Narrow time-to-failure distributions indicate mature product Egon Herr and Steve Dewar, ABB Semiconductors AG, Switzerland The new Soft Punch Through (SPT) 1200V IGBT range in LoPak industry standard packaging has been qualified to the highest reliability standards. The methodology used gives a deeper insight into the useful life of the LoPaks in an inverter environment, allowing users to design systems with narrower margins than before. Fig.1 The LoPak range Fig. 2 IGBT and Diode Wafers and Dies PCIM Europe Magazine page 1 of 6 June 2001
LoPak A New Industry Standard The IGBT module market for inverter applications above 40kVA and up to 690V AC, and for traction inverters up to 750V DC, is served by a sometimes bewildering array of packaging styles. Manufacturers offer 1-packs and 2-packs with similar (but not quite identical) dimensions and layouts, and customers need to approve and design layouts for various different module packages in order to design a line up of inverter power and voltage ratings. ABB recognised the need for a new standard module range, which would offer customers a more rationalised option. LoPak4 and LoPak5 are six-pack modules with ratings from 150A to 300A in 1200V and 150A to 225A in 1700V. The choice of the 6-pack configuration and careful design of the SPT dies used in these packages allows parallel connection of the phase terminals of the modules so that, for example, a 300A 6-pack can be used as a 900A 2-pack. Furthermore these 2-packs can be further paralleled to achieve still higher current levels. In this way a line-up of 400-500V AC drives with power levels from 40kVA into the multi-mw range can be manufactured using only 3 LoPak part numbers. The LoPak Line-up Module 1200V 1700V LoPak3 75A 6-pack NPT and SPT 100A 6-pack NPT and SPT LoPak4 LoPak5 150A 6-pack SPT 200A 6-pack SPT 450A 2-pack SPT 600A 2-pack SPT 300A 6-pack SPT 900A 2-pack SPT 150A 6-pack SPT 450A 2-pack SPT 225A 6-pack SPT 675A 2-pack SPT One of the main aims of the LoPak4 and LoPak5 packages is to minimise the size and assembly time of the system. The devices themselves are extremely compact, however due to the high level of mechanical integration it is possible to manufacture an inverter where all components (excluding heatsink and link capacitors) fit within the envelope of the module itself. The power connections to the module have been inverted when compared to a traditional standard module, this allows a lower profile module and more compact module where, which in turn leads to shorter paths between the external power circuitry and the actual switching devices and therefore lower inductance. The modules are designed to take a PCB containing control and gate drive, which snaps on to the module, and is held in place by rugged hooks. The use of spring contacts eliminates the difficult process of soldering to a big module and produces a more reliable contact. The spring contact also allows the PCB to be removed should servicing be needed. A low-profile channel runs along the centre of the module long dimension, to allow for high PCB-mounted components, like capacitors and magnetics. The auxiliary connections are located as close as possible to the chips inside the module. This avoids unnecessary tracking inside the module, which would make it larger and therefore more costly than necessary. The user has the opportunity to place the low-impedance gate circuit loops of his drive circuit close to these pins, minimising the inductance between the real driver and the chips themselves. In this way the user can achieve gate circuitry with high noise immunity and optimise the switching of the device to his application. In terms of inverter assembly the LoPak4 and LoPak5 connections are made in the now famous and simple 2- step process: drop on the bus-bars, snap on the PCB. Completing the LoPak line-up is the LoPak3 module, which is compatible to the existing industry standard 75A and 100A 1200V packages. SPT (Soft Punch Through) IGBT The packaging benefits of the LoPak line-up are complemented by those of the new Soft Punch Through (SPT) IGBT die generation. SPT combines the losses of a trench device with the thermal resistance, ruggedness, and simplicity of a planar gate IGBT. Both switching and on-state losses are reduced by 20% compared to the current NPT devices. Since the thermal resistance is not increased, SPT offers more actual usable current per rated amp, so that system designers can make equipment, which is smaller, and more cost effective than ever before. PCIM Europe Magazine page 2 of 6 June 2001
Even with an on-state voltage of 20% less than the current NPT product, a strong positive temperature coefficient is maintained over almost the complete range of current. Furthermore the on-state characteristic is very resistive so that at low currents and high switching frequencies the benefit of SPT is even greater. The combination of positive temperature coefficient and the resistive on-state characteristic makes SPT technology very easy to parallel connect. LoPak Reliability Qualification Qualification Program The LoPak range is undergoing extensive reliability qualification testing. A complete set of qualification tests has been performed on LoPak3 and tests are still ongoing on LoPak4 and LoPak5. The power cycling capability is considered to be decisive for the useful life of an IGBT power module. The ABB qualification methodology is that such wear-out mechanisms must be investigated by driving devices into failure and then analysing time-to-failure using Weibull statistics. This gives reliable data to designers on the expected time-to-failure of 1% of a population of those modules. Such data is vitally important to the system design engineer, who can then calculate the useful lifetime as a function of the wear-out failure level that can be accepted from a system point of view. Reliability Qualification Tests for the LoPak Module Family Test Symbol Reference Load Conditions Requirement Qty. test samples High Temperature Reverse Bias High Temperature Gate Bias Temperature Humidity Bias HTRB IEC 60747-9 V CE = 80% V CES, V GE = 0 V, T j = 150 C HTGB IEC 60747-9 V GE = ±20 V, V CE = 0 V, T j = 150 C THB IEC 68-2-3 T c = 85 C, RH = 85%, V CE = 80 V, V GE = -15 V > 1000 hrs 8 > 500 hrs pos. > 500 hrs neg. > 1000 hrs 8 High Temperature Storage IEC 68-2-2 T c = 125 C > 1000 hrs 8 Low Temperature Storage IEC 68-2-1 T c = -40 C > 1000 hrs 8 Change of Temperature TC IEC 68-2-14 T c,min = -40 C, T c,max = 125 C, mounted module Intermittent Operating Life IOL1 IEC 60747-9 T j = 80 C, T j,min 40 C 20 sec t cycle 60 sec > 100 cycles 8 0.01 percentile > 70 kcycles Test until 5 modules failed 8 IOL2 T j = 30 C, T j,min 70 C 1 sec t cycle 3 sec Mechanical Shock IEC 68-2-27 a = 500 g, 1 ms, 3 shocks per direction Vibration (sinus) IEC 68-2-6 10 to 500 Hz, d < 0.35 mm, g < 5 g, 10 cycles per axis 0.01 percentile > 5 Mcycles Test until 5 modules failed 8 8 In the following, some of the available results shall be highlighted. Intermittent Operating Life (Power Cycling) T j = 30 C: Power cycling tests at T j = 30 C are running for the baseless version of LoPak3, which is using Al 2 O 3 as the substrate material. The test conditions are as follows: Currently all test samples have exceeded 17 Mcycles without failure. Therefore, the power cycling capability at low temperature swings is well above the requirements of the qualification plan. The tests will be continued until failure. Intermittent Operating Life (Power Cycling) T j = 80 C: IC 200 A Tj,min 80 C Tj,max 110 C ton 0.6 sec toff 0.7 sec PCIM Europe Magazine page 3 of 6 June 2001
End-of-life power cycling tests at T j = 80 C have been performed for the baseless version of LoPak3, which is using Al 2 O 3 as the substrate material. The test conditions were as follows: IC 72 A Tj,min 40 C Tj,max 120 C ton 23 sec toff 5.5 sec Five samples were tested into failure ( Vce > 10% or Rth > 20%) with the following results: Number of cycles to failure Sample 1 81.2 kcycles Sample 2 79.4 kcycles Sample 3 76.3 kcycles Sample 4 76.1 kcycles Sample 5 79.1 kcycles All failures that were encountered in the T j = 80 C test were analysed to understand the responsible failure mechanism. In all cases solder fatigue between the die and the substrate was the root cause of the failure (see Figure 3). Figure 3: Cross sectional view of die soldered on Al2O3 DBC after failing in power cycling. A crack in the solder layer can be seen on the left-hand side of the picture. Cracked solder layer IGBT die Alumina DBC Vibration Testing LoPak4: The test conditions were as follows: Frequency 10 Hz 500 Hz Sweep rate 1 octave / min Amplitude 5 g Axis 3 (x, y, z) Test duration 11 min 17 sec per axis Figure 5: LoPak4 Set-up for Vibration testing During the test the gate driver PCB was mounted and the continuity of the auxiliary contacts was continuously monitored. None of the spring contacts opened during the test even when some of the available mounting screws were not tightened. SPT Reliability Qualification With the introduction of the SPT chip we used the approach to qualify the SPT IGBT chip independent of a specific package. We used a simple test vehicle that can be considered as representative for many types of modules. With HTRB, HTGB and THB the 3 relevant reliability tests from a chip point of view were conducted using this test vehicle. The test conditions were the same as for the corresponding LoPak tests as given in the LoPak qualification program above. PCIM Europe Magazine page 4 of 6 June 2001
The reliability qualification for the SPT IGBT has successfully been completed and the chip set is released for mass production. In this way the SPT reliability is already qualified to allow it replace NPT types in the LoPak range and in other modules. How to predict the useful device lifetime Results of thermal or power cycling are often reported as data points in graphs with a temperature excursion T on the x-axis and a tested number of cycles N on the y-axis (normally in log scale). Assuming the applicability of a power law these data points are then connected through a straight line, which in turn is used either to extrapolate from test to field conditions or to do comparisons between competitors. We believe that this approach may lead to wrong expectations from a system reliability perspective if the failure statistics behind are not well enough understood. Thermal or power cycling tests are lifetime experiments addressing wear-out type of failure mechanisms (e.g. solder joint fatigue or wire bind lift-off). The relevant lifetime parameter to study is the number of cycles to failure N f, which by its very nature is a random variable. If we wish to analyze this random variable, we have to generate a certain number of failures (using a well-defined failure criterion) and then try to estimate type and parameters of an applicable statistical distribution law. At ABB the generation of five failures is the typical requirement, which we believe is a good compromise between statistical significance and duration respective cost of testing. It turns out that a Weibull type distribution law in most cases fits very well with the observations. Knowing the lifetime distribution one may calculate the number of cycles to a predefined probability of failure ε (the so-called ε -percentile). A proper choice for components seems to be at ε = 0.01, because from a system point of view a 1% probability of failure on component level would results in a 10% probability of failure on system level (assuming on average that a system would contain 12 such components in operation at any one time). In addition more than 1% failures under normal operation would certainly indicate the beginning of the long-term wear-out phase. So the interesting outcome of thermal or power cycling experiments is N f;1% = number of cycles to 1% probability of failure. We believe that N f;1% is a proper metric to characterize the capability of an item under certain cycled load conditions. The proposed statistical approach was used to evaluate the data from LoPak3 power cycling tests at Tj=80 C, for which the numbers of cycles to failure are given in 3.1 (see Figure 6). Figure 6: Weibull fit for end-of-life power cycling tests at T j = 80 C. The curved lines represent the maximum errors for a 90% confidence level. The 1% failure level (0.01 percentile) is predicted to be at 71.7 kcycles. It can be seen that the failures all occurred in a very narrow time interval. The scale parameter of the Weibull distribution (characteristic lifetime) at which the cumulated failure percentage has reached 63% (=0.63 percentile) is at 79.4 kcycles. The point at which the cumulated failure percentage has reached 1% (=0.01 PCIM Europe Magazine page 5 of 6 June 2001
percentile) is with 71.7 kcycles only slightly lower than the characteristic life. This is due to the high shape parameter of the Weibull distribution of β=45. Such a high value of the shape parameter is an indication of a mature technology and well-controlled manufacturing processes and materials. Summary The LoPak module family package design offers high reliability for industrial as well as traction applications. This is assured by an extensive reliability qualification program, which is executed before the product is released for production. A Weibull statistics approach was used for tests, which are considered to represent the most critical wear-out mechanisms in the field, e.g. power cycling. This generated data with a very tight distribution, indicating a mature product. In this way a system designer does not have to consider unknown safety margins for wear-out mechanisms, as is the case when only characteristic life or median life values are offered by the module supplier. PCIM Europe Magazine page 6 of 6 June 2001