Influence of Power Cycling Strategy on IGBT Lifetime - A Case Study
By John Parry, Industry Manager, Mentor Graphics
Vendors are working hard to increase the maximum power level and current load capability of IGBT and other power devices, while still maintaining high quality and reliability. Innovation has brought new technologies such as ceramic substrates with improved thermal conductivity, ribbon bonding to replace thick bond wires, and solderless die-attach technologies to enhance the cycling capability of the modules.
Power modules are also being designed and manufactured by end-users because the chips, the required direct bond copper (DBC) substrate, and a variety of different die-attach materials are all available on the open market. This offers increased flexibility in terms of mechanical design, however it raises severe thermal and reliability challenges as their end use is typically where high reliability is critical such as hybrid and electric vehicles. High junction temperatures and high temperature gradients during operation induce mechanical stress especially at contacting surfaces between materials with different coefficient of thermal expansion, which may lead to the degradation or the complete failure of these components. To avoid premature failures, proper thermal design and material selection is necessary.
Mentor’s Power Tester 1500A is designed to automate the process of qualifying the reliability of parts to make a good estimate for the power module’s lifetime during service, and to identify weaknesses that can be removed during development thereby increasing reliability and lifetime. This case study details the application of the Power Tester with four medium power IGBT modules containing two half bridges, demonstrating the rich data obtained from automated power cycling of the components. This article is abstracted from the two technical papers given in the references. [1,2]
The modules were fixed to the liquid cooled cold plate integrated into the Power Tester with a high-conductivity thermal pad to minimize the interfacial thermal resistance. The coldplate temperature was maintained at 25°C throughout the whole experiment using a refrigerated circulator controlled by the Power Tester. The gates of the devices were connected to their drains (the so-called magnified diode setup) with each half bridge powered using a separate driver circuit. Two current sources were connected to each half bridge. A high-current source that can be switched on and off very fast was used to apply stepwise power changes to the devices. A low current source provided continuous biasing of the IGBT allowing the device temperature to be measured when heating, connecting it to a separate measurement channel of the Power Tester.
An initial set of tests on four samples was conducted using constant heating and cooling times. Heating and cooling times were selected to give an initial temperature swing of 100°C, for a power of ~200W with 3s heating and 10s cooling. This most closely mimics the application environment, where degradation of the thermal structure results in a higher junction temperature leading to accelerated aging. Of the four devices, Sample 3 failed shortly after 10,000 cycles, significantly earlier than the others. Samples 0, 1 and 2 lasted longer, failing after 40,660, 41,476 and 43,489 power cycles respectively. Figure 1 illustrates the structure functions generated from the thermal transients measured on Sample 0 after every 5,000 cycles. The flat region at 0.08 Ws/K corresponds to the die-attach. It can be seen that the structure is stable until 15,000 cycles, but after that point the degradation of the die-attach can be clearly noticed as its resistance increases continuously until the device fails. Again, the immediate cause of the device failure is unknown, but we found that a short circuit is formed between the gate and the emitter and burnt spots could be seen on the chip surface.
A second set of tests were performed on an identical set of samples using the different powering strategies supported by the Power Tester. In this case we kept the current constant for IGBT1, the heating power constant for IGBT2, and the junction temperature change constant for IGBT3. To ensure a fair comparison, the settings were chosen to give the same initial junction temperature rise for all components, with 3s heating and 17s cooling, and ~240W initial heating per device chosen for the test. The whole heating and cooling transient is measured for each device in all cycles, with the following electrical and thermal parameters monitored continuously by the Power Tester:
- Device voltage with heating current turned on, Von
- Heating current applied in the last cycle, ICycle
- Power step, P
- Device voltage after heating current turned off, Vhot
- Device voltage before heating current turned on, Vcold
- Highest junction temperature during the last power cycle, Thot
- Lowest junction temperature during the last power cycle, Tcold
- Temperature swing in the last cycle, ΔT
- Temperature change normalized by the heating power, ΔT/P
Additionally, the full length thermal transient from powered on steady state to powered off steady state was measured after 250 cycles using a 10A heating current, to create structure functions to investigate any degradation in the thermal stack. Again, the experiment was continued until the failure of all IGBTs.
As expected, IGBT1 failed first, as there is no regulation of the supplied power as the part degrades. Interestingly, it showed no degradation in the thermal structure as shown in Figure 2.
In order to find the cause of the device failure we have to examine evolution of device voltage during the experiment. In Figure 3 the forward voltage of IGBT1 at heating current level can be seen as function of elapsed power cycles. In the first three thousand cycles a decreasing tendency can be seen.
This initial change caused by the slow change of the average device temperature that decreased by almost 5°C. Despite the negative temperature dependence of the device voltage at low currents, at high current levels the temperature dependence of the forward voltage become positive. After about 35,000 cycles this tendency changed and the voltage started to increase slowly. This was followed by stepwise changes in the device voltage while the increasing tendency continuously accelerated until the failure of the device. As the structure did not change the increasing voltage can be attributed to the degradation of the bond wires. This also gives an interpretation to the stepwise changes of the voltage when a bond wire finally detaches.
The increasing heights of these steps are caused by the increasing change in the parallel resistance sum of the bond wire thermal resistance as the number of bond wires decreases. If we use constant current strategy, the crack of a bond wire increases the current density in the remaining bonds and accelerates aging.
Figure 4 shows the same type of curve corresponding to IGBT3. Here the increasing tendency of the device voltage starts even earlier but due to the regulation to keep the junction temperature constant, the heating current was proportionally decreased. The decrease in current reduced the load on the bonds and increased the measured lifetime.
To conclude, the two sets of experiments were conducted which showed different failure modes, illustrating how different powering strategies, and possibly electrical setup, can influence failure mode. The first set of measurements at a constant cycle time, that most closely reflects operational use, verified that the Power Tester is able to detect immediately the appearance of degradation within the device’s structure, including the die-attach and other compromised layers.
The second experiment clearly identified degradation of the bond wires as the forward voltage of the device was observed to increase stepwise. While with these powering options, (current constant, constant heating power, and constant temperature rise), the thermal structure did not change for any of the samples tested. Due to the low number of samples we have to be conservative in formulating conclusions. However the results warns us that the measurement results can differ depending on the cycling strategy, and lifetime predictions based on certain strategies can overestimate the real lifetime of power devices.
- Zoltan Sarkany, Andras Vass-Varnai, Marta Rencz (2013) Investigation of die-attach degradation using power cycling tests, Proceedings of 15th IEEE EPTC, pp 780 – 784, Singapore.
- Zoltan Sarkany, Andras Vass-Varnai, Sandor Laky, Marta Rencz (2014) Thermal Transient Analysis of Semiconductor Device Degradation in Power Cycling Reliability Tests with Variable Control Strategies, Proceedings of SEMI-THERM 30, pp. 236-241, San Jose CA.