Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 20, NO.

3, MAY 2005 687

Power Electronics Cooling Effectiveness


Versus Thermal Inertia
Luc Meysenc, Member, IEEE, M. Jylhäkallio, and Peter Barbosa, Member, IEEE

Abstract—Today, the popularity of power electronics integration 3) snubber [6];


is increasing. Despite the prospect of fully integrated module, in- 4) control [7];
cluding features like driving and control electronics, protection, 5) cooling [8].
power integration has not taken-off for medium to high power elec-
tronics applications. Manufacturing issues such as yield, reliability Cooling integration is highly promising from the manufac-
and return-on-investment for a new fabrication line are the major turing, cost, and risk management aspects, as the impact on the
blocking points. As a first step toward integrated modules, inte- manufacturing processes could be minimized and the gain at
gration of the cooling system appears realistic and cost effective. the system level significant. Unlike circuit integration, cooling
Increasing the cooling effectiveness could double the output cur- does not affect significantly the reliability of the device since it
rent of an inverter while using the same amount of silicon. On the
is a passive equipment. Without increasing the complexity of
other hand, integrated cooling leads to small thermal inertia, which
can generate high temperature variation under load cycling con- the power devices, more efficient integrated cooling solution
dition. This paper highlights the relationship between cooling ef- allows increased current density per cm of silicon. To high-
fectiveness and thermal inertia. Typical performances of several light this fact, loss calculations were performed for a two-level
cooling systems are compared under load cycling condition to ex- three-phase inverter containing nonpunch through (NPT)
plain how to take into account the variation of the losses in the 1.7 kV insulated gate bipolar transistor (IGBT) technology,
choice of a cooling technique at the design stage. As an example,
a standard liquid cooled plate performed similar to an integrated considering 1 cm active area per IGBT position (Fig. 1). The
microchannel network for specific load variation frequencies. results, plotted in Fig. 2, were computed for a 900-V dc link
inverter switching at a carrier frequency of 750 Hz, with a mod-
Index Terms—Liquid cooled plate, load variation frequencies,
ulation index of 0.9 and an output power factor of 0.85. This
power electronics integration, silicon, thermal inertia.
operating condition corresponds, for example, to an inverter
driving a motor in traction mode at nominal speed and voltage.
I. INTRODUCTION Increasing the cooling effectiveness from 50 W/cm (typi-
cally a standard forced air cooled heatsink) to 120 W/cm (typ-
T ODAY, there is a growing trend toward power electronics
integration [1], [2]. In most cases, this trend is technology
driven and for a given application, gain in terms of cost, relia-
ically an efficient liquid cooling), allows doubling the output
current of the inverter for the same quantity of silicon (from
bility or manufacturability are often not achieved. As manufac- 40 A/cm to 80 A/cm ). The cost saving is significant and com-
turing quality and stability depend strongly on the production pensates the extra effort needed to integrate the cooling into the
volumes, it is difficult to setup a production line for integrated devices.
power electronics for medium to high power applications (here Integrated cooling is more effective with liquid cooling as
defined as above 200 V–20 kW), where the amount of equip- compact and efficient cooling can be achieved. However, as the
ment sold is relatively small and the number of lost production mass of material involved in the cooling path is significantly
batches before stabilizing the processes have a large impact on reduced, the thermal inertia becomes smaller and the system
the final cost. becomes more sensitive to load variations. Considering 1 cm
However, in a medium term perspective, despite heavy in- of material, uniformly heated on one side and having constant
vestments, significant technological risks and complex business temperature on the other side (Fig. 3), a simple RC thermal
models, power integration is the most promising option to dras- equivalent electrical model can be established, the losses being
tically cut down the cost of power electronics equipments. modeled as a current source injecting power in a parallel RC
Power integration does not have a well defined perimeter and circuit. The thermal response to a 1 W step input losses is
like all integration fields has no limit on the features that can plotted in Fig. 3 for several typical materials used in power
be embedded. However, integration in the area of high power electronics packaging (Table I).
semiconductor modules generally considers embedding: All the materials, except for Al O , listed in Table I possess
a time constant of approximately one second and irrespective
1) protection circuits [3], [4];
of the choice of the material used in the design of an integrated
2) driving circuits [5];
cooling setup, the thermal inertia will remain small and will not
function as a thermal capacitor without significant mass.
Manuscript received April 28, 2004; revised January 10, 2005. Recom- For air-cooled systems to be efficient, large surface areas are
mended by Associate Editor J. A. Ferreira. required to spread the heat and maximize the final exchange
The authors are with the ABB Switzerland, Ltd., Corporate Research Centre, area. Large surface area, combined with thick diffusion layer
Baden-Dättwil, Switzerland (e-mail: luc.meysenc@ch.abb.com; mervi.jyl-
hakallio@ch.abb.com; peter.barbosa@ch.abb.com). results in high thermal inertia, but makes this cooling technique
Digital Object Identifier 10.1109/TPEL.2005.846548 less suitable for integration.
0885-8993/$20.00 © 2005 IEEE
688 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 20, NO. 3, MAY 2005

Fig. 1. Typical characteristic of 1 cm active area NPT IGBT devices.

2) Direct temperature measurements on chip surface using


infrared cameras or contact optical fibers. Several tem-
perature values corresponding to several locations on the
chips are given by these methods, but one has to choose
a single value that could be the average or the maximum
temperature for example.
3) Numerical simulation methods giving the temperature
distribution such as finite elements for example, but one
has still to choose a representative value [11].
4) Mathematical models which could give a simplified
value corresponding to some average.
Knowing all these different definition methods, it becomes clear
that the junction temperature should only be considered as the
most simplified reliability model of a device. If the designer con-
Fig. 2. Output current of a 6-cm two-level inverter versus the cooling siders the operating and cycling temperature ranges specified by
effectiveness. the manufacturers, one can reasonably predict the lifetime and
the reliability of the systems [12].
The purpose of this paper is to present advantages and draw- However, for a deeper understanding of the thermal manage-
backs of the low inertia, efficiently integrated cooling versus ment of power electronics devices it is important to comprehend
standard cooling techniques and to identify the best cooling the fundamental points mentioned hereafter and summarized in
solution for each application. Finally, the expected benefits of Fig. 5.
cooling integration will be shown. The junction temperature must be associated with a
clearly defined method. It can be derived from numer-
ical simulation, mathematical models or measurements.
II. JUNCTION TEMPERATURE IN POWER ELECTRONICS Once the junction temperature is defined, the same def-
It is worth recalling what junction temperature is in power inition should be used to process a set of measurements
electronics. The junction temperature of a device (one should or simulations in order to identify the failure modes of
rather speak about virtual junction temperature or equivalent the system relative to the temperature (absolute value or
junction temperature) is meaningless as long as it is not related variation).
to one or several failure modes of the device. To better under- The failure modes identified must be correlated to the
stand the fuzzy nature of the term “junction temperature,” it is junction temperature in order to have predictive model
important to know that the silicon temperature is not uniform of the thermal behavior of the system.
over a multichip semiconductor module. Variations can occur The designer will have to design the system according to
from chip to chip as well as within the same chip in steady state the chosen definition of the junction temperature.
and in dynamic conditions (see Fig. 4). Today, this process is implicitly applied by the semiconductor
There are various methods to evaluate the so called “junction industry, which provide in the datasheet of the devices thermal
temperature” and each of them will give a different value. resistances, thermal capacitances and operating limits.
1) Indirect temperature measurement methods using The problem of the definition of the junction temperature
electrical “temperature-sensitive” parameters such as reappears with power integration. This time, however, the man-
collector-emitter voltage (Vce) or gate threshold voltage ufacturing process changes, new steps are introduced generating
(Vgeth). These methods result in a single temperature new potential failure modes, and new validation methodology to
value corresponding to a sort of average temperature correlate these failure modes with a junction temperature defini-
[9], [10]. tion are needed. This qualification, time consuming and costly,
MEYSENC et al.: POWER ELECTRONICS COOLING EFFECTIVENESS 689

Fig. 3. Thermal model for 1 cm of material uniformly heated on one side and with constant temperature on the other side.

TABLE I TABLE II
TYPICAL THERMAL PROPERTIES OF CLASSICAL MATERIALS USED RELIABILITY PREDICTION—ACCELERATION FACTORS
FOR POWER ELECTRONICS PACKAGING. THERMAL TIME CONSTANT,
THERMAL RESISTANCE, AND THERMAL CAPACITANCE ARE FOR
THE CASE SHOWN IN FIG. 3

reliability of the silicon chips, interconnections (wire bonds,


electrical terminals, solder bumps), and also all the soldered
and glued elements such as substrates or housing. For each
of these potential failure modes, a predictive methodology
(measurements or simulations) must be in place [13]–[15].
To shorten the duration of these qualification tests, acceler-
ated tests are carried out under high stress level. Then, based on
the results of these accelerated tests, the behavior under normal
stress condition must be extrapolated. Basic rules to define ac-
celeration factors are given in Table II.

III. POWER ELECTRONICS COOLING


Today, for industry or traction applications, depending on the
Fig. 4. Temperature distribution over two IGBT chips belonging to the same availability of coolant, the cooling techniques are usually one of
module in steady state conditions (105 W/chip, reference temperature at 313 K).
those listed as follows:
1) natural air cooled heatsink;
2) forced air cooled heatsink;
3) liquid cooled plate;
4) heatpipe.
As shown in Fig. 6, the thermal time constant of a cooling setup
decreases when the cooling effectiveness increases. Cooling ef-
fectiveness and thermal time constant ranges given in Fig. 6
can commonly be achieved without additional optimization of
cooling design and the performances given here should be con-
sidered as order of magnitude.
Thermal time constants vary over three to four orders of
Fig. 5. Relationship between junction temperature and failure modes. magnitude and will strongly influence both maximum tempera-
ture and temperature variation. Consequently, when the load of
is one of the major drawbacks to the development of power elec- a converter is cycling, increasing the cooling effectiveness will
tronics integration. tend to decrease the average temperature, however, the tem-
Generally, qualification aims at predicting the sensitivity perature variation around this average value will increase and
of the system to the absolute temperature, as well as to the counter the gain. This is especially true when efficient cooling
temperature variation amplitude and to the temperature variation is integrated such as a microchannel network. In this case, ex-
frequency. These parameters influence the lifetime and the tremely low thermal resistance can be achieved ( 0.1 K/Wcm
690 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 20, NO. 3, MAY 2005

which breaks the thermal boundary layer against the channel


wall [24].
With two-phase heat exchanger (Fig. 8), the same level of
cooling effectiveness as that obtained with single phase forced
convection can be achieved at a much lower pressure drop and
flow rate (lower by 10 to 100 times) [25], [26]. The low cost of
the cooling loop is a significant advantage compared to the clas-
sical forced convection heat exchanger. However, some extra
constraints such as critical heat flux, control of the vacuum in
the loop, and pumping cavitations must be taken into account,
which makes two-phase heat exchanger less attractive for power
electronics [27], [28].

IV. THERMAL INERTIA VERSUS COOLING EFFECTIVENESS


The thermal loss in a device is defined as the average energy
dissipated over a pre-defined period of time. If one considers
Fig. 6. Typical Cooling effectiveness and thermal time constant of usual the three-phase inverter in Fig. 1, and plots the losses of one
cooling techniques. IGBT versus the time, the first approximation to consider is the
averaging period. Depending on the phenomena of interrest, the
average losses can be computed over:
1) the switching period (few hundred microseconds);
2) the inverter output period (few milliseconds);
3) the load cycling period (few seconds).
Regarding failure modes that are usually considered in power
electronics (wire bonds lift-off, semiconductor or substrate
soldering fatigue, peak temperature), averaging the losses over
the inverter output period is sufficient. As a study case, the loss
profile given in Fig. 9 is considered. It is a square profile with
an average value of Po and a duty cycle of 0.5. To simplify
the computation of the temperature response, this loss profile
is modeled by the sum of a continuous component and a
sinusoidal component.
Fig. 7. Measured thermal resistance from junction to liquid of a network of
microchannel heated by a single IGBT chip. Cooling systems are modeled according to Fig. 3, using a
parallel first order RC circuit. Then, the maximum temperature
is obtained by superposing a continuous temperature drop and
from junction to outlet liquid, Fig. 7) [16]–[18]. Under this sinusoidal temperature variation
condition, IGBT chips are able to operate with losses around
200 W/cm without the possibility of smoothing junction
temperature variation with the load cycling. with (1)
Several directions are investigated for integrated high effec-
tiveness cooling. As mentioned earlier, a liquid cooling system For improved accuracy, a full harmonics model using the de-
is the best choice for cooling integration, since power integra- composition in Fourier’s series of the losses can replace this
tion is not compatible with large heat diffusion areas required approximation. For each harmonic, the corresponding temper-
by air cooled system. However, one of the main drawbacks of ature drop is calculated using the thermal impedance value at
forced convection liquid microchannels is the high-pressure that frequency. In electrical engineering the resistance and the
drop across the network. capacitance can vary with the frequency (variable permittivity,
With convective heat transfer, the heat exchange coefficient skin and proximity effects), while thermal resistance and capaci-
being proportional to the friction factor, an increase in the tance, for the considered frequency range, are constant, allowing
exchange effectiveness will also increase the pressure drop the decomposition in Fourier series
[19]. To limit the size of the pumping equipment and design a
low pressure drop cooler, one should also consider two-phase
heat exchangers with a boiling-condensing loop [20]. In this
case, the heat exchange coefficient is no longer proportional
to the pressure drop, since the main part of the pressure drop
does not come from the friction but from the acceleration [21],
[22]. Very high exchange coefficients can be achieved using
two-phase heat exchanger [23]. Contrary to common belief,
such high heat exchange coefficient is not due to the energy
(2)
used for the vaporization of the liquid but due to nucleation
MEYSENC et al.: POWER ELECTRONICS COOLING EFFECTIVENESS 691

Fig. 8. Two-phase integrated heat exchanger cooling a single IGBT chip and comparison between measured and computed heat exchange coefficients.

TABLE III
TYPICAL THERMAL RESISTANCE AND THERMAL CAPACITANCE PER
cm FOR DIFFERENT COOLING SYSTEMS

Fig. 9. Typical normalized loss profile in an IGBT of an inverter driving a


cycling load.

The temperature rise versus the time period of the square losses
is computed for several cooling systems given in Table III and
is plotted in Fig. 10. Values are normalized per cm of IGBT
active area. It is interesting to note that despite a higher thermal
resistance (twice), the liquid cooled plate has almost the same
performance of a high effectiveness integrated micro heat ex-
changer when the period of the losses is around 10 s.
Regarding air-cooled systems, natural convection cooling
presents lots of advantages. The implementation of such system
does not require too much design effort or mechanical com-
ponents and there is little maintenance required. However,
cooling effectiveness is limited and for a better utilization of the
semiconductor, most of the time, a fan is installed to enhance
the convective heat transfer. This additional effort and cost are
compensated by the semiconductors controlling more power.
However, under load cycling conditions, depending on the
load variation frequency, the gain might not be as high as ex-
pected compared to the thermal resistance alone (Table IV). Fig. 10. Normalized peak temperature (1T =P ) versus the period of a
square wave losses excitation.
For a load cycling period around 800s, junction temperature
obtained with a forced convection cooling system is only 25%
lower than the junction temperature obtained with a natural con- At the design stage of a load cycling application, it is very
vection system, although, the forced convection thermal resis- important to consider the load variation of frequency before
tance is almost two times smaller. choosing the cooling system.
692 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 20, NO. 3, MAY 2005

TABLE IV temperature varies significantly around its average value. Such


COMPARISON OF NORMALIZED PEAK TEMPERATURE AND THERMAL temperature variation greatly increases stress in the packaging
RESISTANCE FOR NATURAL CONVECTION AND FORCED CONVECTION
of the device (wire bonds, soldering fatigue, etc). The integra-
tion of the cooling then is not of much interest and standard
cooling systems (liquid cooled plate, heatpipe, etc) display
the same level of performance. When the load is non cyclic
or when the cycle period is such that even standard cooling
solution does not smooth the temperature out, power integration
becomes very attractive. As explained in Fig. 2, improving
the cooling effectiveness can significantly increase the output
current of an inverter. This allows significant cost reduction of
equipment and justifies all the research and development effort
made toward integrated cooling. Today, both customers and
manufacturers are becoming familiar with integrated cooling
and power integration, and the acceptance of these solutions
will grow quickly over the coming years. There is still a lot
of work to adapt the semiconductor industry manufacturing
processes, to reach acceptable production yield and quality,
but these obstacles will be surmounted and power devices with
integrated cooling will become more common like today’s
devices.

ACKNOWLEDGMENT
Fig. 11. Position of cooling system corner frequency versus losses spectrum. The authors wish to thank Dr. S. Gunturi for his suggestions
and the help he provided for the correction of this work.
Depending on the relative position of the corner frequency of
the cooling system versus the losses frequency spec- REFERENCES
trum, loss variations may or may not be damped. The damping [1] F. C. Lee, J. D. Van Wyk, D. Boroyevich, G.-Q. Lu, Z. Liang, and P. Bar-
of the junction temperature variation can result in similar power bosa, “Technology trends toward a system-in-a-module in power elec-
tronics,” IEEE Circuits Syst. Mag., vol. 2, no. 4, pp. 4–22, Apr. 2002.
being handled by the semiconductor without the extra cost of [2] J. D. Van Wyk, “Power electronics technology at the dawn of a new cen-
a lower thermal resistance cooling system. In that respect, the tury-past achievements and future expectations,” in Proc. Power Elec-
position of the first harmonic regarding the corner frequency of tron. Motion Control Conf. (PIEMC’00), vol. 1, Aug. 15–18, 2000, pp.
the cooling system indicates if the temperature variation will be 9–20.
[3] T. Kajiwara, A. Yamagaguchi; Y. Hoshi, K. Sakurai, and J. Gallagher,
damped or not by the cooling system. “New intelligent power multi-chips modules with junction temperature
As an example, Fig. 11 shows the position of the losses first detecting function,” in Proc. Industry Applications Conf. (IAS’98), vol.
harmonic versus corner frequency of two different cooling sys- 2, Oct. 12–15, 1998, pp. 1085–1090.
[4] J. Saiz, M. Mermet, D. Frey, P. O. Jeannin, J. L. Schanen, and P. Musz-
tems. Both yield similar maximum junction temperatures de- icki, “Optimization and integration of an active clamping circuit for
spite very different thermal resistances. The low effectiveness IGBT series association,” in Proc. Industry Applications Conf. (IAS’01),
cooling system will fully damp the load variation while the ef- vol. 2, 30 Sep.–4 Oct. 2001, pp. 1046–1051.
[5] K. Ishii, H. Matsumoto, M. Takeda, A. Kawakami, and T. Yamada, “A
ficient cooling system will not damp them at all. high voltage intelligent power module (HVIPM) with a high perfor-
In this particular case, extra cost and effort for a more effec- mance gate driver,” in Proc. 10th Int. Symp. Power Semiconductor De-
tive cooling system does not pay back. Moreover, despite similar vices ICs (ISPSD’98), Jun. 3–6, 1998, pp. 289–292.
maximal junction temperatures, the aging of the devices will be [6] H. J. Beukes, J. H. R. Enslin, and R. Spee, “Integrated active snubber for
high power IGBT modules,” in Proc. Applied Power Electronics Conf.
faster due to larger temperature excursions (Table II). Expo (APEC’97), vol. 1, Feb. 23–27, 1997, pp. 161–167.
[7] R. Chen, F. Canales, B. Yang, P. Barbosa, J. D. Van Wyk, and F. C. Lee,
“Integration of electromagnetic passive components in DPS front-end
V. CONCLUSION DC/DC converter—A comparative study of different integration steps,”
in Proc. Applied Power Electronics Conf. Expo (APEC’03), vol. 2, Feb.
To reduce cost of power electronics equipment, designers try 9–13, 2003, pp. 1137–1142.
to increase the effectiveness of the cooling to get more power [8] L. Meysenc, L. Saludjian, A. Bricard, S. Rael, and C. Schaeffer, “A high
heat flux IGBT micro exchanger setup,” IEEE Trans. Comp., Packag.,
out of the devices. However, increasing the cooling effective- Manufact. Technol. A, vol. 20, no. 3, pp. 334–341, Sep. 1997.
ness reduces the thermal inertia. Consequently, when the load [9] R. Perret, C. Schaeffer, and E. Farjah, “Temperature evolution in power
is cycling, on one hand, the average temperature over the cycle semiconductor devices: measurement techniques and simulation,” Proc.
Inst. Elect. Eng., pp. 10/1–10/7, Oct. 22, 1992.
decreases, but on the other hand, the temperature variation [10] S. Rael, C. Schaeffer, and R. Perret, “Electrothermal characterization
around the average value increases. This is very important of IGBT,” in Proc. Industry Applications Soc. Annu. Meeting (IAS’94),
when the cooling is integrated to the device. Significantly lower Oct. 2–6, 1994, pp. 1269–1276.
[11] T.-Y. Lee, “Design optimization of an integrated liquid-cooled IGBT
thermal resistance can be achieved, but due to the low mass power module using CFD technique,” IEEE Trans. Compon. Packag.
of the system, the loss variations are not filtered and junction Technol., vol. 23, no. 1, pp. 55–60, Mar. 2000.
MEYSENC et al.: POWER ELECTRONICS COOLING EFFECTIVENESS 693

[12] H. Khatir and S. Lefebvre, “Thermal analysis of power cycling effects on Luc Meysenc (M’01) received the M.S. degree in
high power IGBT modules by the boundary element method,” in Proc. electrical engineering and the Ph.D. degree from
17th Annu. IEEE Symp. Semiconductor Thermal Measurement Manage- the National Polytechnic Institute of Grenoble,
ment, Mar. 20–22, 2001, pp. 27–34. Grenoble, France, in 1995 and 1998, respectively.
[13] H. D. Lambilly and H. O. Keser, “Failure analysis of power modules: He was with Alstom Transport SA for three years,
A look at the packaging and reliability of large IGBTs,” IEEE Trans. where he was in charge of the Power Electronics
Compon., Hybrids, Manufact. Technol., vol. 16, no. 4, pp. 412–417, Jun. Thermal Laboratory and managing a research team
1993. dealing with power integration. In December 2000,
[14] J. M. Thebaud, E. Woirgard, C. Zardini, and K. H. Sommer, “High he joined ABB Corporate Research, Baden-Dättwil,
power IGBT modules: Thermal fatigue resistance evaluation of the Switzerland, where he is currently leading a research
solder joints,” in Proc. Int. Workshop Integrated Power Packaging group dealing power electronics systems and appli-
(IWIPP’00), Jul. 14–15, 2000, pp. 79–83. cations. His field of interest covers thermal management, semiconductors and
[15] F. Auerbach and A. Lenniger, “Power-cycling-stability of IGBT-mod- power integration, converter design for industry and transport applications. He
ules,” in Proc. Industry Applications Conf. (IAS’97), vol. 2, Oct. 5–9, is the owner of several patents in the field of power electronics.
1997, pp. 1248–1252. Dr. Meysenc is a Member of the IEEE Industry Application Society.
[16] R. J. Phillips, Microchannel Heat Sinks Advances in Thermal Modeling
of Electronic Components an Systems. New York: ASME, 1990, vol.
2, pp. 109–184.
[17] R. W. Knight, D. J. Hall, and R. C. Jaeger, “Heat sink optimization with
application to microchannels,” IEEE Trans. Comp., Hybrids, Manufact.
Technol., vol. 15, no. 5, pp. 832–842, May 1992.
[18] C. Gillot, C. Schaeffer, C. Massit, and L. Meysenc, “Double-sided
cooling for high power IGBT modules using flip chip technology,”
IEEE Trans. Comp., Packag. Manufact. Technol. A, vol. 24, no. 4, pp. Mervi Jylhäkallio received the M.S. degree in me-
698–704, Dec. 2001. chanical engineering from the Helsinki University of
[19] V. Gnielinski, “New equations for heat and mass transfer in turbulent Technology (HUT), Espoo, Finland, in 2000.
pipes and channels flow,” Int. Chem. Eng., vol. 16, 1976. From 2000 to 2003, she was a Design Engineer
[20] X. F. Peng and B. X. Wang, “Forced convection and flow boiling heat and a Specialist with ABB. The work was related to
transfer for liquid flowing through microchannels,” Int. J. Heat Mass thermal management and reliability of LV ac drives.
Transfer, vol. 36, no. 14, pp. 3421–3427, 1993. Presently, she is a Scientist in the Corporate Research
[21] H. Muller-Steinhagen and K. Heck, “A simple pressure drop correlation Centre of ABB, Baden-Dättwil, Switzerland, where
for two-phase flow in pipes,” Chem. Eng. Process, no. 20, pp. 297–308, her area of research is thermal management of power
1986. electronics.
[22] D. R. H. Beattie and P. B. Walley, “A simple two-phase frictional pres-
sure drop calculation method,” Int. J. Multiphase Flow, no. 8, pp. 83–87,
1982.
[23] T. N. Tran, M. W. Wambsganss, M. C. Chyu, and D. M. France, “A
correlation for nucleate flow boiling in small channels,” in Proc. Int.
Conf. Compact Heat Exchangers Process Industries, 1997, pp. 353–363.
[24] C. Gillot, L. Meysenc, C. Schaeffer, and A. Bricard, “Integrated single
and two-phase micro heat sinks under IGBT chips,” IEEE Trans.
Compon., Packag. Manufact. Technol. A, vol. 22, no. 3, pp. 384–389, Peter Barbosa (M’95) received the B.S. degree in
Sep. 1999. electrical engineering from the Federal University
[25] M. B. Bowers and I. Mudawar, “Two-phase electronic cooling using of Uberlandia, Brazil, in 1991 the M.S. degree from
mini-channel and micro-channel heat sinks: Part 1—Design criteria and the Federal University of Santa Catarina, Brazil,
heat diffusion constraints,” Trans. ASME, vol. 116, pp. 293–297, 1994. in 1993, and the Ph.D. degree from the Virginia
[26] , “Two-phase electronic cooling using mini-channel and Polytechnic Institute and State University (Virginia
micro-channel heat sinks: Part 2—Flow rate and pressure drop Tech), Blacksburg, in 2002.
constraints,” Trans. ASME, vol. 116, pp. 298–305, 1994. From 1993 to 1995, he was a Research Assistant at
[27] Y. Katto, “Prediction of critical heat flux for annular flow in tubes taking the Institute for Power Electronics, Federal Univer-
into account the critical liquid film thickness concept,” Int. J. Heat Mass sity of Santa Catarina, and from 2001 to 2003 he was
Transf., vol. 27, no. 6, pp. 883–891, 1984. the Technical Director at the Center for Power Elec-
[28] , “A generalized correlation of critical heat flux for the forced con- tronics Systems (CPES), Virginia Tech. Since 2003, he has been with ABB Cor-
vectice boiling in vertical round uniformly heated round tubes,” Int. J. porate Research, Baden-Dättwil, Switzerland, where he develops technology for
Heat Mass Transf., vol. 21, pp. 1527–1542, 1979. medium voltage drives and wind power converters.

You might also like