Advanced Power Modules with AlN-Substrates Extending Current Capability and Lifetime

Transcription

1 Advanced Power Modules with AlN-Substrates Extending Current Capability and Lifetime U.Scheuermann SEMIKRON Elektronik GmbH, Sigmundstraße 2, 9431 Nuremberg, Germany Tel.: Fax: Abstract: AlN with its high thermal conductivity allows to increase the current capability of power modules while maintaining a high insulation strength. The advanced power module design based on the pressure contact technology to establish the thermal contact between module and heat sink enables the implementation of AlN-substrates without any change in architecture. The results of end-of-life power cycling tests with advanced power modules using AlN, presented here for the first time, show that the lifetime can be improved by more than a factor of 2 compared to the Al 2 O 3 version. I. INTRODUCTION The market for power modules has a demand for a scalable output power of a module family, especially in the field of motor drives. This allows a line-up of various output power devices on the same application platform technology. While down-scaling can be achieved by reducing the implemented Silicon area, up-scaling is the challenge for the module designer. One option is the transition to a ceramic insulator with a higher thermal conductivity, for example to replace an Al 2 O 3 -DBC by an AlN-substrate. Due to the reduced coefficient of thermal expansion of AlN, this option has severe consequences for the module lifetime in a base plate system with a standard copper base plate. For system designs on the basis of the pressure contact technology without base plate, the implementation of different insulator materials is possible without any additional change of the construction and therefore represents a flexible module design. II. PRESSURE CONTACT TECHNOLOGY The pressure contact technology enables the design of a module without any base plate. The thermal interface to the heat sink is established by a multitude of pressure columns, which transmit the contact force from a pressure plate to the substrate. These pressure columns are connected by a flexible lateral structure in the so called bridge element. By positioning the pressure columns close to the chips, a good thermal interface can be achieved in those regions, where it is really important for the device performance. Fig. 1 shows the substrate layout and the bridge element with the pressure columns of the SKiM 5 module an IGBT sixpack module with 3 chips of 12.5 x 12.5mm size in the maximum configuration. The displayed version is the SKiM61GD126DM with 3 Trench IGBTs in parallel per switch on AlN- DBC substrates. Fig. 1: SKiM 5 substrate layout and bridge element with pressure columns AlN version with Trench IGBTs A pressure storage layer the spring pad is located on top of the bridge element and ensures an even distribution of pressure onto the pressure columns. This spring pad also minimizes the effects of local thickness variations of the substrate or tolerances in length of the pressure columns. It even permits the implementation of substrates with differences in the total thickness with negligible influence on the pressure system. Fig. 2 gives a cross section view of the SKiM 5 module, showing the pressure contact system in

2 PCB mounting positions PCB spring contact bridge element spring pad steel inlay pressure plate pressure columns substrate Fig. 2: Cross sectional view of the SKiM 5 pressure contact system and the snap-on features for the driver PCB assembly a mounted condition. By mounting the module on a heat sink, the pressure plate which is enforced by a steel inlay compresses the spring pad and thus applies tension to the pressure system. The pressure plate moves downward while tightening the mounting bolts until it reaches a mechanical stop supplied by the module frame and compresses the spring pad from the initial thickness of 4.5mm to a nominal remaining thickness of 2mm. Fig. 3 displays the spring characteristic of the spring pad. At the designed remaining thickness, the characteristic is still in the range of a nearly linear relationship between the remaining thickness and the force, thus making the system tolerable to small variations in thickness. An accelerated aging test was performed by a 1h storage at a temperature of 176 C. In compliance with the UL 1557 standard, a force [N] mm diameter 1h 176 C + 3h 25 C in SKiM 5 initial characteristic designed thickness remaining thickness [mm] Fig. 3: Spring characteristic of spring pad initial and after accelerated aging test 1h/176 C storage condition is used to evaluate the insulation integrity of the module for 3,h at 125 C. This high temperature storage beyond the application temperature range was followed by a room temperature storage for 3h of the mounted SKiM 5 module. The aged spring pad material shows a plastic deformation of approximately 1%, but also leads to a harder characteristics and even increases the pressure at the designed thickness. This flexibility of the system design enables the implementation of different substrates even if the total substrate thickness is not exactly the same. Table 1 shows the two options available today in the SKiM 5 module. The difference in total thickness of.7mm can be tolerated by the pressure contact system. Since no base plate is needed in the module design, the exchange of substrate material requires no changes in the module architecture. Table 1: Optional substrates for SKiM 5 module layer thickness Al 2O 3-DBC AlN-DBC Cu circuit side.4 mm.3 mm ceramic.5 mm.63 mm Cu bottom side.4 mm.3 mm total 1.3 mm 1.23 mm Fig. 2 also shows the snap-on features for the assembly of the driver PCB. The auxiliary contacts are designed as spring contacts and establish the electrical connections to the gates and auxiliary emitters and to the temperature sensor. This provides an easy assembly process and together with the optional driver board forms a modular intelligent power module, which can

3 be disassembled for system inspection. To secure a safe electrical contact even in vibrating environments, additional PCB mounting positions are supplied close to the contact positions for the mechanical fixture of the PCB by screws. III. POWER CYCLING TEST The active power cycling test is an important tool to characterize the lifetime of a power module under defined load conditions and therefore allows a comparison of different module designs. For the test, the module is mounted on a watercooled aluminum heat sink. A nominal DC load current is applied at the start of each cycle, heating up the device until a maximum temperature is reached at a selected reference position in the heat sink. This maximum temperature is adjusted during the setup of the test by measuring the heat sink temperature in the center of the chips, 2mm underneath the surface of the heat sink as described in [1]. The junction temperature is calculated using the specified data sheet value for the thermal resistance of the module. The duration of the heating phase is then adjusted, so that the nominal junction temperature reaches the desired? T j,nom. This procedure eliminates variations of the thermal resistance of the device under test and allows to determine comparable test conditions for a group of samples, that can be analyzed by statistical methods. After the maximum temperature is reached at the reference position, the load current is switched off. By applying a small sense current to the chips, the maximum junction temperature at the end of each heating phase can be recorded, using the temperature dependence of the forward voltage drop V CE (T). The forward voltage drop at load current level at the end of each V CE [mv] ? T j,nom =11K SKiM61GD126DM sample no. 4 total: cycles T j Bot T j Top V CE Bot Fig. 4: Power cycling test result at? T j,nom=11k (AlN version with Trench IGBT), V CE Top number of cycles T j,max [ C] heating phase is also recorded. In the cooling phase the device cools down to the start temperature. By activating the coolant flow only in the cooling phase, the cycle time can be shortened considerably. Fig. 4 shows a record of an active power cycling test of a SKiM61GD126DM, a sixpack module with 3 Trench IGBTs of 12.5mm x 12.5mm per switch on an AlN-DBC substrate. In this test, both switches of a single phase leg were driven in series. During the first couple of thousand cycles, the maximum junction temperatures decrease for both switches. The forward voltage drop decreases correspondingly, due to the positive temperature coefficient of the IGBT at nominal current. This typical feature of the pressure contact system is caused by a decreasing thermal resistance, which can be seen in detail in Fig. 5. thermal resistance R th(j-s) [K/W],1,8,6,4,2? T j,nom =11K SKiM61GD126DM sample no. 4 Thermal resistance junction - heat sink BOT IGBT TOP IGBT number of cycles Fig. 5: Decreasing thermal resistance at the start of the test (AlN version with Trench IGBT) Before the module is mounted a homogeneous layer of thermal grease is applied to the heat sink. When the pressure system is activated during the mounting process, the multitude of pressure columns is pressing the substrate to the heat sink and displace the grease material underneath the pressure column positions. In the areas between the pressure columns, excess thermal grease material is enclosed. The excess grease material has to be transported outward form the chip position through the small gaps left between the substrate and the heat sink at the pressure column positions. This flow process has a high time constant and is assisted by the relative movement of the substrate on the heat sink during thermal cycles caused by the thermal expansion of the system. For the module shown in Fig. 4, the displacement of thermal grease

4 under the chip positions is observed during the first 8, cycles and leads to a reduction of the initial thermal resistance of approximately 25%. After this initial thermal resistant decrease saturates, the maximum junction temperature at the end of the heating phase remains stable for more than 1, cycles. At approx. 11, cycles, a step in the forward voltage drop of the bottom switch indicates the first bond wire failures. The failed bond wires cause a higher resistance in the current path and increase the losses generated in the system. More steps in the forward voltage drop follow at 12, cycles and increase the resistance even more. The only moderate increase of the maximum junction temperature in this phase of the module life, resulting from the change of the local current distribution in the three parallel chips, leads to the conclusion that all three IGBTs are still contributing to the current transport. Otherwise, a more drastic increase of the thermal resistance would be expected. However the failure of bond wires induces a higher current density in the remaining bond wires and the unbalance of the current distribution leads to locally increasing temperatures, which cause more bond wires to fail in a positive feedback loop. Finally after 129, cycles, the bottom switch fails completely. IV. FAILURE ANALYSIS OF END-OF-LIFE TEST A failure analysis was performed to verify the interpretation of the data monitored during the power cycling test. The thermal resistance had shown no increase before the first bond wire failures were observed. A scanning acoustic microscope (SAM) analysis of the chip solder layer was engaged to investigate the thermal interface. Fig. 6 shows the SAM image of the failed bottom switch of the device tested in Fig. 4. In comparison to the diodes, that were not actively cycled, the IGBT solder interface shows a slightly lighter shading towards the diode positions, which could indicate some local fatigue in the solder interface, that is not jet effecting the thermal resistance of the IGBTs. But no severe fractures can be seen at the edges of the solder interface. The interfaces inside the AlN-DBC substrate were also inspected by SAM, but no trace of delamination could be detected. The ceramic layer was also scanned for fractures, but no deviation was observed. This result is in compliance with the test recording during the power cycling test, which showed no indication for an increasing thermal resistance. Since the bond wire failures were suspected to be the dominant failure mechanism, the IGBT Almetallization was inspected. A severe reconstruction of the emitter and gate contact areas was found as shown in Fig. 7. This reconstruction is caused by the difference of thermal expansion between the silicon bulk and the grains of the Al-metallization. If the maximum junction temperature during active temperature cycles is above 11 C, the stress in the Almetallization exceeds the elastic limit and causes grain boundary sliding and plastic deformation effects [2], which changes the surface structure of the metallization and thus destroys the contact interface to the bond wire. The comparison between the surface of the failed chip to the metallization of a chip from a phase leg not cycled shows this reconstruction effect. diodes IGBTs a) b) Fig. 6: Scanning acoustic microscope image of failed switch in Fig. 4 IGBTs in test, diodes not cycled Fig. 7: Reconstruction of IBGT emitter metallization a) Failed chip in power cycling test of Fig. 4 b) Chip from not tested phase of the same module

5 V. STATISTICAL ANALYSIS OF THE TEST RESULTS The power cycling test was performed on a group of 4 samples with the same nominal junction temperature swing. Different phase legs were cycled, but the monitored parameters showed in principle the same behavior for all tested modules. After an initial decrease of the maximum junction temperature due to the distribution of thermal grease, the value remained stable until the first bond wire failures occurred, indicated by steps in the forward voltage drop. All results are collected in Table 2. no. Table 2: Power cycling results for the SKiM 5 module? T j,nom [K] tested phase Failed switch Cycles to failure] 1 11 V Bot 86, W Bot 113, U Top 98, 4 11 W Bot 129, V Bot 223,5 AlN version with 12V Trench IGBT The four test results for the nominal junction temperature swing? T j,nom =11K were statistically analyzed. The resulting Weibull distribution is shown in Fig. 8. According to this analysis, the failure probability of 1% is reached after 53,3 cycles. probability density,25,15,5 Weibull-Distribution SKiM 5 PC? T=11 C AlN-DBC with Trench IGBT,2,1 5%: 17.2 kcycles 1%: 78.5 kcycles 1%: 53.3 kcycles? = 6.5? = 114 kcycles PC kilocycles probability density Weibull-distribution PC results Fig. 8: Weibull-analysis of power cycling test results An additional test was performed at a nominal junction temperature swing of? T j,nom =9K. This module failed after more that 22, cycles. VI. EVALUATION OF THE TEST RESULTS Since the power cycling test for? T j,nom =11K was performed on a group of 4 samples, a 1,8,6,4,2 accumulated probability statistical prediction of the end-of-life failure of the SKiM 5 module with AlN can be derived. For? T j,nom =9K, only one sample was tested. In a first order approximation, this test result was assumed to represent the 5% failure probability. With this assumption, the parameters of the lifetime prediction function known from the LESIT program [3] can be extracted. The LESIT results performed on standard base plate modules in the mid 9s stated the following relationship between the number of cycles to failure N f and the cycling temperature swing? T j. N f? A?? T? j? Ea? exp? k B? T (1) This lifetime function includes an additional temperature dependence on the medium test temperature T m, which was found to have a considerable impact on the lifetime of standard base plate modules. This medium temperature moderates a thermal activation process, described by the exponential function with the Boltzmann-constant k B, and an activation energy E a. Since the LESIT results represent only single tests, no statistical evaluation can be performed on the test data. Therefore, the lifetime curve was assumed to give the 5% failure probability in a first order approximation. The LESIT results are given for reference in the comparison displayed in Fig. 9. Since all tests reported here were performed with the constant minimum temperature of 4 C, the lifetime functions are calculated for this boundary condition. The statistical lifetime predictions for a failure rate of 1% and 5% are shown for the SKiM 5 module with AlN along with the previously published 1% probability for the SKiM 3 module with Al 2 O 3 substrates [4], which is also based on a pressure contact technology. The 23k cycles for this module was confirmed by test results on the Al 2 O 3 version of the SKiM 5 module with 21k cycles for the 1% failure probability. Compared to the lifetime of a base plate module, the pressure contact technology with Al 2 O 3 already improves the lifetime considerably. The AlN version increases the lifetime by a factor of 2 compared to the same module construction with Al 2 O 3 substrates. These results confirm the theoretical investigations of the influence of different thermal expansion coefficients on the reliability of power modules [5]. m??

6 1.. T j,min =const.=4 C number of cycles LESIT result SKiM3 (Al2O3) Weibull 1% SKiM5 (AlN) Weibull 1% SKiM5 (AlN) Weibull 5% delta T j [ C] Fig. 9: Results of power cycling end-of-life test for the SKiM 5 module with AlN and Trench IGBT comparison to the SKiM 3 module with Al 2O 3 and to the LESIT results for standard base plate modules. VII. CONCLUSION The SKiM series, a family of power modules based on the pressure contact technology, is a flexible module design that permits the implementation of different substrate materials without additional design changes. By replacing the standard Al 2 O 3 -DBCs with AlN substrates, an up-scaling towards higher current capability is possible. The power cycling tests on the AlN-version of the SKiM 5 module show an additional advantage in module lifetime, which was already predicted from the investigation of thermal expansion in the system. For typical application conditions, an extension of lifetime by a factor of 2 compared to pressure contact modules with Al 2 O 3 can be expected. Compared to the LESIT results on standard modules with Al 2 O 3 substrates and a copper base plate, the lifetime of the AlN pressure contact system is extended by a factor of 1 for a temperature swing of? T j =9K. VIII. REFERENCES [1] U.Hecht, U.Scheuermann: Static and Transient Thermal Resistance of Advanced Power Modules, Proc. PCIM 21, PC1.3, [2] M.Ciappa: Some Reliability Aspects of IGBT Modules for High Power Applications, Thesis ETH Zürch, 21. [3] M.Held, P.Jacob, G.Nicoletti, P.Scacco, M.H.Poech, Fast Power Cycling Test for IGBT Modules in Traction Application, Proc. Power Electronics and Drive Systems 1997, [4] U.Scheuermann: Power module design for HV-IGBTs with extended reliability, Proc. PCIM 1999, PC1.4, [5] U.Scheuermann, E.Herr: A Novel Power Module Design and Technology for Improved Power Cycling Capability, Microelectronic Reliability 41, 9-1 (21),