Traction applications are a major driving force pushing IGBT module technology to higher demands on temperature cycling capability and general reliability improvements. Developments like AlSiC baseplates and AlN substrates that improve thermal management have aided in this progress. Other improvements include the bonding and baseplate technology as well as partial discharge immunity.
The Photo shows the three new housing designs of the 6.5kV module family. Because their footprints are identical to the existing baseplate standard sizes, you can easily adapt existing inverter designs for IHM and IHV modules to the new modules. A module redesign was essential because values for clearance and creepage distances needed adjustments to this new voltage. The modules meet standard EN50124-1 and fulfill the following classifications:
Overvoltage category OV2 — circuits not directly connected to the contact line and protected against overvoltages. Traction circuits protected by filters or inherently protected by components (e.g. semiconductors) are considered as corresponding to OV2 conditions.
Pollution degree PD3 — dust deposit → low conductivity caused by condensation; humidity → frequent condensation; location → indoor rooms, outdoor rooms protected from weather conditions; ventilation → natural ventilation, forced ventilation with clean, filtered air from outdoor.
Considering these operational conditions and rated insulation voltages of up to 4500V, a clearance of 26mm and a creepage distance of 56mm was chosen for the housing design.
For many years the reliability of the bond contact has been a concern, especially for traction applications. Considerable work has been concentrated on accelerated power cycling tests, analysis of failure mechanisms and improvements in material and bonding technology. Heel cracks, bond lift-offs, reconstruction of Al-metallization on the chips and corrosion of wires were step-by-step identified as reliability-limiting weak points. Development activities on composition of wires, shapes of bonding tools, bonding parameters, metallization of chips and leads and the introduction of protective coatings have led to considerable improvements in the reliability of the bond contact.
Solder, Substrate Technology
A vacuum process has made it possible to obtain a void free solder structure. This ensures a homogenous contact between substrate and baseplate and an appropriate thermal resistance.
Highest demands have to be fulfilled to cover the requirements on insulation capability and partial discharge behavior. The design of the insulation refers to the international standard IEC 1287.
To meet requirements of all applications, the modules must withstand a voltage between the electrical terminals and the baseplate for 60 sec (insulation voltage test) of:
UP = Test voltage
Um = Rated voltage, 6500V
The partial discharge level of below 10pC (Fig. 1) is given for a test voltage of:
These values are based on the worst-case conditions for 6.5kV-IGBT-modules in operation, which means that even a repetitive voltage peak of 6500V between any electrical terminal and the baseplate does not exceed the limits given by IEC 1287.
Detailed investigations done when qualifying the insulation materials showed that partial discharge (PD) measurements are of higher significance than the pure pass/fail insulation test itself. The PD behavior gives more insight of stability of the used materials. Fig. 2 shows a long term PD test (>10 min.) at the level of the insulation test voltage 10.2kV (instead of the standardized 1 min/5.1kV/10pC). Although the PD-level is slightly increased to 25pC, the material shows an excellent stability during the test.
The use of copper as baseplate material is common for its well-known advantages with regard to high thermal conductivity, easy mechanical handling, galvanic plating and adequate pricing. Disadvantages are nonreversible changes of mechanical properties above 300°C and the mismatch of the coefficient of thermal expansion (CTE) to the ceramic substrate. Due to this mismatch, thermal stress occurs between the materials and generates mechanical strain on the solder. Repetitive, heavy load cycling will create solder cracks and therefore an increase of the thermal impedance between chip and baseplate.
Efforts have been made to mitigate the bimetallic effect of the soldered system metal/ceramic by an adequate shaping. A machined convex bow is standard to improve the heat transmission between baseplate and heat sink.
The first four sections of Fig. 3 show the ultrasonic images of solder layers and ceramic substrates for modules with copper baseplates after 200 to 4000 temperature cycles. The progress of delamination in the solder layer due to the repetitive plastic deformation of the substrate edges is obvious. The root cause is the mismatch of the CTEs between Cu (17ppm/K) and AlN (7ppm/K).
A relatively stiff material with low deviation of its CTE to the ceramic could solve both described problems. Fig. 4, on page 42, shows the microscopic structure of a metal matrix compound (MMC) material: AlSiC, an aluminum infiltrated ceramic, offers extreme stiffness and a CTE close to that one of the AlN ceramic (7.3ppm/K).
The last section of Fig. 3 shows the result of 20,000 cycles for AlSiC baseplates, which are exclusively used for modules in the upper voltage and power range. By matching CTEs, no DCB delaminations occur between AlSiC and AlN even after 30,000 cycles. The failure mechanism no longer exists.
Cycling capability of modules is a major consideration. Thermal cycling specifies the material matching between baseplate and substrate in regard to their capability to sustain thermal stress while the “power cycling” measured by temperature excursions of the junction depicts the quality and lifetime of the bond wires.
Standard modules are designed for the needs of general industrial applications while the curves labeled “traction” represent modules that feature all the previously described improvements. They fulfill the severest specifications commonly required by traction applications or industrial applications with heavy load cycles. Traction modules are available in the voltage classes from 1700V up to 6500V.
Fig. 5 is published and confirmed by continuous qualification tests. Tests are performed by temperature swings that commonly reach a Tjmax of 125°C (e.g. the value for ΔTj = 40K is derived from cycles between 85°C and 125°C). The failure indicator is an increase of the forward voltage by 5%. The curves in Fig. 6 consider a safety margin, so that hardly any changes are observable in the given points.
For applications that do not reach 125°C, Fig. 6 only allows a worst-case estimate. To eliminate this lack of knowledge, a second set of measurements was performed testing temperature swings starting at a Tjmin of 25°C. This second curve is given in Fig. 6.
Further temperature levels in-between the cycling capability can be evaluated with the help of acceleration factors given in Fig. 7.
Example: what is the power cycling capability of a traction module for repetitive cycles of ΔTj = 50K from Tjmin = 50°C to Tjmax =100°C?
Fig. 6 in combination with Fig. 7 gives the following numbers:
0.95Mio (1Mio.=1 million) cycles for ΔTj=50K and Tjmax=125°C → operation at 100°C instead of 125°C (-25K) gives a deceleration by a factor of about two, resulting in 1.9Mio cycles.
4.7Mio cycles for ΔTj=50K and Tjmin=25°C → operation at 50°C instead of 25°C (+25K) gives an acceleration by a factor of 2, resulting in 2.35Mio cycles. Interpolation between the two results finally allows a prediction of roughly 2.1Mio cycles.
Fig. 5 only gives information about tests performed with temperature cycles of 30K and above. Assuming a duty cycle of 1 sec on and 1 sec off, which is necessary to activate the failure relevant time constants of the system, testing 10Mio. cycles already takes 231 days of continuous test bench operation. Further real-time testing with lower temperature swings and widely increased cycle numbers is unrealizable.
The Coffin-Manson model helps describe plastic deformation for power cycling at low temperature swings. This inverse power law predicts that higher stress at a shorter time interval provokes the same damage as lower stress for a longer time interval. The strain is given by the temperature cycles ΔT½ and the time corresponds to the number of cycles n½. The value for the exponent k is determined by respective investigations. Fig. 8 shows the model's prediction of cycling capability across the full temperature range. At 30K and above, we find a good correlation between the model and the tested points shown in Fig. 5, on page 42.
The improvements, receivable by changing the baseplate material from Cu to AlSiC, are documented in Fig. 9. The thermal cycling capability can be substantially increased, whereby AlSiC shows absolutely no indication of any delamination or weakening of the solder structure.
Power and thermal cycles are widely considered for load pulsations due to drive acceleration or braking. Additional investigations have been done to go further into the question of load pulsations not generated by the mechanical load itself but by low output frequencies of the inverter. A general tendency can be derived from Fig. 10, on page 46, which shows an example for measured and calculated upper and lower junction and case temperature excursions for different switching times at constant power dissipation.
While keeping the heat sink temperature at a stable level, the case no longer shows relevant excursions for switching periods below about 0.5 sec. Therefore, you don't have to consider low frequency cycles >2Hz for thermal cycling.
Due to the much smaller thermal capacity of the chips, the junction still shows temperature excursions at higher switching frequencies. As a rule of thumb, you can consider frequencies above 20Hz as no longer relevant. Because of the thermal capacity of the chips, the ΔTj has then dropped to 20K or below and several 100Mio cycles are expected before end of module life, a number that will not be reached in standard applications.
System considerations for highly reliable traction applications require a failure rate for IGBT modules in the range of 100FIT (1FIT corresponds to one failure in 109 hr of operation). To confirm this value by explicit measurements, more than 1100 modules would have to be tested 365 days for 24 hours. However, realizing such an experiment is unlikely because of cost and capacity reasons.
By tracing back shipped modules, estimating accumulated operation hours and surveying field returns for the last few years, we can calculate the tendency in the development of failure rates as shown in Fig. 11. The FIT rates are evaluated from all devices in the field and all rejects.
On the basis of some 100,000 modules in the field and more than five years of field experience, we see the results of continuous improvements leading to rates well below 100FIT for high power IGBT modules.
T. Laska, L. Lorenz, A. Mauder, “The Field Stop IGBT Concept with an Optimized Diode,” PCIM 2000 Nuremberg.
H. Berg, M. Hierholzer, T. Schütze, “Further Improvements in the Reliability of IGBT Modules,” IAS St. Louis 1998.
For more information on this article, CIRCLE 333 on Reader Service Card