Coupled electro-thermal simulation of a DC DC converter

Coupled electro-thermal simulation of a DC/DC converterMiquel Vellvehi *,Xavier Jorda`,Philippe Godignon,Carles Ferrer,Jose ´Milla ´n Centre Nacional de Microelectro`nica (CNM-CSIC),Institut de Microelectro `nica de Barcelona,Systems Integration Department,Campus U.A.B.Bellaterra,08193Cerdanyola del Valle`s,Barcelona,Spain Received 3April 2006;received in revised form 24October 2006Available online 19December 2006AbstractElectro-thermal simulation of a DC/DC converter using a two-step methodology and its experimental validation are presented in this paper.The simulation technique is based on the coupling of a thermal simulator (FLOTHERM)and an electrical one based on VHDL-AMS language.This work is mainly focused on the description of the thermal component with special emphasis in the modelling of the active power devices (MOSFETs and Schottky diodes)included in the converter.Once these devices have been electro-thermally mod-elled,they are implemented in the DC/DC converter including a 10layer PCB and some passive components.The modelling results have been experimentally checked by means of infrared measurements.Ó2006Elsevier Ltd.All rights reserved.1.IntroductionThe performance and characteristics of semiconductor devices in electronic packages can be considerably affected by temperature variations.For this reason,accurate circuit simulation requires to take into account the static and dynamic effects induced by the heat dissipation.In this way,new modelling methodologies are being applied in complex power systems to study these effects [1–3].The modelling using electro-thermal interactions in inte-grated circuits has been addressed in a variety of ways.The existing methods can be broadly classified into two groups:direct and relaxation methods (Fig.1)[4].In both cases,standard electrical models became electro-thermal models,by including an additional thermal electrode in each model.In the direct or fully coupled method a single electrical simulation tool is used to model both electrical and ther-mal behaviour [5].The coupling is possible through the exchange between both models of an electrical parameter (instantaneous dissipated power)and a thermal variable (instantaneous temperature).The problem of this method is the thermal modelling due to the complex structure ofthe overall system.Often finite-difference approaches,Fou-rier series,analytical solutions and thermal networks are employed.Most of them are valid only in the steady-state cases for linear heat equations.Different contributions have been previously reported using this representation together with 1-D heat transfer equations,where voltages are equivalents to temperatures and currents to thermal power.As an example,we can mention the electro-thermal model of a commercial MOSFET from IR (IRPF254)with SPICE software [6].In the relaxation method,electrical and thermal equa-tions are solved separately exchanging periodically temper-ature and power parameters until thermal and electrical convergence is reached [7].The trade-offof this technique is clear:a more accurate representation of the thermal behaviour of the system is available,at the expense of an increase of the computation time.Modelling tasks requires a good knowledge of the geom-etry and physical properties of the different materials involved in the system.In addition,depending on the sys-tem to be modelled,the mesh generation can be complex and time consuming.In our case,the modelling of the DC/DC converter requires more than two million mesh points.Although these tasks are laborious but not concep-tually complex,the problems can appear in data transfer0026-2714/$-see front matter Ó2006Elsevier Ltd.All rights reserved.doi:10.1016/j.microrel.2006.10.009*Corresponding author.Tel.:+34935947700;fax:+34935801496.E-mail address:miquel.vellvehi@cnm.es (M.Vellvehi)./locate/microrelAvailable online at Microelectronics Reliability 47(2007)2114–2121between both simulators(synchronization,convergence and mathematical approximation).This paper shows how the methodology based on the relaxation method can be applied to analyse relatively com-plex industrial systems,such as a DC/DC converter,in order to predict their thermal behaviour.The whole approach has been experimentally validated by means of an exhaustive thermographic analysis method.In afirst stage,we have used and demonstrated the simulation method on single power devices(MOSFETs and Schottky diodes)used in the converter.The methodology used in this work willfinally allow predicting the chip temperature of the power devices used in the DC/DC converter,hence improving the system design and increasing the reliability level of the whole system.Other passive components such as transformers,self-inductances and capacitors have been also included in thefinal model of the converter.2.Modelling methodologyOne of the main objectives of our work is to build a model for a DC/DC converter usable for electro-thermal simulation.The block diagram of the relaxation method (usually called two-step method)is shown in Fig.2.The electrical model is defined taking into account the tempera-ture dependence of the more relevant parameters,extracted from the characterisation of the components at different operating temperatures.On the other hand,the thermal model is based on the physical and geometrical description of the packages of the components,mainly extracted from the datasheets.The coupling between both simulators is achieved in the following way:the results obtained from thefirst iteration of the electrical simulator at a given initial temperature are exported in a textfile containing the value of the instantaneous power dissipated in the relevant com-ponents(main heat sources).Then,these power values are used as inputs in the thermal simulator.Next,new values of the components temperature are obtained,which are also exported in a textfile and used as a new input for the elec-trical simulator.New values of dissipated power are obtained,being the new inputs for the thermal simulator. This process is iterative and once the convergence criterion is reached(i.e.any significant evolution of powers and temperatures is observed),we get thefinal values of instan-taneous dissipated power and temperature in each com-ponent.The used convergence criterion has been that the components’temperature does not change more than 0.05°C from a coupling loop to the next one.2.1.The electrical simulatorThe electrical simulations have been performed with VHDL-AMS(Analogue Mixed Signal)modelling language using ADVanceMS from Mentor Graphics[8].To represent the power devices(MOSFETS and Schottky diodes),we have used SPICE models described with VHDL-AMS,as Fig.1.Example of both main electro-thermal coupling methods used for power system modelling.M.Vellvehi et al./Microelectronics Reliability47(2007)2114–21212115well as the passive components(transformer,self-induc-tance,capacitors...).A detailed explanation of the electri-cal modelling of the DC/DC converter has been reported in[9].The advantage of using this strategy of modelling is that it is very convenient to extract the model parameters from the characterisation measurements at different tem-peratures.However,some limitations arise:the models are adjusted in specific conditions of measurements and they can only be used in these conditions.In our case,the models have been developed to be compatible with real operating conditions of the considered converters(input voltage of 48V,an output voltage of5V,an output current of3A,a commutation frequency of500kHz and a duty cycle of 40%).For example,power MOSFETs are modelled only in their lineal region,which is their normal working condi-tion in the DC/DC converter.This modelling approach will be detailed for the power devices in Section3.2.2.The thermal simulatorThe thermal models developed for the MOSFET,Scho-ttky diodes and the complete PCB of the DC/DC converter have been implemented with the FLOTHERM TM package from FLOMERICS[10].This software uses Computa-tional Fluid Dynamics(CFD)technique and is addressed to solve thermal problems in electronic equipments.It pre-dicts airflow and heat transfer in and around electronic equipments,including the coupled effects of conduction, convection and radiation.In particular,we have used this software to model the thermal behaviour of semiconductor chips and their packages,as a series of layers of materials characterised by their thermal conductivities and specific heats.These layers must be properly connected to describe the correct thermal path from chip to environment.The main heat transfer mechanism from the different devices to the PCB is the conduction.The convection to the air is considered around the whole DC/DC converter and the radiation phenomena can be neglected because the maximum temperatures reached in the system are low enough.Geometrical data,material properties and bound-ary conditions of the full system have to be well known for an efficient modelling.3.Power devices modellingAs afirst step,the two-step electro-thermal modelling method using VHDL-AMS and FLOTHERM,has been applied to the discrete power devices of the commercial DC/DC converter from ASCOM(SMC CX15C48S5). The active components integrated in this converter are basically of two types:power MOSFETs(BUK482-100A) and Schottky diodes(EC21QS04).3.1.MOSFET modelThe power MOSFET has been electrically modelled with a SPICE Level3model,using experimentally extracted parameters of the BUK482-100A.Electrical characterizations at several working temperatures and gate voltages have been performed.Fig.3shows the I(V)char-acteristics of the MOSFET at different ambient tempera-tures(from25°C to150°C)and an applied gate voltage of7V.As the MOSFET will work as a switch in its linear region,the model considers only the parameters related with this working zone(on-state resistance,threshold volt-age,etc.).Apart from the static characterisation,the dynamic electrical behaviour of the MOSFET is described by its parasitic capacitances.As it has been demonstrated [6],parasitic MOSFET capacitances are practically inde-pendent of temperature,being the main temperature dependant parameters the on-state resistance(R ON)and the threshold voltage(V TH)as it is shown in Fig.4.Electro-thermal iterations have been performed between the electrical model described above with the aid of VHDL-AMS and the thermal model of the MOSFET packaged in a SOT-223,developed with FLOTHERM (Fig.5).As regards the boundary conditions,the MOSFET has been mounted on a Cu layer with afixedtemperature Fig.3.VDMOS forward I–V characteristics at a gate voltage of7V extracted at different temperatures.2116M.Vellvehi et al./Microelectronics Reliability47(2007)2114–2121of 25°C.As explained in the previous section,the main heat transfer mechanism is the conduction from the chip to the Cu layer,and the radiation and convection to the air can be neglected.Under these conditions,we obtain an equilibrium junc-tion temperature under different input powers,considering a minimum number of iterations.As an example,the results obtained for the different iterations with V gs =5V and I ds =1.25A biasing,are shown in Table 1.As we can see from this table,the convergence criterion is reached after four iterations between both simulators and we obtain a final chip temperature of 31.7°C and a dissipated power of 0.5105W.From the thermal simulation results,the ther-mal resistance of the device can be deduced as the relation-ship between the silicon temperature increase and the dissipated power.The value obtained in the simulation is 13.1°C/W being the maximum value obtained from the manufacturer’s data sheet of 15°C/W.Although these results represent a first indirect validation of our models,we have performed a direct experimental validation.In this sense,the power VDMOS BUK482-100A has been mounted on an IMS (Insulated Metal Substrate)applying a uniform temperature on the IMS backside.The measurement conditions are:an applied power of 0.51W,an input gate voltage of 5V and an output current of 1.2A.When the backside temperature is fixed at 25°C,an equilibrium temperature of 28.6°C is measured on the MOSFET top surface.This value is in good agreement with the simulated value (28.8°C),hence validating our modelling approach.3.2.Schottky diode modelAn equivalent model of a Schottky Diode (SPICE Level 1)has been developed,adjusting the simulation parametersof the EC21QS04Schottky diodes included in the system.As for the MOSFET,an electro-thermal characterization has been performed to extract the temperature dependence of the parameters of the diode (forward voltage drop,leak-age current and junction capacitance).Then,the electrical model has been developed considering the temperature dependence of the electrical parameters.In parallel,the thermal model of the Schottky diode using FLOTHERM TM (Fig.6)has also been developed and checked,taking into account the DO-214AC package structure of the EC21QS04.4.DC/DC converter modellingThe modelled DC/DC converter is mainly composed of an input filter (two capacitances),power switches stage (including two MOSFETs in parallel),transformer,output rectifier Schottky diodes and filters (the inductance and the capacitance)and finally,isolation stage and PWM Control (Fig.7).The power consumption of the latter is negligible compared to other stages and will not be taken into account in the modelling.All these elements are mounted on both faces of a 10-layer PCB board.The electrical circuit of the converter modelled with ADVanceMS is depicted in Fig.8[9].Only the components dissipating significant power have been considered.The electrical characteristics of the converter are:an input volt-age of 48V,an output voltage of 5V,an output current of 3A and a commutation frequency of 500kHz.A duty cycle of 40%has been chosen taking into account the optimum operating ing these parameters as input val-ues,we have calculated the dissipated power mean values of the different components (Table 2).Complete thermal modelling of the DC/DC converter has been developed using FLOTHERM TM ,describing the complete PCB.The geometry and components location have been obtained from the manufacturer’s datasheet.The active components (MOSFET and Schottky diodes)thermal models previously developed have been included in the full converter system.The 10-layer board PCBisFig.5.FLOTHERM model of the BUK482-100A MOSFET packaged in a SOT-223.Table 1Simulation results after successive iterations of the MOSFET coupled electro-thermal modelling V gs (V)I d (A)T amb (°C)T j (°C)Pot (W)#Iteration 51.2525250.492131.450.5098231.690.5105331.70.51054Fig.6.FLOTHERM model of the EC21QS04Schottky diode packaged in a DO-214AC.M.Vellvehi et al./Microelectronics Reliability 47(2007)2114–21212117modelled with an orthotropic thermal conductivity.The power converter works at an output voltage of5V and an output current of3A.That brings us to the consider-ation that the most important devices related to power con-sumption will be the switches,the rectifier diodes and the transformer.It is worth to point out that the passive com-ponents included in the PCB have not been thermally mod-elled.In this sense,such components have been considered as cuboids placed at the right locations but without any functionality.The active and passive components are placed in the PCB as shown in Fig.9:•Face A includes the half part of the transformer and the inductor,two MOSFETs in SOT-223package and fourcapacitors.•Face B includes the second half part of the transformer and the inductor,four Schottky diodes in DO-214AC package,a small transformer,a bipolar diode andfive capacitors.The values of the power dissipated by each component extracted from the electrical simulator(see Table2)are used as input parameters in the thermal simulator.The out-put of the thermal modelling is the internal temperature of each component.The values of the temperature of the active devices obtained by modelling at ambient tempera-ture(20°C)are shown in Table3.Fig.10shows the simu-lated temperature distribution in both faces of the PCB for an output current of3A andfixing20°C as ambient tem-perature.As it can be observed in Table3and in Fig.10, although the two MOSFETs temperatures are similar, some differences are observed between the four diodes, being the coolest the external ones(CR2and CR5).This will be corroborated with the infrared measurements.Table2Calculated dissipated power in the main components of the converter(T amb=20°C)Component Dissipated power(W)Transformer0.834Inductor0.025MOSFET0.205Schottky diodes0.8Fig.9.FLOTHERM DC/DC converter model.(a)Face A and(b)FaceB.2118M.Vellvehi et al./Microelectronics Reliability47(2007)2114–21215.Experimental validationTo validate the simulation results,temperature distribu-tion has been measured using Infrared (IR)thermography with AGEMA Thermovision THV900equipment.This technique allows the monitoring of surface temperatures and thermal patterns while the converter is operating.To obtain the IR measurements on both sides of the PCB,a test board has been developed to obtain the temperature distribution both on top and back of the PCB (Fig.11).Since the various components of the PCB have different emissivity values,a direct measurement of the surface tem-perature could result in unrealistic temperature values.For this reason,the IR measurements methodology includes two complementary techniques for comparison:•Emissivity equalization:Initially,an IR image of the PCB is made without biasing the circuit and considering a fixed external temperature;this gives a reference image for the emissivity equalization (i.e.an emissivity map-ping is obtained).After that,we obtain a second IR image of the system under working conditions.Consid-ering the two images,a surface map of temperature is obtained by software processing (Fig.12).•Black paint:the board is painted with black paint (Nex-tel 811-21Velvet Coating)with a known emissivity (e =0.97).Then,all components are supposed to pres-ent the same value of emissivity.Consequently,a direct temperature distribution can be measured on the IR pic-ture when the converter is biased (Fig.13).Fig.10.Simulated surface temperature distribution of the DC/DC converter at 20°C:Face A (a)and Face B(b).Fig.11.PCB connections for IR thermographymeasurements.Fig.12.Experimental temperature distribution on Face A (a)and Face B (b)of the converter,using an emissivity equalization technique (output current of 3A).Table 3Simulated surface temperatures of the active components of the DC/DC converterAmbient temperature 20°CMOSFET Q274.23°C MOSFET Q374.29°C Diode CR273.94°C Diode CR376.77°C Diode CR475.66°C Diode CR574.46°CM.Vellvehi et al./Microelectronics Reliability 47(2007)2114–21212119As mentioned before,the converter has been biased under the same conditions used for the simulation.The experimental results obtained for 1A and 3A output cur-rents can be seen in Tables 4and 5,respectively.These tables show the values of the surface temperatures of the different active components in both faces (A and B).As can be observed in both tables,the two methods used to measure the temperature distribution give similar results,with maximum differences of 2°C.The most important dif-ferences are found for the 3A output current case and for Face B components.If we compare the results of the measurements with the simulated ones (see Table 5)for the 3A output current (T amb =20°C),we observe a difference on the temperature distribution over the transformer.The reason is that pas-sive components have not been thermally modelled in a first approximation as explained previously.However,regarding the MOSFETs (Q2and Q3)the temperatures measured in both devices are very similar using both mea-surement techniques and almost equal to the temperatures calculated by simulation.In the case of the Schottky diodes,they present slight differences between experimental and simulated tempera-tures.For the external diodes (CR2and CR5)the experi-mental temperature values are similar to those obtained by simulation.However,for the internal ones (CR3and CR4),which are the hottest,the temperatures differences can reach 5°C,corresponding to an error of 6%.Then,we can conclude that,although the slight discrepancy on the values of temperatures in two of the active devices,the results obtained from the simulations agree with the measured results,corroborating the efficient electro-ther-mal modelling of the active devices.Then,using the pro-posed simulation methodology,we can also predict the junction temperature of the active devices included in the converter,hence improving its design in order to decrease the maximum temperature and increasing its reliability.In our case,under the operating conditions described above,the obtained chip temperatures of the different power devices are around 1°C above the temperatures measured in the surface of its packages.6.ConclusionsIn this work,the electro-thermal model of a power sys-tem using a modelling methodology based on relaxation method is presented and experimentally validated.The demonstrator is a DC/DC commercial converter and its modelling presents two main characteristics:the first one is the inclusion of power devices and the second one is the presence of a large number of active and passive com-ponents mounted in a 10-layer PCB.Electrical modelling performed with VHDL-AMS (using ADVanceMS)is cou-pled with thermal simulations using FLOTHERM TM .This coupled electro-thermal modelling was applied to single power devices (MOSFET and Schottky diode)to validate the methodology,before considering the complete circuit simulation.Special emphasis has been done on the thermal simulations and their experimental validation by means of an exhaustive infrared thermography analysis.The results obtained from the measurements show a good agreement with those obtained by simulation,hence showing the efficient electro-thermal modelling of the active devices.In addition,using this simulation methodology,we can predict the junction temperature of the devices.In thisTable 4Experimental surface temperatures of the active components obtained for an output current of 1A Op @1A Emissivity equalisation (°C)Black paint (°C)FACE A T Q237.137.3FACE A T Q337.137.3FACE B T CR236.835.2FACE B T CR338.836.9FACE B T CR438.337.0FACEB T CR536.534.6Table 5Experimental surface temperatures of the active components obtained for an output current of 3A Op @3A Emissivity equalisation Black paint FACE A T Q274.2°C 75.4°C FACE A T Q374°C 74.4°C FACE B T CR274.073.0FACE B T CR381.078.7°C FACE B T CR480.878.0°C FACEB T CR574.369.4°CFig.13.Experimental temperature distribution on Face A (a)and Face B (b)of the converter,using a black paint of known emissivity (output current of 3A).2120M.Vellvehi et al./Microelectronics Reliability 47(2007)2114–2121way,we can improve the system design in order to decrease the maximum junction temperature and to increase the reli-ability of the whole system.AcknowledgementsThe work presented in this paper was supported by EU under SPARTE Euclid contract and by the Spanish Ministerio de Educacio´n y Ciencia under contract TEC2005-087392(SPACESIC Project).References[1]Gutierrez H,Christoffersen CE,Steer MB.An integrated environ-ment for the simulation of electrical,thermal and electromagnetic interactions in high-performance integrated circuits.In:Proceedings of the IEEE sixth topical meeting on electrical performance of electronic packaging,September1999.[2]Kojima T,Yamada Y,Ciappa M,Chiavarini M,Fichtner W.A novelelectro-thermal simulation approach to power IGBT modules for automotive traction applications.R&D Review of Toyota CRDL 2004;39(4).[3]Miao L,Runde Z,Yuanquing G.A New electro-thermal simulatorbased on relaxation method for integrated circuits with distributed temperatures.In:Proceedings of international conference on chip design automation proceedings ICDA,2000.[4]Wu¨nsche S,ClaußC,Schwarz P.Electro-thermal circuit simulationusing simulator coupling.IEEE Trans Very Large Scale Integrat VLSI Syst1997;5(3):277–82.[5]Hefner AR,Blackburn DL.Thermal component models for electrothermal network simulation.In:ninth IEEE semiconductor thermal symposium1993.p.88.[6]Jorda`X,et al.Electro-thermal VDMOS SPICE model for thethermal design of power converter.In:IEEE international sympo-sium on diagnostics for electrical machines power electronics and drives(SDEMPED’99),Gijo´n(Spain),Septiembre1999.[7]Petegem WV,Geeraets B,Sansen W.Electrothermal simulation anddesign of electrical circuits.IEEE J Solid-State Circuits1994;29(February):143–6.[8]Heinkel Ulrich et al.The VHDL reference,a practical guide tocomputer-aided integrated circuit design including VHDL-AMS.John Wiley&Sons,LTD.;2002.p.197–232.ISBN047189972. [9]Martin E,Ferrer C.DC/DC Converter Modelling with VHDL-AMS.In:Proceedings of IEEE ICIT(International Conference on Indus-trial Tecnology)2003–Maribor,Slovenia,p.766–71.[10]FLOTHERM TM User’s Guide.Flomerics Inc.www.fl.M.Vellvehi et al./Microelectronics Reliability47(2007)2114–21212121。