac, v. a., and lee, t. y. t., 1999, "thermal strategy for modelling the wirebonded pbga packages," in advances in electronic packaging, editors, agonafer, d., saka, m., and lee, y.c., asme, new york, eep-vol. 26-1, pp. 287-293.

[17] rosten, h. i., and viswanath, r., 1994, "thermal modelling of the pentiumtm processor package", proceedings of the 44th electronics components and technology conference, pp. 1140-1150.

[18] burgos, j., manno, v. p. and azar, k., 1995, "achieving accurate thermal characterisation using a cfd code - a case study of plastic packages," ieee transactions on components, packaging, and manufacturing technology - part a, vol. 18, no. 4, pp. 732-738.

[19] teng, s., lee, t. y. t., mahalingam, m., and joiner, b., 1997, "thermal model of a component package for system level applications", journal of electronics manufacturing, vol. 7, no.2, pp. 115-127.

[20] ramakrishna, k., thomas, t. r., lee, t. y., trent j. r., and hause, j. v., 1998," thermal performance of an air-cooled plastic ball grid array package in natural and forced convection," proceedings of the sixth intersociety conference on thermal and thermomechanical phenomena in electronic systems, pp.27-34.

[21] behnia, m., and nakayama, w., 1998, "numerical simulation of combined natural convection-conduction cooling of multiple protuding chips on a series of parallel substrates," proceedings of the sixth intersociety conference on thermal and thermomechanical phenomena in electronic systems, pp. 135-142.

[22] anderson, a., 1997, "a comparison of computational and experimental results for flow and heat transfer from an array of heated blocks", transactions of the asme, journal of electronic packaging, vol. 119, march, pp. 32-39.

[23] drabbels, j. p. a., 1997, "natural convection heat transfer of metal cuboids flush mounted in a horizontal plate," in thermal management of electronic systems ii, editors, beyne, e., lasance c. j. m., and berghmans, j., kluwer academic publishers, dordrecht, pp. 273-280.

[24] rodgers, p., eveloy, v., lohan, j., fager, c.m., tiilikka, p., and rantala, j., 1999, "experimental validation of numerical heat transfer predictions for single- and multi-component printed circuit boards in natural convection environments," proceedings of the fifteenth ieee semiconductor thermal measurement and management symposium, pp. 55-64.

[25] rodgers, p., eveloy, v., lohan, j., fager, c.m., and rantala, j., 1999, "experimental validation of numerical heat transfer predictions for single- and multi-component printed circuit boards in a forced convection environment: part i - experimental and numerical modelling," 33rd asme national heat transfer conference, nhtd99-167.

[26] rodgers, p., eveloy, v., lohan, j., fager, c.m., and rantala, j., 1999, "experimental validation of numerical heat transfer predictions for single- and multi-component printed circuit boards in a forced convection environment: part ii - results and discussion," 33rd asme national heat transfer conference, nhtd99-325.

[27] rodgers, p., lohan, j., eveloy, v., fager, c.m., and rantala, j., 1999, "validating numerical predictions of component thermal interaction on electronic printed circuit boards in forced convection air flows by experimental analysis," in advances in electronic packaging, editors, agonafer, d., saka, m., and lee, y.c., asme, new york, eep-vol. 26-1, pp. 999-1009.

[28] eveloy, v., lohan, j., and rodgers, p., 2000, "a benchmark study of computational fluid dynamics predictive accuracy for component-printed circuit board heat transfer," ieee transactions on components and packaging technology, volume 23, number 3, pp. 568-577.

[29] saulnier, j. b., 1997, "the numerical modelling of heat transfer in electronic systems: challenges and ideas for answer," in thermal management of electronic systems ii, editors, beyne, e., lasance c. j. m., and berghmans, j., kluwer academic publishers, dordrecht, pp. 3-15.

[30] mack, b., and venus, t., 2000, "thermal challenges in the telecom and networks industry," electronics cooling, vol. 6, no. 2, pp. 44-49.

[31] lasance, j. m., 1995, "the need for a change in thermal design philosophy," electronics cooling, vol. 1, no. 2, pp. 24-26.

about the authors:

peter rodgers holds the ph.d. degree in mechanical engineering from the university of limerick, ireland and has over ten years experience in electronics thermal management. his current research activities include the application of numerical analysis to electronics cooling.   he was formerly with the nokia research center, finland, where he consulted on all levels of electronics thermal management and lead research on benchmarking the predictive accuracy of commercial cfd codes dedicated for the thermal analysis of electronic equipment.   for this work, dr. rodgers was awarded the 1999 harvey rosten award for excellence. other research interests include the application of hot-wire anemometry to measure natural convection flow fields. dr. rodgers has authored or co-authored over twenty five referred conference and journal publications.

valérie eveloy received the m.sc. degree in physical engineering from the national institute of applied science (insa), france in 1994.   she was previously a research engineer at nokia, finland, working on both applied research and product development. her activities related to both thermal and electrical performance of rf packaging, and electronics thermal management.   she is currently pursuing graduate studies in thermal-fluid sciences applied to electronics cooling. she has authored or co-authored more than ten referred conference and journal papers.