A new Validated Physically Based IGCT Model for Circuit Simulation of Snubberless and Series Operation

H. Kuhn and D. Schroder

Institute of Electrical Drives Technical University of Munich Arcisstr. 21 80333 Munich, Germany
AbsIracl-A new physically based model which is implemented in Saber MASTS and usable for circuit simulation in order to calculate the static and dynamic behaviour of IGCT devices correctly is presented in this paper. The model is verified by comparing simulation with experimental results of a 4.SkVl3kA-IGCT in hard switch snubherless operation. Furthermore simulation results of two series connected IGCT switches are analyzed especially concerning the prDblem oPnon-symmetrically distributed blocking voltages between the switches if snubbers are omitted due to inevitably existent differences concerning the exact device properties and the gate drives respectively. An idea for dimensioningthe snnhher capacity is given.

I. INTRODUCTION The GCT (Gate Commutated Thyristor) is a special structured GTO including a n-doped buffer layer and a transparent emitter. The buffer layer enables a trapezoidal distribution of the electric field in the case of high blocking voltages. Thus the low doped drift-zone (n--Zone) can be designed shorter compared to the n--Zone of a standard GTO without such a buffer layer concerning a certain blocking voltage with the positive effect of smaller on state voltages. The very thin (only few pm) p"-anode, the transparent emitter, causes a noticeable reduction of the injection efficiency of holes into the n-Zone [6]. Thus the GCT is ideal for application of the hard drive concept even without a RCD-snubber. Hard drive means that the device is turned off with dI,,t,/dt 2 1000 A/ps gate current rise. As a consequence the anode current commutates quickly and completely from the cathode (C) to the gate electrode (G), i.e. the cathode current turns equal to zero and the GC-junction is blocked before a depletion layer can arise. Because of the fact that the device destruction mechanism caused by hot spots due to current filamentation cannot occur without cathode current during the expansion of the depletion zone, a hard driven GCT withstands higher loss power densities than a conventional (soft) driven GTO. After the current commutation described above, the p+nn-ptransistor (anode-buffern--Zone-p-base) determines the dynamic behaviour. Hence the GCT is well suited for one dimensional physical modeling. To reach the aim of a hard drive the GCT is produced with an integrated low inductive gate drive (Lgate F;! 5 nH). Then the GCT is called IGCT (Integrated GCT). The advantages of the IGCT compared to the HV-IGBT prevail for applications in the range of 1 MVA and more [7], [IO]. The paper will present a new ID-Buffer-GTO model based on semiconductor physics which is suitable for correct calculation of the static behaviour and the fast transients of hard

driven IGCT devices without RCD-snubber. Furthermore the new model is valid in the temperature range 100 - 400K due to the application of a new mobility theory [ 9 ] . When using conventional GTO devices the capacity of the obligatory RCDsnubber determines the blocking voltage rise dV/dt during turn off process. In contrast to that the transient voltage rise depends on device internal physics only if the snubber will be omitted. In this case the anode current remains high whereas the blocking voltage rises so that the dynamic avalanche effect may influence the transients. Thus it is important to consider device internal physics especially regarding the expansion of the depletion zone as correct as possible . In section I1 both the modeling fundamentals in general and the specific modeling principles of the IGCT model are discussed. Furthermore it is discussed, why a new mobility description including electron-hole scattering (EHS) due to the theory of Mnatsakanov [8] is advantageous to use especially regarding the static behaviour. In section I11 the received simulation results using this model are compared with measured transients and the current-voltage characteristics (CVC). Preliminary investigations regarding series connection of IGCT devices are presented in section IV. The conclusion in section V completes the paper and indicates current and future research activities.
OF 11. MODELING THE IGCT

A. Basic Modeling

The GTO model of Metzner and Vogler [3] was used as a starting point for the GCT modeling. Especially the very effective numerical approach solving the continuity equation (diffusion equation) in the low doped drift zone is taken from [3]. The boundaries of the high injection region move due to the expansion of the depletion zone so that only few sampling points are necessary for a correct calculation of the spatial distribution of the excess charge carriers for each time step. For this drift zone module the assumption of high injection condition p M n is an adequate approach so that apart from the drift zone width tuy. the high injection lifetime TH is an important parameter describing the distribution of the injected carriers. The saturation current densities j,, and j,, are the physical parameters describing the injection behaviour of the anode and the cathode [3]. Furthermore the previous GTO model includes expressions describing the temperature dependent behaviour [4]. Nevertheless parts of this description especially the formulation of the

0-7803-6401-S/00/$10.000 2000 IEEE

2866

carrier-carrier scattering effect has to be substituted by a new 1D. one (comp. 1 - )

B. Modeling of the buffer layer and the dynamic behaviour of the electricfield
In [3] the buffer layer was modeled by a quasistatic formulation, or in other words the buffer layer excess charge was considered as one lumped charge which has to be diminished by recombination with the buffer lifetime qbuf during turn off process. Apart from the inherent inaccuracies of the lumped charge approach, the fact of appearing convergence problems especially regarding the turn on process due to the difficult handling of the the boundary conditions at the buffedn--zonejunction (lumped charge vs. numerical approach) seems to be the main disadvantage of this approximation. Therefore the buffer layer is integrated into the drift zone module in the new model. In addition to that the possibility to enter lifetimeprofiles is implemented which is necessary to consider the reduced buffer lifetime within the common drift zone module. Of course other regions with reduced lifetime (e.g. as a result of applying electron beams) within the original n--Zone may be regarded in this way. As mentioned above it is very important to describe the dynamical behaviour of the electric field in the depletion zone during the turn off process as accurate as possible. The slope of the electric field function E ( x ) (triangle) is directly proportional to the effective drift zone doping N , , f f which may exceed noticeably the physical doping N D , during dynamic turn off processes. N , , f f can he expressed as

vergence problems may also occur. Therefore the new IGCT model contains a simple avalanche description which offers sufficient accuracy without notable enlargement of the simulation effort. The hole current density jpv multiplied by a is factor MaVal which is calculated by an equation of the form MaVal= f ( E , , j t , E , S ) , with S as a sensitivity parameter. Both and S are model parameters and can be modified within the netlist. Since the avalanche multiplication generates pairs of electrons and holes they compensate each other concerning their contribution to N v e f f in (l), i.e. the avalanche current does not increase the slope of the electric field and thus no additional amplification but self-limitation of the avalanche effect takes place. Therefore the device is not destroyed unless the thermal stress as a result of the increased switching losses is too high.

D. Modeling of the mobilities

The carrier transport in semiconductors assuming 1-Dgeometry is determined by the commonly accepted current density equations j , for electrons n and j , for holes p

with the total mobilities p,,,, the elementary charge qo and the diffusion coefficients D,,,. The ambipolar diffusion equation, valid for the high injection approximation (n M p ) applied at conventional bipolar power device models (e.g. [3]) for describing the charge carrier profile (CCP) in the low doped drift zone can be derived from (2) with the total current density jtot = j , j , inserted into the balance equation for holes

with the basic doping ND", the hole and electron current densiat ties jp,nr the n-p-junction (drift zone/p-base-junction), the unity charge qo and the drift velocities for holes and electrons up,,. In general j,, M 0 during expansion of the depletion zone except the dynamic avalanche effect influences the turn off process. If E > 1 0 4 V / c m ,up,, are constant and called saturation Below E = 104V/cm, U , depend on ,, velocities up)psat,nsat. E . Thus the expression to determine the electrical field is an implicit equation. Sufficient results may also be obtained if substituting up," with upsat,nsat in (1) and applying a correction factor instead of the accurate implicit description. Then a triangle approximation for calculating E ( x ) can be used. When E ( x ) ,i.e. the depletion zone arrives the n--zone/buffer layer junction the shape of E ( x )distribution changes from a triangle into a trapezoid.

ap

dip
dx

P
TH

at
y ieI din g

(3)

. .

(4)

where D A = ZkTpnpp/qO(pn+pp) the ambipolar diffusion is coefficient and TH = T + T, the high injection lifetime. , The electric field along the drift zone depending on the distribution of the excess charge carriers (the CCP) may also be derived from (2) resulting for high injection condition

C. Modeling of the Avalanche Effect

Since the anode current remains high whereas the electric field rises, the maximum electric field near the n-p-junction may exceed the critical electric field [Link] (comp. (1)) so that the dynamic avalanche will affect the turn off process. If a detailed physical description for the avalanche effect is implemented the computation effort rises immensely and con-

with the mobility ratio b = pn/p,. From ( 5 ) the contribution of the drift zone to the static voltage of a device may be calculated by integration over the drift zone width. Within this description no carrier-carrier scattering effect reducing the overall mobilities for n and p is included. That means merely lattice scattering and impurity scattering are considered, e.g. due to [2]. In the model from [3], [4] a conventional carrier-carrier scattering description due to [ l ] was implemented to obtain correct simulation results at low temperatures. Within this physical description p p depending on the

2867

excess carriers p and the temperature T is defined. This special mobility is combined with the mobility due to lattice and impurity scattering by the so-called Rule of Matthiesen

Since ppis a function of p. the mobility ratio b and the diffusion coefficient DA are also functions of p. Equation (2) combined with (3) yield instead of (4)

If (1 1) is compared to (5) it can be seen that an additional term on the right side containing the EHS-mobility ppn appears in (1 1). Especially the absence of this additional term in (5) is responsible for receiving wrong simulation results which do not correspond to measurement both regarding switching behaviour (proportional to the CCP) and the CVC with the same set of parameters even for 300K. In contrast to that the model including the EHS description yield consistent simulation results. Therefore this new mobility theory was implemented also in the GTO and IGCT model. The effect of applying EHS or not can be seen in section 111-C (Fig. 7). 111. VALIDATION

(7)

with the lifetime

A. Schematic for Validation and Model Adoption Process

1
T

1
TH

+ -1 TA

which includes the Auger Effect. Due to T A lip3 the Auger Effect reduces the lifetime noticeably for high carrier densities
(p 2 1 0 1 7 ~ ~ - 3 ) .

(8)

As a result of investigations, conventional carrier-carrier scattering models as described above failed in calculating correct static voltages of bipolar power diodes especially for low temperatures, i.e. simulated current-voltage characteristics (CVC) and simulated charge carrier profiles (CCP) did not correspond to the measured ones at the same time with the conventional mobility description even for 300K [9]. Therefore a new theoretical approach by Mnatsakanov describing electronhole scattering (EHS) [SI was used within a new diode model yielding much better results [9]. Within this approach an EHS-mobility ppnwhich also depends on p and T is defined. The overall mobilities pp,pn for p n are calculated as follows

In order to verify calculations of the static and dynamic behaviour of a 4,5kV/3kA-IGCT switch a simple buck converter consisting of an IGCT, a free wheeling diode, a clamped turn on snubber inductance but without any turn off snubber circuit for both experiment and simulation is used (Fig. 1). For simulation purposes the load inductance of several mH is substituted by an ideal current source and the gate circuit, which is more complex in reality, by the effective gate inductance of a few nH and the effective gate resistance only. It is understandable that there have to be many physical parameters (w 30) within such a physically based model. An adequate parameter set was found in the following manner: first of all already known data' were used. The unknown parameters were adopted by using three different measured turn off characteristics. Then the comparison of measured and simulated CVC was performed and as expected before no additional change of the parameter set was necessary. Most of the figures within this section include both experimental data' and the simulated curves so that the model performance can be seen directly from the figures.

without applying Matthiesen's rule (6). From (9) and the definitions of 6 and DA(comp. above) it can be seen that b and DA are independent of the hole density p. Taking this into account the continuity equation for calculating the correct CCP for high injection condition can be expressed as [XI

+-b - 1

b + 1 ~n

+ ~p + ppn

PpnDA

""(2)"P
T

P'

1 F7 V,... n
Fig. 1. schematic for measurement and simulation

Equation (5) which determines the voltage along the CCP in the drift zone has to be modified also and can be written as

E ( x )=

jtot

qo(Pn

+ PP)P

jtot +qoPnpP

b-lldp

b + 1 P dx '

(11)

'with courtesy of ABB Semiconductors. Lenzburg. Switzerland

2868

B. The Turn Off Behaviour

Turn off transients for Iloa,j= 1kA vs. VDC = 2.8kV are depicted in Fig. 2. The noisy dotted lines represent the measured transients, the solid lines the simulated ones (also in Fig. 3 , 4 and 6).
V[kVl,

The dynamic avalanche effect can be observed now due to the reduced rise of the blocking voltage from 3kV until its first maximum. To make this obvious an additional simulation with disabled avalanche formula was performed. These curves with neglected avalanche multiplication are represented in Fig. 3, 4 and 6 by the dashed lines.
VlkVl, 5.5

W I
I

W I

10 12 time [IS]

14

16

18

20

Fig. 2. snubberless hard tum off 1kA vs. 2.8kA

Fig. 4. snubberless hard turn off 3kA vs. 2.8M

In this figure experimental and simulation curves match nearly exactly. For Iload = 1kA the maximum electric field corresponding to the first maximum voltage in Fig. 2 near t = 5 , 2 p s is noticeably smaller than the critical electrical field Eerit so that the dynamic avalanche effect does not set in. Therefore the rise of the blocking voltage is relatively linear. The second peak in the voltage curve in Fig. 2 near t = 9ps comes from the transition from the triangle shaped field characteristic to a trapezoidal one. Fig. 3 indicates the transients for Iload = 2kA vs. VOC= 2.8kV.
V[kVI,

From Fig. 4 which depicts anode signals for Ilaad = 3kA vs. VDC= 2.8kV the avalanche effect, can also be seen by comparing full and dashed lines and it is no surpnse that the influence of this effect is more evident than for Ilond = 2kA. In Fig. 2, 3 and 4 a quite good agreement between simulation and measurement even if the avalanche effect influences or predominates the turn off behaviour can be observed Another feature within this model is the possibility to plot the excess charge carner distribuhons during dynamic processes. Fig. 5 depicts for every time step one single CCP.
hole density [ I O "

WI

c' m]

3.5

15

05

time

10
[S I]

12

14

16

18

20

0%

2
Fig 5 distnbution o excess holes dunng turn off f

Fig 3 snubberless hard turn off 2kA vs 2 8kA

2869

As mentioned in section (comp. 11-B) the n-buffer layer is a part of the drift zone module and thus the excess holes both in the drift zone and the buffer layer are calculated numerically and thus plotted in Fig. 5. The fact that the electrical field has to stop when the depletion zone arrives the n-buffer layer and the residual excess hole density in the n-buffer has to decrease merely by a recombination process can be seen in this figure very well. In Fig. 6 the energy losses of a single turn off process for all three currents (comp. Fig. 2, 3 , 4 ) are indicated.

represented by the triangle shaped symbols. The full line corresponds to the simulation including EHS and a little resistor (caused by the measurement equipment) of 73pR. If EHS is neglected the on state voltages are always too small (dashed line). The difference of the static voltage between the full and the dashed line corresponds to the additional term in (1 1) compared to (5). Finally it is worth remembering that the static calculation was performed with the same model parameters as the tum off simulations.

_.
20~

E ,

[Jl

IV. SERIES CONNECTION IGCT OF

Within section I11 only a simple circuit with one switch was considered. Of course more complex circuits with more than one IGCT may be simulated. The series connection of IGCT is a popular measure to raise the DC-link voltage, e.g. of an inverter [5]. Unfortunately snubberless operation cannot he applied to these circuits because during tum off processes an asymmetric distribution of the blocking voltage among the particular switches due to inevitable differences between geometrical or physical data and distinctive gate signals would happen. This phenomenon can also be analyzed by simulation using the presented model. Within preliminary activities simulations two series connected IGCT switches (IGCT 1.2) were performed. The schematic corresponds to a buck converter similar to Fig. 1 but now with two series connected IGCT and two series connected free wheeling diodes (Fig. 8). It is easy to understand theoretically that in case of using two switches with identical geometry and physical characteristics driven by identical gate signals. the problem of asymmetry would not occur. Exactly this unrealistic situation can be confirmed by calculation (see Fig. 9).
3250 -

16~

12-

64-

0 ,

30002750 2500 2250-

20001750150012501000-

750

500250-

O1

I
Fig. 8. series connection of IGCT schematic for simulation

2870

l[WV W l
3.0- 7.5

................... ................... 1 8- 4.5 16140 1.0- 2.5 0.8- 2.0

v-tot . v-gct2 . . ...... . -I-gct1 V-gctl __ 0

0.n- n n

IO

15 20

25 30 tlusl

35

40

45

Fig. 11. two IGCT series connected. with RCD-snubbers

Even if there are small differences between the gate drives or the devices, both voltages at the two switches V,,,, and Vgetz will differ notably. Fig. 10 shows transients of two identical IGCT switches driven by gate signals whereby the gate signal of IGCT 2 has a delay of 2011s compared to the gate signal of IGCT I . It can be seen from this figure that the voltage drop at IGCT 1 exceeds the one at IGCT 2 since the IGCT 1 is turned off earlier due to the delay of the gate signal of IGCT2. It is very interesting that the two voltages diverge not before the anode current is almost equal to zero. Consequently relatively small snubber capacities might be sufficient to ensure symmetrically distributed voltages. For the simulation depicted in Fig. 11 snubbers with Csnub= lOOnF are used.

In this case these snubbers are sufficient to ensure symmetry. In reality the differences between two or more series connected IGCT devices and the gate signals are larger than in the example above, so that the effect of asymmetry is also more predominant. That means that the snubber capacity has to be chosen an order of magnitude higher than in the example (e.g.

2 W).
V. CONCLUSION
Within this paper a new model for calculating the static and dynamic behaviour of hard switched IGCT is presented. The important differences, i.e. the innovative components, compared to older versions of GTO models are discussed with special emphasis to the avalanche effect and EHS according to Mnatsakanov. The most important part of this paper is the validation section in which the simulation performance of this model is demonstrated. This means that calculation and experimental results yield an excellent agreement. In this context it is worth mentioning that for one simulation run of a single turn off, a computer with a 500 Mhz Pentium 111processor needs approximately 14 seconds even if the avalanche formula is considered. Consequently the claim from 11-B that calculations yielding satisfactory results without increasing the simulation effort noticeably can be proven. Finally simulations of series connected IGCT devices are presented. These calculations confirm the necessity of applying RC(D)-snubbers to ensure symetrically distributed voltages after an turn off process due to differences both of particular properties among the switches or between the particular gate drive signals. This field and if there are other possibilities to reach the aim of symmetrically distributed voltages without using snubbers causing additional device effort and costs, e.g. the idea of applying intelligent (active) gate drives, is object of current and future investigations.

IO

15 20

25

30

35 40

45

50

tCsl
Fig. 10. two IGCT series connected, snubberless. gate signal of GCT2 20ns earlier

2871

REFERENCES
[I] J.M. Dorkel and Ph. Leturcq, "Canier Mobilities in Silicon SemiEmpirically related Temperature. Doping and Injection Level", Solid Srafe El.. bf 24, No. 9, pp. 821425 1981. (21 N.D. Arora, [Link] and D.J. Roulston. "Electron and hole mobilities as a function of concentration and temperature", IEEE Trans. El. Dev., 29 2, pp. 292. 1982. [31 D. Metzner, T. Vogler and D. Schroder, "A Modular Concept For The Circuit Simulation Of Bipolar Power Semiconductors", Proc. of EPE, Brighton, 1993. [4] T. Vogler, A.E. Schlogl, N. Kasahara and D. Schriider, "Modeling and Characterizing Power Semiconductor Devices at Low Temperatures", Proc. of IPEC, Yokohama, 2, pp. 1232-1238. 1995. [51 P. Steimer, H. Gdning, J. Werninger and D. Schroder, "State-of-theArt Verification of the Haed Driven GTO Inverter Development for a 100 MVA Intertie". Proc. of PESC, Baveno, 1996.

[61 H. Grilning, B. 0degbd and J. Rees "High-Power-Driven GTO Module for 4.5kVI3kA Snubberless Operation", PCIM, Nilmberg, 1996. [71 S. Klaka, M. Frecker, H. Grilning "The Integrated Gate-Commutated Thyristor: A New High-Efficiency, High-Power Switch for Series or Snubberless Operation", PCIM, Niimberg, I997 [81 Mnatsakanov, T.T.. D. Schriider and A.E. Schlogl, "Effect of High Injection Level Phenomena on the Feasibility of Diffusive Approximation in Semiconductor Device Modeling", Solid Stare El., bf 42, No. I , pp. 153163, 1998. [91 A. Schl6gl. T. Mnatsakanov, H. Kuhn and D. Schroder "Temperature Dependent Characterization of Silicon Power Semicondutors - A New Model Validated by Device-lntemal Probing between 400K and IOOK', Pmc. of PESC, Kyoto, 1998. [IO] S. Bernet, R. Teichmann. A. Zuckerberger and P. Steimer "Comparison of High Power IGBTs and Hard Driven GTOs for High Power Inverters", APEC, Anaheim, 1998.

2872