IGBT GATE-DRIVE WITH PCB ROGOWSKI COIL FOR IMPROVED SHORT CIRCUIT DETECTION AND CURRENT TURN-OFF CAPABILITY

Transcription

1 IGBT GATE-DRIVE WITH PCB ROGOWSKI COIL FOR IMPROVED SHORT CIRCUIT DETECTION AND CURRENT TURN-OFF CAPABILITY D. Gerber, T. Guillod, and J. Biela Laboratory for High Power Electronic Systems ETH Zurich, Physikstrasse, CH-89 Zurich, Switzerland This material is posted here with permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any of ETH Zürich s products or services. Internal or personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by writing to pubs-permission@ieee.org. By choosing to view this document you agree to all provisions of the copyright laws protecting it.

2 IGBT GATE-DRIVE WITH PCB ROGOWSKI COIL FOR IMPROVED SHORT CIRCUIT DETECTION AND CURRENT TURN-OFF CAPABILITY D. Gerber, T. Guillod, and J. Biela Laboratory for High Power Electronic Systems ETH Zurich, Physikstrasse, CH-89 Zurich, Switzerland Abstract In this paper, a gate drive using gate boosting and double-stage turn off including voltage clamping as well as with detection of overcurrent and a too high di/ during turn on is discussed in detail. Besides the gate drive, also the design of a PCB-Rogowski coil, which is used for measuring currents and for di/ detection, is explained and different designs are compared. The presented coil has a bandwih of more than 8 MHz and a propagation delay of ns. Load I. INTRODUCTION Currently, a compact and cost effective X-ray free electron laser facility (SwissFEL) is planed at the Paul Scherrer Institute (PSI) in Switzerland. For this laser system, high power modulators meeting the specifications given in table I are required. There, an important feature of the modulator is the pulse repetition accuracy, which must be better than. To fulfill these challenging requirements, a solid state modulator based on a matrix (split core) transformer with six cores as shown in Fig. a) is designed. In the modulator in total twelve 4. kv Press-Pack IGBTs manufactured by ABB are used and each IGBT is connected to a separate primary winding and always two primary windings are around one core. Due to the matrix transformer, the current balancing between the IGBTs on different cores is inherently provided as explained in [], so that only the current balancing between the two IGBTs connected to windings on the same core must be guaranteed by the gate drive circuit/control. As the considered pulse length is relatively short, the switches are operated at pulse currents much higher than the rated switch current, in order to minimize the number of required IGBTs and fully utilize the semiconductors. This operation condition is not very critical with respect to thermal issues and/or thermal cycling due to the relatively large thermal capacitances and short pulses, but with the high pulse currents, the margin between currents during normal operating and maximum switch current (latching current []) becomes much smaller. Therefore, the current TABLE I SPECIFICATIONS FOR THE SOLID STATE MODULATOR (V In = KV). Output Voltage Output Power Repetition Rate Repetition Accuracy 7 kv MW Hz Reset-Circuit Main Switch Pulse Generator Pulse Transformer (a) (b) Bouncer Fig.. a) Schematic of the solid state modulator with: generator circuit (incl. active reset), pulse transformer and bouncer circuit. b) Photo of one generator circuit consisting of energy storage capacitors, Press-Pack IGBT switch and active reset circuit. through the switch could relatively quickly rise to levels, where the IGBT could not be turned off any more in case of fault conditions or arcing of the klystron load. A method to avoid this problem is to apply gate voltages, which are just large enough, so that the IGBT could only conduct the nominal current, but enters the linear operation in case of an overcurrent. However, with the small gate voltage, the turn on behavior of the IGBT is slow, limiting the rise time of the pulse during turn on. In order to limit the short circuit currents also in case of high gate voltages, a fast and reliable overcurrent / short circuit detection is required. In [], a di/ short circuit detection utilizing the parasitic inductance of the kelvin contact has been investigated and the reliability of the method has been demonstrated.

3 Unfortunately the considered 4. kv press-pack IGBTs do not provide a kelvin contact, so that in this paper a di/ short circuit detection method based on PCB- Rogowski coils is proposed. Such Rogowski coils offer a high bandwih and can be used up to very high currents due to the lack of a ferromagnetic core which can saturate. Furthermore, the fabrication of a PCB-coil is simple and cheap and it is easy to integrate it in a press-pack stack. In the proposed system, each gate drive is able to detect overcurrents in both IGBTs connected to windings on the same transformer core, in order to reduce the turn off time of the two IGBTs. In order to minimize the turn on time, a high gate voltage is used at the beginning of the pulse, which is reduced after the IGBT is fully turned on. Additionally, the gate drive provides a -stage turn off and an overvoltage clamping [4] in order to reduce the pulse fall time during turn off and lower the switching losses. In the following, first, the gate drive circuit and the short circuit detection are described and thereafter, the design of the coil is presented in detail in section III. II. GATE DRIVE In the considered solid state modulator, a matrix (split core) transformer, which consists of six cores is applied. The two parallel connected secondary windings both enclose all six cores. Furthermore, each core carries two primary windings and only one IGBT is connected to a primary winding. Consequently, there are twelve primary windings and twelve IGBTs in the considered MW modulator. As explained in [], the current distribution between windings mounted on different cores is inherently provided by the matrix transformer. However, the current balancing between the two switches/primary windings on the same core must be controlled by the gate drive/control circuit. This could be for example achieved by scheduling the gate pulses as described in [4]. As already mentioned, for achieving a fast rise/short pulse rise time, a high gate-emitter voltage is required during turn on. After the IGBT is turned on, the gate voltage could be reduced to limit the maximal current through the IGBT before the IGBT enters the linear operation mode. Such a technique is known as gate boosting and is for example described in []. While the gate voltage is boosted, a too high current could flow through the IGBT in case of a short circuit, resulting in a latch-up of the parasitic thyristor in the IGBTs pnpn-structure. If the parasitic thyristor is ignited by a too high current, the IGBT could not be turned off via the IGBT-gate and will be destroyed in case of a short circuit. Such a situation must be avoided by a fast short circuit detection and a fast turn off of the IGBT. In case of a short circuit, turning off of one of two IGBTs on the same core causes the current to commutate to the other IGBT on the same core. Since the pulse currents are high compared to the nominal current, this could lead to latching and/or destruction of the IGBT. A detection and synchronous turn off via the main control would be relatively slow due to the optical link, which is used for transmitting the control signals. Therefore, both IGBTs must be turned off at the same time. For achieving a short delay time between short circuit detection and IGBT turn off, the presented gate drive circuit is capable to detect a short circuit in both IGBTs at the same time as. The implementation is relatively simple because the current measurement with a Rogowski coil is floating and two Rogowski coils providing two independent current measurements can be implemented on the same PCB. A further feature implemented in the discussed gatedrive is a double-stage turn off circuitry with active clamping as described in [6] and as could be also seen in the block schematic of the gate drive given in Fig.. In the following, the operating blocks are shortly discussed. a) Turn On Stage: The basic idea of the turn on circuit is using a higher gate voltage during turn on. To do so, two supply voltages are used. Because the gate voltage has to be reduced after turn on, an additional switch is required to discharge the gate to the lower voltage rail when the IGBT is turned on. b) Turn Off Stage: At the beginning of the turn off action, a low gate resistance is used to achieve fast switching speeds. Due to parasitic inductances in the power circuit, an overvoltage occurs across the IGBT during turn off. To limit this overvoltage, the switching speed is reduced by using a higher gate resistance as soon as the voltage reaches a certain limit. This is done by using two gate resistors a high value resistor, which is used continuously during turn off, and a low value resistor, which is connected in parallel only at the beginning of the turn off action (cf. also [6]). Additionally, an active clamping circuit is used to limit the over voltage at a fix level by slightly turning the IGBT on again (linear mode). c) Status Detection: To properly control turn on and turn off process, the status of the IGBT, i.e. if it is fully turned on or off, must be known. For detecting the status, the value of the DC link voltage, which could be varied during operation to change the output power, must be known. This detection is performed by a simple track and hold stage (THA) and two comparators, which track the DC link voltage V CE,m before the main IGBT turns on and compares the actual IGBT voltage with V CE,m. With the detection three control signals are generated: Signal indicates if the IGBT is turned off, signal indicates if the switch is turned on and signal is used to determine if the low or the high gate resistance is required during turn off. The references for the first two signals are adjusted from pulse to pulse by the track and hold stage. This is not necessary for the third signal because the overvoltage limit is independent of the DC link voltage. d) Integrator & Short Circuit Detection: Since the Rogowski coil only measures the di/, an integrator circuit is required for obtaining the current amplitude. As the analog integrator circuit is sensitive to DC-offsets, the presented gate drive uses an active offset compensation, which uses an LF analog PI-controller that feeds the integrators output voltage back to its input (not shown in Fig. ). Alternatively, the integrator could be reset at the beginning of each pulse. To detect short circuits two comparators are used to indicate an overcurrent and two comparators to indicate a too high di/ value.

4 Vgate, Status Detection Vgate, Turn On Stage C Rfb VCE,m THA Ron, Ron, Active Clamping FPGA G VCE,m Turn Off Stage Vce Roff, Roff, VCE,m C-E Voltage Measurement Vref,off Trigger for Double-Stage Turn-OFF Integrator Vlim,I VCoil E Integrator Vlim,I Vlim,dI/ VCoil Vlim,dI/ Short-Circuit Detection Fig.. Block schematic of the gate drive circuit where the functional blocks are summarized in the grey highlighted areas. e) Control Circuitry: For obtaining a flexible and compact gate drive, the control circuitry is implemented in a FPGA. There, also a CPLD could be used in order to reduce costs, which, however, limits the flexibility during the development of the code. To avoid ground loops, the status signals as well as the switching signal are transmitted via plastic fibre optic links and the power supply is galvanically isolated. due to a failure. When the IGBT is turned off, both turn on voltage rails are disconnected from the gate and both turn off resistors are connected to reduce the gate voltage rapidly. As soon as the collector-emitter voltage reaches a given level, one turn off resistor is disconnected. This reduces the switching speed and therefore also the overvoltage caused by parasitic inductances. Finally, the track and hold amplifier is set into the track state as soon as the switch is turned off. During the switching action, the current status is transmitted to the main control unit. This can be used to synchronize the edges and in case the higher gate voltage A. Gate Drive Operation When a switching signal is received, the track and hold amplifier is put into the hold mode. As soon as this is done, the turn on process begins and the MOSFET connecting the higher gate voltage to the IGBT gate is turned on. Because the n-channel MOSFET connecting the lower voltage rail to the gate has a diode in series, it is turned on at the same time, in order to reduce the rise time of the gate voltage. As soon as the IGBT collector-emitter voltage has dropped, the higher voltage rail is disconnected and a switch is closed to discharge the IGBT gate capacitor to the level of the lower gate voltage. For safety, there is also a time-out circuit implemented on the FPGA which reduces the gate voltage after a certain amount of time in case the IGBT voltage does not drop Fig. 4. Photo of the gate drive. IC VGE V VCE or Time Controlled V, R Vgate, VCE VGE Vgate, VCE Turn On Fig.. Roff t Turn Off Gate drive operation during turn on and turn off. VCE t Fig.. Voltage and current waveforms for a µs pulse. The scale for VCE is V per division, for IC A per division and for VGE it is V per division. The time scale is ns per division.

5 can be adjusted by the main control also to synchronize the switching speeds during turn on. In Fig., measured waveforms of the gate drive are shown nicely matching with the theoretical curves. M di L coil C coil R d V coil III. ROGOWSKI COIL For detecting a short circuit, a current measurement system based on PCB-Rogowski coils is used. With this system not only the amplitude of the current could be measured, but also the di/ of the current enabling a faster short circuit detection. The details of the coil design, measurement accuracy and simulation results are presented in the following. A. Coil Design The operation principle of a Rogowski coil is based on induction, i.e. a flux variation induces a voltage in the turns of the Rogowski coil. The induced voltage at the coil terminals is: V ind = N i= dφ n = M di where N is the number of turns and M the mutual inductance between the loop conducting the current to be measured and the Rogowski coil. Assuming that the magnetic field is constant inside the area enclosed by the turns of the Rogowski coil, the mutual inductance is given by M = N l () A µ, () where l is the length of the coil and A the area enclosed by a single turn. This equation is only valid for currents flowing in the middle of the coil. For currents not flowing in the middle, the mutual inductance has to be calculated with a more complex method which is described later. This assumption is valid because the wih of the area enclosed by a single turn is small compared to the distance of the current to be measured. In the considered gate drive, two separate coils are used for a faster short circuit detection as described above. In order to obtain the same measurement result for both of the Rogowski coils, which are realized side by side on the same PCB, the mutual inductance for both coils must be the same. Considering the equation above this means that N l = N l () must be fulfilled if the area is the same for both coils. There, the number of turns is mainly determined by the desired measurement accuracy and bandwih. In the considered case, the outer coil on the PCB has 46 windings, the inner coil has 6 windings. This corresponds to.4 and.4 turns per cm, which results in only a deviation of the mutual inductance of.47 %. The cross-sectional area of the turns is. mm, so that a mutual inductance of.4 nh results for both coils. Fig. 6. Electrical model of a Rogowski coil valid up to the first resonance frequency. The distributed inductance and capacitances are summarized in two lumped components. R d represents an additional damping resistor. Voltage (V)... ζ =. ζ =. ζ =.7 ζ = M di Time (μs) Fig. 7. Coil response on a square wave input voltage for different damping. ) Coil Bandwih Besides the self-inductance each turn of the Rogowski coil has, there are also parasitic capacitances between the turns. These can be modeled by an equivalent circuit consisting of many Ls and Cs, which result in a high number of resonances. However, in normal operation the Rogowski coil is only used at frequencies below the first resonance frequency, so that the electrical model of a Rogowski coil could be simplified to a simple LC network as shown in Fig. 6. Based on the equivalent circuit given in Fig. 6, the second order transfer function of the coil up to the first resonance frequency is given by: V out M di = s LC + s L Rd + = ω s + ζω s + ω The resonance frequencies of the two designed coils the inner and the outer one on the PCB as shown in Fig. are MHz and 8 MHz. With these values, the time domain response of the measurement system could be determined. In Fig. 7, the time domain response of the outer coil with a resonance frequency of 8 MHz for a rectangular input is shown. As could be seen in Fig. 7, the coil response strongly depends on the damping ζ. Because the coil should be also used for a di/ short circuit detection, no overshoot is allowed in order to avoid false short circuit detections. Therefore, the damping resistor R d has to be selected properly to achieve ζ =. The delay for a ramp voltage, which results due to the finite resonance frequency f res, could be calculated by: t d = (4) ζ πf res. ()

6 .. e>% e>%..4 e<.%.4. e<.% Error e (%)..... Fig. 8. Measurement error of the inner PCB Rogowski coil with constant winding density. The given error is for perpendicular line currents flowing at the considered location of the display error. For ζ = and fres = 8 MHz this results in a time delay of ns, which shows that the coils can be used to detect a short circuit very fast. Error e (%).. Increased Winding Density. Fig. 9. Measurement error of the inner coil with increased winding density in the corners. formed by analytical calculations. Therefore, -D PEEC simulations are performed, which are discussed in the following section. C. Coil Simulation B. Measurement Accuracy Besides the bandwih, also the measurement accuracy of the coil is an important issue. There, two factors influence the accuracy: first if the current, which should be measured, is not exactly flowing in the middle of the Rogowski coil, and second, other currents flowing in the vicinity of the Rogowski coil, which also could result in an induced voltage/measurement signal. For determining the achievable accuracy for measured currents not flowing in the middle of the Rogowski coil, first the mutual inductance between each turn of the Rogowski coil and the loop, where the measured current is flowing, is calculated for each turn for the considered location of the measured current as described in [7]. The total mutual inductance is obtained by adding up the values for the individual turns. By comparing the calculated mutual inductance to the value for the ideal mutual inductance, which is obtained, when the measured current is flowing in the center of the Rogowski coil, the measurement accuracy is obtained. The ideal value for the considered coil given in Fig. is.4 nh. The same method could be applied to determine the influence of currents flowing outside the coil, which should not result in a measurement signal, i.e. the ideal mutual coupling for these currents is nh. In Fig. 8 the achievable accuracy for measured currents flowing at different positions through the coil are given. The shown error value is for an ideal line current, which is flowing perpendicular through the coil at the considered position. In the figure it can be seen that the measurement error for currents flowing close to the corners of the coils is relatively high. By shifting turns to the corners and therefore increasing the number of turns in the corners with a constant total number of turns, the accuracy can be improved significantly as shown in Fig. 9. To increase the measurement accuracy further, more windings would be required. However, this would reduce the bandwih. In a next step, the measurement accuracy of currents flowing not perpendicular through or outside the coil is investigated. Unfortunately, this could not easily per- For non-perpendicular currents, the induced voltage/signal can be relatively large since the Rogowski coil has one big turn in addition to the small turns, which results due to the large loop the Rogowski coil encloses between the start and the end point (cf. ALoop in Fig. ). For coil () in table II, the total area enclosed by all turns is 6 mm and the area ALoop is mm resulting in a high sensitivity to noise generated by non-perpendicular currents. For avoiding these disturbances, an additional return conductor from the end of the coil to the start, so that both connection points of the coil are on the same Rogowski coil point, can be used. An example is coil () in table II, where the area enclosed between the starting and the end point of the Rogowski coil is reduced to 7 mm. The second method is by continuing the turns at the end of the coil back to the starting point along the same path instead of just using a single wire. An example is coil () in table II, which has an area of approximately 4 mm between the starting and the end point of the Rogowski coil. In table III the mutual inductance for the three different designs for different locations of flowing currents are given (yellow boxes and red arrows). In row the ideal situation with the ideal mutual inductance is given. For currents flowing outside the Rogowski coil, the mutual inductance should be as mentioned above. These values have been obtained with GeckoEMC, which uses the PEEC method. The coil without compensation shows the best frequency performance, but the noise immunity is very poor. With the compensation, the bandwih reduces to approximately MHz, which is relatively independent of the compensation method. Since the noise immunity of the coil () is slightly better, this geometry is used for the Rogowski coil. In Fig., the real test setup is shown. As could be seen, the current is not flowing only through the middle of the coil, but also on the outside in order to achieve a low inductive setup for the IGBT. Therefore, the mutual inductance in the test setup is a combination of the investigated cases in table III. This leads to an increased

7 TABLE II D IFFERENT WINDING TOPOLOGIES FOR THE ROGOWSKI COIL : () WITHOUT COMPENSATION, () WITH A SINGLE RETURN CONDUCTOR AND () WITH TURNS FROM THE STARTING POINT TO THE END AND BACK. Load Capacitor + Fig IGBT Emitter Collector D model used for the simulation of the IGBT test setup. TABLE III C OMPARISON OF DIFFERENT COMPENSATION METHODS. Winding topology L (µh) fres (MHz) nh.4 nh.9 nh.4 nh. nh.7 nh.9 nh.94 nh.89 nh.94 nh.87 nh.66 nh Fig.. Picture of the presented Rogowski coil. the current amplitude but also the di/ for detection, PCB-Rogowski coils are used that could be easily integrated in the press-stack. The design of these Rogowski coil is investigated in detail in the paper by analytical calculations and by D simulations based on the PEEC method. Furthermore, different designs for the coil are compared. There, a bandwih of more than 8 MHz and a propagation delay of ns has been achieved for the coil. ACKNOWLEDGMENT 7.79 nh.497 nh. nh 6.7 nh.4 nh.78 nh 7.9 nh.9 nh.9 nh.6 The Authors would like to acknowledge the support of ABB Semiconductor, who provided IGBTs for the experimental system, and the strong support of PPT in relation with the practical realization of the project. References total mutual inductance in the real circuit which is around % higher than the ideal inductance. A picture of the PCB-Rogowski coil is shown in Fig.. IV. CONCLUSION In this paper, a gate driver circuit designed for 4. kv press-pack IGBTs used in a MW/7 kv solid state modulator is presented. The gate drive implements gate boosting for achieving faster turn on and a double-stage turn off with additional voltage clamping to minimize turn of time and limit the occurring overvoltage. For a fast over-current detection, which uses not only [] D. Bortis, J. Biela and J.W. Kolar, Transient behaviour of solid state modulators with Matrix Transformers, 7th IEEE International Pulsed Power Conference (PPC), 9, pp [] N. Mohan, T. M. Undeland and W. P. Robbins, Power Electronics, Third Edition, John Wiley & Sons, United States of America,, pp [] M.N. Nguyen, R.L. Cassel, J.E. delamare and G.C. Pappas, Gate drive for high speed, high power IGBTs, Digest of Technical Papers Pulsed Power Plasma Science,, pp [4] D. Bortis, J. Biela and J.W. Kolar, Active Gate Control for Current Balancing in parallel connected IGBT Modules in Solid State Modulators, 6th IEEE International Pulsed Power Conference (PPC), 7, pp. -6. [] A. Volke and M. Hornkamp, IGBT Modules, Technologies, Driver and Application, Infinieon Technologies AG, Munich,, pp [6] D. Bortis, P. Steiner, J. Biela, and J.W. Kolar, Double-Stage Gate Drive Circuit for Parallel Connected IGBT Modules, Proceedings of the IEEE International Power Modulators and High Voltage Conference, 8, pp [7] N.Karrer, P. Hofer-Noser and D. Henrard, HOKA: A New Isolated Current Measuring Principle and its Features, Conference Record of the IEEE Industry Applications Conference, 999, pp. 8.