Selected failure mechanisms of modern power modules
Introduction
Thanks to the recent improvements made in handling large currents at high voltage and at high switching frequencies, insulated gate bipolar transistors (IGBT) have almost completely replaced bipolar power transistors (BPT) and they are challenging the position of gate turn-off thyristors (GTO) in their traditional fields of application. In the last five years, the need of increase the reliability of high-power IGBT multichip modules has been one of the most powerful drivers that forced engineers to design new products, especially intended for traction, for power transmission, and for power distribution applications. In order to cope this demand, various wide-band research projects, which involved both industry and academia, have been promoted on an international base. This is for example the case of the high-power semiconductors for railway traction applications (RAPSDRA) project and of the high-temperature IGBT- and MOSFET-modules for railway and automotive electronic applications (HIMRATE) project, both funded by the European Union. One among the most relevant goals of such research programs is the definition of realistic reliability requirements based on dedicated mission profiles. These studies confirmed that the reliability figures of merit required from power devices for railway traction are much more stringent than those needed by the components intended for the usual microelectronic applications. As an example, the useful lifetime of an IGBT module for railway traction is specified in at least 30 years, while the failure rate of every single IGBT module is not allowed to exceed 100 FIT. Under previous assumptions, the traditional reliability growth programs based on a posteriori failure rate assessment procedures are unuseful, as it is already well known from similar experiences with ULSI devices. In fact, even a simple characterization of the failure rate as function of the time with a reasonable level of confidence, would require many millions of cumulated component hours, what is almost impracticable in the case of complex devices and in particular for power modules (even at accelerated conditions). The strategy used for circumventing this limit bases on the well-known concept of built-in reliability. The main idea behind this approach is the continuous control of those process parameters, which may affect the reliability of the final product. This initial phase is followed by dedicated experiments and characterizations, which are intended to investigate the system response over the variation of a given parameter. Finally, the obtained information is returned into a feedback loop for finely tuning the process conditions. A very important step when realizing this strategy is the characterization and the classification of the failure mechanisms, which occur either during accelerated tests or in field applications. While wearout failure mechanisms can be attacked by adequate design rules (and represent essentially a cost optimization problem), random failures are not necessarily related to a dominant failure mechanism. In fact, they express the random character of both the occurrence of physical processes and of the quality of manufacturing processes. Nevertheless, random failures play a very relevant role in determining the survival probability of a mature system.
In the following, we will shortly review the most frequent failure mechanisms observed to affect mainly IGBT modules. However, since the majority of the failure mechanisms listed here are package related and are driven by thermo-mechanical stresses, they can also be encountered in power modules using devices (e.g. power MOSFETs or diodes), which exhibit high densities of power dissipation.
Section snippets
Module architectures
Multichip modules for high-power IGBT devices are complex multilayered structures consisting of different materials, which have to provide a good mechanical stability, good electrical insulation properties, and good thermal conduction capabilities. The schematic cross-section through a module of type A (e.g. a standardized E2 package) is represented in Fig. 1a, and the related physical parameters are listed in Table 1. Starting from the bottom one can recognize the base plate, the direct copper
Materials and thermomechanics
When considering thermal cycling of these multilayered structures and the consequent thermo-mechanical fatigue induced failure mechanisms, it is important to take into account all the factors, which play a role in determining thermo-mechanical stresses.
In first approximation, there are the mismatch in the coefficient of thermal expansion (CTE), the characteristic length of the layer, and the local temperature swing. The first two parameters for the relevant materials and layers in a multichip
Bond wire fatigue
Multichip IGBT modules for high-power applications typically include up to 800 wedge bonds (Fig. 3). Since about half of them are bonded onto the active area of semiconductor devices (IGBT and freewheeling diodes), they are exposed to almost the full temperature swing imposed both by the power dissipation in the silicon and by the ohmic self-heating of the wire itself. Emitter bond wires are usually 300–500 μm in diameter. The chemical composition of the wire can be different from manufacturer
Aluminum reconstruction
Although reconstruction of the aluminum metalization is an effect, which has been encountered since the early times of microelectronics [11], [12], the occurrence of this degradation mechanism in IGBT multichip modules has been firstly reported in [13], [14].
During thermal cycling of IGBT devices and of freewheeling diodes, periodical compressive and tensile stresses are introduced in the thin metalization film by the different CTEs of the aluminum and of the silicon chip. Due to the large
Brittle cracking and fatigue crack propagation
The brittle materials used in advanced IGBT multichip modules are the single crystal silicon, the thin insulating layers on it, and the ceramic substrate. One among the main assumptions in fracture mechanics of brittle materials is that the sharp stress concentration at pre-existing damages leads to the rupture under the influence of external mechanical stresses. Ultimate brittle fracture can occur suddenly without any plastic deformation, when an initial crack is present, whose length exceeds
Corrosion of the interconnections
Corrosion of aluminum is a well-known failure mechanism since the early times of microelectronics. When pure aluminum (e.g. bond wires) is exposed to an oxygen containing atmosphere, a thin native Al2O3 surface layer is grown that passivates the metal. Aluminum is self-passivating also in pure water, where the native aluminum oxide is converted into a hardly soluble layer of aluminum hydroxide Al(OH)3. When exposed to other solutions, aluminum hydroxide is amphotheric, i.e. it is dissolved both
Solder fatigue and solder voids
A main failure mechanism of IGBT multichip modules is associated with the thermo-mechanical fatigue of the solder alloy layers. The most critical interface is represented by the solder between the ceramic substrate and the base plate, especially in the case of copper base plates [19]. In fact, at this location one finds the worst mismatch in the CTEs, the maximum temperature swing combined with the largest lateral dimensions (see Table 1 and Fig. 3). Nevertheless, fatigue phenomena occurring in
Burnout failures
Device burnout is a failure mode, which is very frequently observed either as the final act of wear out, or as consequence of a failure cause occurring randomly. Burnout is often associated with a short circuit condition, where a large current flows through the device (or through a portion of it), while it is supporting the full line voltage. Sustaining a short circuit over a long time interval inevitably leads to thermal runaway and finally to a fast destruction of the device. In fact, since
Conclusions
Electric traction has provided a particularly demanding application domain where severe thermal and power cycle constraints are engendered by the manyfold application profiles. The increased demand in terms of device performances and reliability has to be faced concurring with the continuous trends towards cost reduction. In the past, the selection of power modules was primarily based upon current and voltage ratings, only. Today's high-power applications have to take in the same consideration
Acknowledgements
I would like to thank Paolo Malberti for his continuous technical advise and Giovanni Nicoletti and Paolo Scacco for preparing some among the samples presented in this paper.
References (32)
- et al.
Thermal characterization of IGBT power modules
Microelectron. Reliab.
(1997) - et al.
Crack mechanism in wire bonding joints
Microelectron. Reliab.
(1998) - et al.
Reliability and lifetime evaluation of different wire bonding technologies for high power IGBT modules
Microelectron. Reliab.
(1999) - et al.
Substrate-to-base solder joint delamination in high power IGBT modules
Microelectron. Reliab.
(1997) - et al.
Modelling thermal effects of large contiguous voids in solder joints
Microelectron. J.
(1999) - et al.
Static and dynamic finite element modeling of thermal fatigue effects in IGBT modules
Microelectron. Reliab.
(2000) - et al.
Reliability of AlN substrates and their solder joints in IGBT power modules
Microelectron. Reliab.
(1999) - et al.
Extrapolation of cosmic ray induced failures from test to field conditions for IGBT modules
Microelectron. Reliab.
(1998) Cosmic ray induced failures in high power semiconductor devices
Microelectron. Reliab.
(1997)- et al.
The plasticity and creep of metals and ceramics
(1982)