DEVELOPMENT OF AN EFFICIENT IGBT SIMULATION MODEL

Loïc Michel, Ahmed Chériti and Pierre Sicard

GRÉI - Groupe de Recherche en Électronique IndustrielleDépartement de génie électrique et génie informatique

Université du Québec à Trois-Rivières, C.P. 500, Trois-Rivières, G9A 5H7, CanadaE-mail: {Loic.Michel, Ahmed.Cheriti, Pierre.Sicard}@uqtr.ca

ABSTRACT

We propose, in this paper, the development of an efficientIGBT model especially designed for simulation and the as-sociated free wheeling diode model that can describe the re-covery current. Although, the IGBT model can describe theswitching mechanism with a good precision, a procedure ispresented in order to make this model able to reproduce bet-ter switching losses. Both models have the advantage to beconfigured exclusively from the datasheets and are built aselectrical circuits, which can be realized easily in classicalsimulators.

Keywords - Power electronics, Power system modeling,DC-DC power conversion, Insulated gate bipolar transis-tors, Diodes

1. INTRODUCTION

Modelling the IGBT component is a significant challenge forsimulation in power electronics and many models have al-ready been proposed [1]. The most important issue is thecompromise between the physical representativeness of themodel, defined by its accuracy degree compared to the physi-cal component, and the time needed by the simulator to solvethe model equations [2] [3]. We can make the following ob-servations regarding the classical models that can be found inthe simulators:

• Some models are too simple and are usually representedby an ideal switch, with eventually a resistor to repre-sent the conduction losses. This lack of physical rep-resentativeness allows to have fast simulations but doesnot necessarily give simulation results true to the real-ity. It means that some significant physical phenomenacould be hidden and are not represented by the simplemodel.

• Other models are too complex, affect the simulationspeed and need many parameters, which describe notonly the main characteristics, but also some internalfeatures (at the semiconductor level). Although this

kind of model can describe exactly the real component,it is still difficult to obtain those internal parameters andconsequently, it can be difficult to configure [4] [5].

• The models are generally designed for a particular sim-ulator and some internal functions are created by usinga proper function or sub-model of this simulator. So itmay be difficult to transpose those models to anothersimulator [6].

The purpose of this work is to develop an IGBT modelbased on the Alonso Model [7], which is easy to configure,has a good physical representativeness and can be applied inmany simulators, in order to extend the modeling to most ofIGBT components. Although the original model has satis-factory physical representativeness, our contribution aims toadapt the model in order to represent with a good accuracythe switching losses, which are described by the energetic di-agrams in IGBT datasheets. Moreover, it has been proved in[2] that the Alonso model provides almost the same results asthe Hefner Model [4] for the switching mechanism behavior.Both models could be considered as behaviorally equivalentand that proves the efficiency of the Alonso model. Becausean IGBT is generally built with a free wheeling diode (FWD),we have developed a simple diode model, based on the Batardmodel [8], to reproduce recovery current losses.

Section II of the paper introduces the Alonso model andthe associated FWD model. Section III presents the specifi-cations of the models and a procedure that we developed todetermine the model parameters. Section IV presents simu-lation results of a switching cell and section V concludes thepaper.

2. IGBT MODEL STRUCTURE

2.1. Alonso Model

The Alonso model has the particularity to be defined by anelectrical circuit composed exclusively by basic componentsi.e. by components we can easily find in a simulator. Figure1 presents the circuit associated to the IGBT model.

978-1-4244-3508-1/09/$25.00 ©2009 IEEE 252

Fig. 1. Original Alonso IGBT model.

According to the IGBT concept, we can separate in thismodel, the MOSFET transistor part and the BJT transistorpart. Both are linked and three capacitors Cge, Cgc and Cce,whose values depend on the Vce voltage, represent the interelectrodes capacitors. Although all these capacitors are nonlinear, extensive simulation and experimentation has shownthat only one non linear capacitor Cgc is necessary to rep-resent the whole capacitors non linearities. Because Cgc isdefined non linear, it admits a circuit to create the non linear-ity.

All the parameters are well defined and a complete pro-cess allows to calculate them by considering the output char-acteristic Ic − Vce diagram given in datasheets. In order tocalibrate the switching losses, the globalCge capacitor will beadjusted in order to adapt the energy losses vs. collector cur-rent (E − Ic) diagram, which is also given by the datasheets.

2.2. Diode Model

Since IGBTs usually have an internal FWD, the IGBT modelis incomplete without designing a diode model that can takeinto account the recovery current of the diode. The recoverycurrent takes an active role in the global switching process(including the diode) and is generally included in the evalua-tion of IGBT turn-on losses. We have developed a completeelectrical circuit, like the Alonso Model and the Batard diodemodel, that enables to reproduce the recovery current (Fig. 2).

Define VA and VK respectively by the anode potential andcathode potential. The model is composed of two parts. Thefirst part is the classical direct diode model and the secondpart represents the recovery current equivalent source. Wedescribe the direct diode model by the dynamic resistor rdin series with a threshold voltage vd. To realize the recoverycurrent source, a RC derivative circuit has been employedin order to detect the time when the diode switches. Duringthe time the diode turns on or turns off, the decreasing or in-creasing VAK voltage is converted into a voltage peak whosesign depends on the VAK slope sign. The circuit composed ofAd, Rdelay and Cdelay constitutes the derivative circuit (thevoltage controlled source Ad is used to adapt the impedance

Fig. 2. Diode model used to reproduce the recovery current.

between the direct diode circuit and the derivative circuit). Fi-nally, a second voltage controlled Ar source uses the deriva-tive voltage VRdelay

in order to charge theCr capacitor, whichrealizes the recovery current source. The D2 - Rshunt cir-cuit eliminates the negative peak, which is created during thediode turn-on. Notice that the rr resistor determines the leak-age current of the diode. As a result, the simulation presentedin Fig.3 shows the behavior of our diode model : the diode isincluded in a buck converter whose input voltage and outputcurrent are respectively equal to 100 V and 300 A.

Fig. 3. Recovery current IAK of our diode model when thediode turns off.

We will now consider our diode model included in theAlonso Model and describe the configuration of the globalmodel in order to reproduce the switching losses described inthe datasheets (including the FWD).

3. SPECIFICATION OF THE IGBT

In both models, although all parameters are easily determinedfrom the datasheets, the goal of the specification process is toadapt the switching losses IGBT diagram vs. collector cur-rent of the IGBT model with the diagram provided by the

253

datasheets. To perform simulations, the SEMIKRON 603 GB[9], will be used as an example and all parameters of theAlonso model will be extracted from the datasheets. Thespecification process is an optimization procedure, commonto the diode and the IGBT, whose purpose is to adapt cer-tain components values of each model to verify, with the bestprecision, the switching losses vs. current diagram of thereal components. We have to consider the optimization ofthe diode at first because it is important that the diode hasthe optimized recovery losses before considering the globalswitching mechanism.

3.1. Notations

Define the following energies, all functions of Ic :

• Eon and Eon , which are respectively the datasheetsIGBT (and FWD) turn-on losses and Alonso model turn-on losses;

• Eoff and Eoff , which are respectively datasheets IGBT(and FWD) turn-off losses and Alonso model turn-offlosses;

• E and E, which are respectively datasheets IGBT (andFWD) global switching losses and Alonso model globalswitching losses;

• Ew and Ew, which are respectively datasheets FWDrecovery current losses and our diode model recoverycurrent losses.

3.2. Optimization of the Diode Model

Adjusting the recovery losses is equivalent to tuning the peakvalue of the recovery current and its extinction delay. Accord-ing to our model, these two parameters are defined respec-tively by the Ar and Cr values. The following algorithm hasbeen used to determine the optimal (An

r , Cnr ), which better

approach the Ew − Ic diagram.

1. Define an interval for each parameters such that {Ar} =[Armin, Armax] and {Cr} = [Crmin, Crmax].

2. Define an interval of Ic, spanned by theE−Ic diagram,such that {Ic} = [Icmin, Icmax].

3. Divide the intervals in order to have k + 1 values suchthat a value is identified by the subscript i . Conse-quently, a value Ai

i ∈ {Ar}, Cir ∈ {Cr} or Ii

c ∈ {Ic}can be written as:

Air = Armin +

i

k(Armax −Armin)

Cir = Crmin +

i

k(Crmax − Crmin)

Iic = Icmin +

i

k(Icmax − Icmin)

(1)

For each ith value of Ic, we have the correspondingrecovery energy losses Ei

w and Eiw and therefore the

complete {Ew} − {Ic} and {Ew} − {Ic} diagrams.

4. For each (Air, C

ir) with 0 < i < k, simulate the model

for each Iic. The quadratic error:

σ =k∑

i=0

({Ew} − {Ew})2 (2)

is then calculated. There exists optimal parameters (Aor,

Cor ) such that σo = min(Ew − Ew)2.

Restart the 1-3 optimization steps by reducing the originalintervals; a dichotomic iteration is used in order to obtain theconvergence. As a result, having performed n iterations, eachoriginal interval is considerably reduced and can be approxi-mated by an unique value (Ao

r, Cor ). These values verify that

the error σ reaches a minimum.

3.3. Optimization of the Alonso Model

In the IGBT model, we have to distinguish the static part,whose purpose is to represent both the output Ic − Vce di-agram and the transfert Ic − Vge diagram, and the dynamicpart, whose purpose is to represent the dynamic behavior, likethe switching mechanism. The function of the capacitors isprecisely to describe how the operation point moves. Thisinvolves that the static part must be configured before the dy-namic part.

To configure the static part, only the Ic−Vce and the Ic−Vge diagrams provided by the datasheets are necessary. Onthe Ic − Vce diagram, two curves for two different Vge areneeded and these curves must show the saturation area for alarge Vge. According to this diagram, there exist two cases :

• The IGBT datasheets contain all the data needed to per-form the complete configuration of the model.

• The IGBT datasheets contain only a part of the dataneeded to perform the complete configuration of themodel. In this case, we have to rebuild the completeIc − Vce diagram.

This last case corresponds to the SEMIKRON 603 GB becauseonly one curve is complete and shows a part of the saturationarea [9]. Figure 4 presents the Ic − Vce diagram of this IGBTwhere only the curve for Vge = 11 V is complete. It wasintroduced in [5] that for a constant Vge, the Ic−Vce diagramsatisfies :

V L

m(Vce) = Vsat −Vsat ln

{1 + exp

[Ats

(1− Vce

V sat

)]}ln(1 + exp(Ats))

Ic = AtVLm(1 + λ(Vce − V L

m))(

1− V Lm

2Vsat

)(3)

254

Fig. 4. Ic − Vce diagram for the SEMIKRON 603 GB [9].

where At, Ats, λ and Vsat are real parameters. To determinethese parameters according to the datasheets, a non linear op-timization process can be employed. Figures 5 and 6 presentsthe simulated Ic − Vce diagram that extends the Ic − Vce dia-gram provided by the datasheets.

Fig. 5. Simulated Ic − Vce diagram for Vce < 25 V.

Fig. 6. Simulated Ic − Vce diagram for Vce < 4 V.

Although the Alonso model is very complete, some ap-proximations have been made during the creation of the model.One of these approximations consists on supposing that onlyone capacitor is non linear and the choice of Cgc is justifiedbecause Cgc presents the most variations relating to Vce. Fig-ure 7 shows a modification of the original model, which in-cludes an additional capacitor called C∗

ge whose purpose is tobe configured and optimized to adjust switching losses relat-ing to the E − Ic diagram. To configure and optimize thedynamic part, the same approach as for our diode model willbe used (Section 3.2).

Fig. 7. Modification of the original model to include an addi-tional configurable capacitor.

Define an interval {C∗ge} = [C∗

gemin, C∗gemax] with its k

divisions:

C∗ge

i = C∗gemin +

i

k(C∗

gemax − C∗gemin), 0 ≤ i ≤ k (4)

The algorithm provides an optimized valueC∗ge

o such thatthe error (E− E)2 is minimum. Figure 8 presents, for severaltime step, the simulated E− Ic and Ew − Ic diagrams for theSEMIKRON 603 GB compared to the datasheet.

Fig. 8. Comparison between the E− Ic and E− Ic diagrams.

We observe that for a smaller time step, the switchingmechanism is better described and is evaluated with a betteraccuracy. For a large large time step, the switching mecha-nism is considered as ideal.

4. SIMULATION RESULTS

Simulation of the switching mechanism allows to verify if themodified Alonso model is correctly configured. Figures 9, 10and 11 present the behavior of the modified Alonso modelduring one switching period of a simple switching cell.

We can see that both turn-on and turn-off reproduce withhigh accuracy the behavior of a real IGBT during the tran-sient time. The most important point is that the power dissi-pated during the switching process is significant and provesthe efficiency of the modified Alonso model with our diode

255

Fig. 9. Ic(t) for an IGBT in a switching cell.

Fig. 10. Vce(t) for an IGBT in a switching cell.

Fig. 11. VceIc(t) for an IGBT in a switching cell.

model. Simulations have been realized using PSIM R© in co-simulation with MATLAB R©.

5. CONCLUSION

The structure of the Alonso model allows an implementationin any simulation platform with a good level of representative-ness. As a result, this model, including the FWD, constituesa good starting point to investigate real phenomena in powerconverters without performing heavy simulations. Moreover,some optimizations have been applied to this model in orderto have a better description of the switching losses.

6. REFERENCES

[1] K. Sheng and B.W. Williams, “A review of IGBT mod-els“, IEEE Trans. Power Electronics, vol. 15, pp. 1250-1266, Nov. 2000.

[2] A. Agbossou, I. Rasoanarivo, and B. Davat, “Compar-ative study of high power IGBT model behavior in volt-age source inverter“, Power Electronics Specialists Con-ference, 1996. PESC ’96 Record., 27th Annual IEEE, vol.1, pp. 56-61, 23-27 Jun 1996.

[3] A. N. Githiari, B. M. Gordon, R. A. McMahon, Z.-M Liand P. A. Mawby, “A comparison of IGBT models for usein circuit design“, IEEE Trans. Power Electron., vol. 14,pp. 607-615, July 1999.

[4] A. R. Hefner, “Analytical modeling of device-circuit in-teractions for the power insulated gate bipolar transis-tor (IGBT)“, in Conf. Rec. IEEE Industry ApplicationsSoc. Meet., pp. 606-614, 1988; also IEEE Trans. IndustryAppl., vol. 26, pp. 995-1005, Nov. 1990.

[5] S. Pittet, “Modélisation physique d’un transistor de puis-sance IGBT - traînée en tension à l’enclenchement“(infrench), Ph.D. Dissertation, EPFL - Lausanne (Switzer-land), 2005.

[6] R. Kraus, P.Turkes, J. Sigg, “Physic-based modelsof Power semiconductor devices for circuit simulatorSPICE“, Power Electronics Specialists Conference, 1998.PESC 98 Record. 29th Annual IEEE, vol. 2, pp. 1726-1731, May 1998.

[7] C. Alonso, “Device modelling for simulation in powerelectronics“ (in french), Ph.D. Dissertation, INPT -Toulouse (France), Dec. 1994.

[8] C. Alonso, T.A. Meynard, H. Foch, C. Batard and H. Pi-quet, “A model of GTO compatible with power circuit sim-ulation“, Fifth European Conference on Power Electronicsand Applications, vol. 2, pp. 232-237, Sept. 1993.

[9] SEMIKRON c© SEMIX R© 603 GB, “07-11-2006 CHD“.

256