Datasheet switching losses of IGBTs and their freewheeling diodes are determined in a double pulse test with a purely inductive load. This closely resembles the application conditions in most cases and is therefore also set out in the applicable semiconductor standard (IEC 60747-9). For drives with a low or medium output power with long, shielded cables, however, a substantial capacitive load occurs for the power semiconductor used which significantly changes the switching behaviour. The capacitive effects increase the total losses of the power inverter and neglecting this can therefore lead to incorrect product design and to overloading of components. Some of the results presented here are taken from Denis Richter's Master's thesis undertaken at the Otto von Guericke University Magdeburg entitled "Detaillierte Untersuchung des Schaltverhaltens von Leistungshalbleitern bei langem Motorkabel" [Detailed investigation of the switching behaviour of power semiconductors for long motor cables].
Measuring switching losses
Test circuit and test object
The effect of the shielded cable for a drive application with a three-phase voltage inverter and motor load is to be investigated. Image 1 shows the switching diagram with the main EMC components (mains interference filter and Y-capacitors) of a drive. The switching processes of the IGBTs cause a potential jump on the motor side with a high change of voltage per time between the phases and between the phase and earth (PE, earth), which forces a current flow across the parasitic capacitances between the conductors.
A typical IGBT module in this power range is a MiniSKiiP. For the results presented here, a 1200V IGBT with 8A nominal current was selected (see Image 2). Depending on the cooling and pulse frequency, a module of this kind can be used in the MiniSKiiP1 case for drives up to 5.5kW. For switching loss measurements, the module is connected to a low-inductance test PCB using the DC link capacitor and the driver. A double shielded 4x4mm² cable connects the power inverter directly to the motor.
The cable capacitances are between 100 and 150pF/m and, due to a parallel connection effective in the switching process, for a two metre cable are already ten times larger than the semiconductor and module capacitances. They dominate the switching behaviour as the length increases. The proportion of the current through the motor capacitance is already < 15% at a cable length of 10m. However, the motor capacitance should still be included in the investigation, for example a smaller and longer servo motor may have capacitances which differ from those of the standard asynchronous motor used here.
The arrows in Image 3 show the current flow when turning-on T2 as an example. The capacitive current flows from the IGBT across the Y-capacitors to the case earth and then across the cable capacitances back to the DC link and the IGBT.
For the high-frequency capacitive currents, the DC link and Y capacitors act as short circuits. Since the motor housing is also connected to the earth potential, some of the current flows across the earth potential. For shielded cables, it is unimportant whether the cables are routed straight or on a drum because there is no capacitive coupling with a changing electrical potential outside the cable.
Measuring switching losses in a double pulse test
Switching losses are measured in a double pulse test. In the example, the component being tested is the lower IGBT of phase U (T2) with the upper freewheeling diode (D1). The other five IGBTs are either permanently turned on (T3, T5 = + DC to phase V and W) or permanently turned off (T1, T4 and T6). The desired current is set by the pulse duration of the first impulse. After turning-off the IGBT, a freewheeling phase takes place until the IGBT is turned-on for the second time. The current remains virtually constant during freewheeling. The first turn-off and the second turn-on are used to measure the current and voltage variations and the switching energy (see Image 4).
Influence of various circuit parameters
The cable length has been varied between 2m and 50m. During turn-on, the capacitive current superimposed on the load current dominates the switching behaviour. The amplitude of the capacitive current increases as the length increases. At a length of 50m, the IGBT is actually desaturated and limits the current to approx.
24A (3x IC(nom)). In the following half-wave, the current becomes negative, the antiparallel diode conducts and the voltage across the switch becomes negative. The additional current, and sometimes also the slow fall in the collector-emitter voltage VCE, increases the turn-on power dissipation Eon to more than 250% compared to the value without a cable.
During turn-off, unloaded switching can be observed since the capacitances at the output reduce the voltage increase. The main part of the capacitive current flows through the DC link. The IGBT turn-off power dissipation Eoff is reduced to approx. 50%, but the high-frequency current causes additional losses in the DC link capacitor. The level in the output voltage can be explained by some wave running effects in the cable, which are independent of the semiconductor switching speed or the gate driver conditions. The longer the cable, the more distinctive the level. The plateau falls with the current and at low currents the voltage may have multiple plateaus.
The load current level has a significant influence on the percentage increase in the turn-on losses and the voltage rise time during turn-off. While Eoff approaches zero in proportion to the current, turn-on power dissipation is caused despite no load current (0A) is flowing. The offset to the switching losses (absolute value) is virtually constant across the entire inverter current range.
The amplitude of the superimposed capacitive current is similar for low and high load current levels. At low currents, this rapidly leads to the current direction changing during oscillations and the antiparallel diode becoming conductive. During turn-off, the low current is not capable of quickly charging the cable capacitances parallel to the IGBT. The voltage increase slows in proportion to the reduction in the load current.
DC link voltage
The influence of the DC link voltage on the switching losses is roughly the same as under standard switching conditions with a purely inductive load. Eon at 400V is around 50% of the value at 600V and Eoff at 400V is 78% of the 600V value.
Both at room temperature and in hot conditions, the losses increase with longer cables. The percentage increase compared to the purely inductive losses is smaller at high temperatures than in the cold. The reason for this is that the semiconductor losses increase as the temperature rises but the capacitive part remains virtually constant. Out of the 39% additional idling losses at 25°C, around 23% remain at Tj = 125 °C and roughly 20% at Tj = 150°C in relation to the switching losses at nominal current in each case.
Additional switching losses in inverter mode
Idling losses are neglected in the conventional formula for PWM power inverters. These are assumed to be zero at zero current. It is necessary to extend the calculation with an expression for the capacitive cable current which adds an offset to the switching losses (energy multiplied by the switching frequency) regardless of the load current. This offset varies for different cable lengths (see Image 5). An example should clarify this influence. Let's assume we have a 3.7kW power inverter with 400V output voltage and a switching frequency of 12kHz at 700VDC voltage. Without cable load, an 8A/1200V IGBT causes approx. 6W conduction losses + 10W switching losses and is thus 30K hotter than the heat sink temperature in the MiniSKiiP case. With a 50m cable, the switching losses almost double to 18.5W and the IGBT becomes 45K hotter than the heat sink.
Extension of the interlock time
Interlock times are necessary in order to prevent a dynamic short circuit between the switches in a power inverter phase. It should be ensured that an IGBT is completely turned-off before the other IGBT is turned-on. The rise time of the voltage during the turn-off process at virtually zero current can increase in the μs area. In this case, the minimum interlock time is determined by the cable capacitances and not by the semiconductor.
The state-of-the-art for short-circuit protection involves monitoring the on-state voltage of the IGBT, which is compared to a reference voltage of typically between 5 and 7V. If the on-state voltage exceeds the reference, the IGBT is turned-off. Monitoring is activated with a delay of a few μs after the turn-on command so that the IGBT has reached its stationary forward voltage. If, however, the capacitive oscillations last several μs, the delay must be extended. For the latest generation of IGBTs with short circuit pulse durations of ≤10 μs, extensions such as these can be implemented. For future IGBT generations with higher current densities, short circuit pulse duration is reduced. As a result, it may be that it is no longer possible to use this kind of current short circuit protection or that unwanted erroneous turn-offs take place.
Shielded cables are often required in order to meet EMC requirements. The coupling capacitances of the cable increase the turn-on energy of an IGBT more than the turn-off energy is reduced. An increase in switching losses should therefore be taken into consideration, the level of which depends on the cable length. The additional losses can be determined during idling at zero current and are added as a simple but good starting point as a constant offset for the entire current range.
At higher inverter power, the additional losses are less important since the switching losses caused by the semiconductor are then dominating, e.g. Esw from a 10A IGBT compared to a 100A IGBT increase nearly linear by a factor of 10 while the cable capacitances increase by less than factor 3 in the same current level. It is not possible to specify up to what current the effect of shielded cables is to be taken into account since it depends on the maximum cable length, the ratio between the switching and conductive losses and motor operation. This effect should be included if a few individual watts per IGBT make a significant contribution to the total losses. The cable capacitances also have an impact on the losses in the DC link capacitor, the switching and interlock times and the signal suppression time of the short-circuit protection with VCE(sat) monitoring.