YVAN AVENAS,

LAURENT DUPONT,
NICK BAKER,
HENRI ZARA,
and FRANCK BARRUEL

P
ower conversion systems
are dependent on the per-
formance and reliability
of static converters. How-
ever, they are subject to
frequent functional and
environmental strains, which can
induce failures. The anticipation of
these failures is difficult but impor-
tant so the operation of a system can
be halted before a breakdown occurs.
In the case of photovoltaic (PV) power
plants, the system can be simplified
into two distinct blocks: the solar pan-
els/modules and the power inverter. A
breakdown in either of these blocks
can cause significant downtime in the
system. Nevertheless, multiple solar
module failures can often be toler-
ated and not lead to a total breakdown
of the entire array, whereas a single
component failure in the inverter can
lead to a collapse of the entire sys-
tem. Furthermore, solar panel/module
manufacturers often offer warranties
of up to 20 years, while warranties
for inverters rarely reach the ten-year
mark [1]. As such, the overall cost of a
PV inverter can increase by a factor of
two or three if it undergoes one or two
failures during the life of the system.
Although the market competition be-
tween PV inverter manufacturers has
©artville, llc

Condition Monitoring
A Decade of Proposed Techniques

Digital Object Identifier 10.1109/MIE.2015.2481564

Date of publication: 21 December 2015

22  IEEE industrial electronics magazine  ■  December 2015 1932-4529/15©2015IEEE

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 traditionally focused on the efficiency
of their product, a failure that induces
User
a downtime of just a few days can eas-
ily negate the yield attained through a Life Status Action to Be Taken Maintenance
1% efficiency improvement. Oversiz-
ing the inverter or introducing redun-
dancy into the inverter system is one Condition-
Prognostic
option to improve reliability; however, Unit
Monitoring Sensors:
often, this is not economical. Unit
State Variables
A 2011 industry-based survey [2]
found 93% of respondents placed reli- Diagnostic Unit Health
Power Converter
ability as a primary issue in the field Failures
of power electronic converters. More
than 50% believed that condition- Power Gate Drivers
monitoring (CM) tools should be im- Devices Control Unit
proved. The survey also indicated Control System
that semiconductor devices are often
the weakest link in converters, com- User
prising 30% of total failures. This is fol-
Action to Be Taken
lowed by capacitors and gate drivers Fixing
at under 20% and 15%, respectively.
In 2010, Yang et al. [3] classified FIGURE 1 – The complete health-monitoring system.
health-monitoring techniques for
power converters into three different
categories: (IGBT) that underwent catastrophic failures. Like failure indicators,
■■ Diagnosis—Identifying the root short-circuit failure is shown. On the these could also be integrated in the
cause of failure contrary, ­Figure 2(b) displays an IGBT gate driver.
■■ Prognosis—Predicting the health that experienced a short-circuit failure In Figure 1, the goals of the different
of a component at some point in when a protection circuit in the gate units in the control system are comple-
the future driver was present [4]. mentary. Through the measurement of
■■ CM—Real-time measurement of For CM purposes, however, the failure indicators, the diagnostic unit
the current condition of a compo- sensors should monitor damage indi- can select a suitable control strategy.
nent or a system, and taking appro- cators that signify gradual wear-out This can, for example, be to operate
priate action. of the device rather than short pe- the converter in a degraded mode, or
Figure 1 presents a schematic of riods of overstress. These can then to halt converter operation [5]. The CM
what could be considered a complete be used to assess the current device unit processes measurements of dam-
health monitoring system for a power condition. Nevertheless, it should be age indicators (signals that indicate de-
converter. The two main parts are noted that gradual wear-out can also vice degradation). In the case of a drift
the sensors and the control system. lead to similar types of catastrophic away from healthy conditions, the user
Sensors understandably include the
necessary voltage or current measure-
ments for normal converter operation,
common in many power electronic
converters. Additionally, sensors
would be present to monitor signals
that indicate failures or critical con-
ditions that could cause catastrophic
failure. Typically, the monitoring of
these failure indicators can be pro-
posed in the gate driver. A common ex-
ample today is a VCE monitoring circuit,
which is often present in commercial
gate drivers and prevents short-circuit
breakdown of the devices. Figure 2
(a) (b)
displays the benefits of detecting such
failures. In Figure 2(a), the aftermath FIGURE 2 – An example of driver protection. (a) a catastrophic IGBT failure caused by an uncon-
of an insulated-gate bipolar transistor trolled short circuit, and (b) an IGBT chip after short circuit and the action of gate [4].

December 2015  ■  IEEE industrial electronics magazine  23

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 is notified, and the appropriate action
can be taken. In other cases, the prog-
Intermittent misfiring of the gate driver can also
nostic system can, through lifetime
models, estimate the remaining life lead to catastrophic failures or repeated short
of the device and predict its health at
some point in the future [6].
periods of overstress.
This article is dedicated specifically to
the presentation and analysis of CM sys-
tems for power semiconductor modules. a traditional wire-bonded power module driver can also lead to catastrophic
In the past, these systems have generally along with some of the common failure failures or repeated short periods of
been confined to laboratory conditions. locations. These failures are primarily overstress [15].
This article also attempts to focus on tech- attributed to the differences in the coef-
niques proposed for operation in condi- ficients of thermal expansion (CTE) of Damage Indicators
tions that are close to real operational the different materials in the chip and A CM system requires the measure-
power converters. Since 2010, a large package construction, and the tempera- ment of damage indicators to estimate
amount of research has been conducted ture swings they experience. These tem- the current aging status of the power
in this area. Therefore, this article focuses perature swings are induced by both the semiconductor device. A large number
on new literature that has been available environment and the application (mis- of indicators have been well known
since the review performed by Yang et sion profile). since the end of the 1990s. These are
al. [3]. In 2015, Oh et al. [7] proposed a One of the largest mismatches in “classical indicators” and are briefly
review of IGBT CM and prognostic prin- CTE is between the semiconductor detailed in the following paragraphs.
ciples and related physics-of-failure. This chip (silicon), and the bondwires and Nevertheless, the majority of these
article is therefore more concentrated on surface metallization (aluminum). indicators are dedicated to character-
implementation issues. This is exacerbated as one end of the izing power devices when they are not
bondwire is bonded directly onto the integrated in a real power converter.
Power Module Failure active area of the chip surface and is This is the reason why, since the end
Mechanisms subject to large temperature swings of the 2000s, a substantial amount of
Before introducing the main damage in- due to the power dissipation in the new research and new indicators have
dicators for CM, a brief reminder about chip as well as the ohmic self-heating been proposed in scientific literature.
some common power module failure of the bondwire. Repeated thermal These newer indicators will also be de-
mechanisms will be provided. Press- cycling then causes stress that leads tailed in this section.
pack high-power devices, which have to one of the most common failure
a higher reliability than wire-bond de- mechanisms: bondwire fatigue, and Classical Indicators
vices, are not discussed in this article. eventually bondwire failure or lift-off Table 1 presents a simplified cross
Power module failure mechanisms from the semiconductor chip itself. matrix of some classical damage in-
are generally separated into two cat- The failure of a single or small dicators and their related degrada-
egories: chip-related failure mecha- number of bondwires can alter the tion modes for IGBT devices [10]–[12],
nisms and package-related failure current distribution in the module [16]–[22]. Because of the large number
mechanisms [8]. Wu et al. [9] defines and increase the ohmic resistances. If of publications dealing with this area,
chip-related failure mechanisms into operation of the IGBT is not halted be- this survey is therefore not exhaus-
two further categories: short-circuit fore a suitable threshold, catastrophic tive. However, the list is sufficient to
and open-circuit failures. Many of failure can occur that can manifest note that we can observe that a dam-
these failures appear to be due to over- itself in a number of ways. This could age indicator (for example, the for-
stress conditions and hence outside of include the melting of remaining bond- ward voltage drop) may be influenced
the realm of CM for wear-out fatigue. wires, thermal runaway in paralleled by numerous types of degradation,
Nevertheless, the likelihood of occur- chips, or triggering of parasitic compo- and, conversely, a single degradation
rence for some chip failure mecha- nents that lead to device destruction. mechanism may influence several
nisms can be increased through the Outside of these categories, some damage indicators. Therefore, it is like-
accumulation of thermal fatigue. system-level degradations can also ly that the use of a single parameter to
The failure mechanisms that are have a large impact on power semi- monitor degradation is insufficient to
most frequently observed in power conductor health. For example, Perpi- develop a complete CM system.
devices are due to the thermomechani- na et al. [10] show how thermal grease Many of the degradations of pow-
cal stress experienced by the packag- displacement can occur during ther- er modules that are presented in
ing materials. These mechanisms are mal cycling, leading to a degradation Figure 3 can readily be observed un-
fairly well known for traditional power in the thermal conductivity between der laboratory conditions using clas-
module packaging technology. Figure 3 the heat sink and the power mod- sical damage indicators. For example,
[10]–[14] displays a typical structure of ule. Intermittent misfiring of the gate reconstruction of the aluminum

24  IEEE industrial electronics magazine  ■  December 2015

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 Direct Bonded Copper Substrate
Ceramic Crack
500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000

Thermal Interface
Bondwire Lift Off

Bondwire Lift-Off

Down
Down

FIGURE 3 – The main degradations observed in IGBT power modules after aging tests [10]–[14].
Number of Cycles (k–)

Bondwires

Up

Up
M3

M3
Thermal Grease Degradation
High Control IGBT

Baseplate

Heat Sink
Semiconductor
Chip Surface Reconstruction

Chip Chip Solder

M4

M4
0

Solder
2.15
2.145
2.14
2.135
2.13
2.125
2.12
2.115
2.11
2.105
2.1

Baseplate Solder

310 µm
Voltage VCE (V)
Ceramic Substrate (AIN)

Plating Base Plate (AISiC)

Solder Fatigue

Copper (DCB)
Solder

Nickel-
Crack

surface metallization increases the a bondwire lifts off from the IGBT, as of the power module to remove heat from
ohmic sheet resistance of the device. the remaining bondwires are forced to the semiconductor and into the heat
As such, this can generally be viewed carry a higher current load. sink. An increase in the thermal resis-
as a steady increase in the on-state On the other hand, degradation in the tance generally indicates the presence
voltage drop of the IGBT as the ther- form of solder or ceramic cracks in the of this type of degradation. The thermal
mal cycles accumulate. On the other power module structure can be viewed resistance necessitates the knowledge
hand, sharp step increases in the through an evaluation of the thermal re- of the junction temperature, which is
voltage drop can often be seen when sistance. These cracks lessen the ability generally made using a thermosensitive

December 2015  ■  IEEE industrial electronics magazine  25

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 parameter (forward voltage under low paralleled devices. It is unclear wheth- is characterized, implemented, and mod-
current—several mA) [23]. er this is a realistic degradation condi- eled by Ginart et al. [30]–[33]. In these
Several alternative methods have tion for devices under real-life stresses. studies, the aging methodology consists
been proposed for damage evaluation A variation of the collector–emit- of the following steps.
in power modules. These methods use ter voltage transient during switching ■■ IGBTs are aged by thermoelectrical
indicators that are closely related to is also observed when wire bonding stress until latching was observed.
operational conditions and will now [26] or thermal path [27] are degraded ■■ Once latching occurs, the transis-
be outlined in the following sections. [Figure 4(c)]. In the first case, the vari- tors are turned off for several min-
Principally, these indicators are the ation can be attributed to an increase utes (recovery period).
switching transients or parasitic ring- of the stray inductance [25]. The sec- ■■ Both steps are repeated until
ing variations in the device. Several ond case is attributed to a tempera- changes are observed in the nomi-
authors have proposed to use case tem- ture increase induced by the thermal nal operating parameters.
peratures or other indirect methods. path degradation by [28] and [29]. Figure 5 shows the evolution of
These damage indicators appear to voltage ringing during switching for
Switching Transients be in the preliminary stages of investi- different health states. The ringing
Research has shown that the majority of gation—to use these indicators in CM appears to be clearly sensitive to ag-
the switching properties of power semi- systems, a critical degradation level for ing. This phenomenon is explained by
conductors fluctuate with aging. Their each must be investigated with the help increased losses in the gate capaci-
use as damage indicators are appealing of aging studies. Furthermore, they are tances of the IGBT chips [31].
since switching properties can potential- all temperature dependent and suffer Like the switching properties,
ly be viewed while the device is in opera- from the same issue as the classically these ringing techniques perhaps re-
tion. However, with present knowledge, used forward voltage as they depend quire more clear and robust specifi-
it may be difficult to distinguish if the on more parameters than just a single cation for critical degradation levels,
degradation in these properties is due to degradation mode. Additionally, many and a sensitivity analysis toward oth-
the semiconductor or the gate driver if of these parameters are also dependent er parameters in a complete system.
applied to a functioning converter. on the health of the gate oxide. Thus,
For example, in the case of a power knowledge of the junction temperature Case Temperature Measurements
module with multiple paralleled IGBTs, is necessary to use these damage indi- Several authors have proposed a non-
a severe degradation of wire bond- cators during the operation of the con- electrical method in a measurement of
ing can induce a reduction of the gate verter. It could be interesting for future the case temperature to estimate the
capacitances Cgc and Cge. If one chip research to characterize these addi- degradation of the device.
is completely disconnected or fails tional influences and validate whether Tian et al. [34] characterize a power
due to total bondwire lift-off, the gate or not they induce non-negligible fluc- module by integrating one thermocouple
capacitance reduces and induces a tuations in comparison to the fluctua- under each paralleled die. In this exam-
notable variation in the gate current tion caused by degradation. ple, there are nine dies in parallel to form
(Figure 4(a) [24]) and of the gate–emit- a single IGBT switch. Figure 6(a) shows
ter voltage (Figure 4(b) [25]) during Current or Voltage Ringing that, before aging, the temperature under
switching. Nevertheless, these studies In relation to the previous para- each die is reasonably uniform. However,
perform an artificial aging of the power graphs regarding switching transients, after aging the temperature between the
module by completely cutting all the degradation can be viewed through volt- hottest and the coldest points is close to
emitter-connected bondwires on one age and current ringing during the opera- 7 °C (to simplify the understanding, we
chip inside a power module with two tion of the converter. This phenomenon only present four thermocouple temper-
atures in this figure). This unbalance is
thought to be due to several wire bond-
TABLE 1 – THE MAIN DEGRADATION MODES OF IGBT DEVICES AND THEIR INDICATORS. ing degradations, which induce a current
DEGRADATIONS
disequilibrium between the chips. Even
though this investigation revealed some
WIRE
EMITTER BONDING interesting observations, the use of nine
METALLIZATION LIFT-OFF OR HEAT PATH GATE OXIDE thermocouples per switch is question-
INDICATORS RECONSTRUCTION FATIGUE DEGRADATION DEGRADATION
able in terms of real-life implementation
Forward voltage with high
current
[10], [16], [11] [17], [12], [18] [22] and cost.
Xiang et al. present a more com-
Thermal resistance or
[10], [19], [20] plex method based on the measure-
impedance
Threshold voltage [11], [21]
ment of the case temperature [35].
The goal is to estimate the power
Gate leak current [16]
loss variations, which are due to
C(V) characteristic [22] the degradation of the thermal path

26  IEEE industrial electronics magazine  ■  December 2015

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 Variation of Switching Properties

Gate Current Gate-Emitter Voltage Collector-Emitter Voltage

(Turn On) (Turn On) (Turn Off)

1.2 180
160
1.0
140
0.8
120
Before Aging
0.6 100
Before Aging
80
0.4
60

Gate Current (A)

After Aging
0.2 40
20

Collector-Emitter Voltage (V)

0.0
After Aging 0
–0.2 –2.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time (µs)
16 Time (µs)
14

12
After Aging
10
8

6
Before Aging
4
2

Gate-Emitter Voltage (V)

0

–2
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Time (µs)

FIGURE 4 – The switching transients as damage indicators [24], [25], [27] (simplified waveforms).

December 2015  ■  IEEE industrial electronics magazine  27

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 and seventh harmonic of the output
100 current. They show a reduction of the
90
fifth harmonic amplitude (in the range
Before Aging from 3 to 5.5%) due to the variation
80
of the transistor switching behavior
70 with temperature [38]. To increase
60 this relatively low sensitivity, an origi-
Vce (V)

nal method is proposed. A controlled

50
harmonic resonance is introduced
40
into the inverter output circuit. This
30 amplifies the fifth harmonic for easier
20 After Aging detection and an outer control loop
suppresses the output current har-
10
monic. The compensation voltage
0 produced by this control loop is then
0 100 200 300 400 500 600
Time (ns) used as the CM signal.
The use of global methods is of great
interest because, generally, no addi-
FIGURE 5 – The VCE voltage ringing after IGBT turn off for different health states [33] (simplified
waveforms). tional sensors have to be implemented.
In particular, the method based on the
fifth harmonic variation seems to be a
of the power module. As shown in an accurate electrothermal model is good way to imagine new CM systems.
Figure 6(b), the case and ambient also an issue in this area. However, specific tools have to be de-
temperatures are recorded with time. veloped to define which specific device
With an electrothermal model, the au- Global Methods in the system is damaged.
thors are able to estimate the power There are some global methods that Another example is presented by
loss in the converter and, therefore, have been proposed by a variety of Mohagheghi et al. [39] who recommend
monitor its variation during aging. authors. The overall efficiency of the use of neural networks. This neural
The main issue of this method is the the converter, for example, has been network estimates the output voltage
low variation of the case temperature proposed for degradation monitor- of a diode rectifier from given values
with aging (around 1 °C), which is dif- ing [36]. In the case of a power train, of input voltage and current, as well as
ficult to observe. The requirement of Xiang et al. [37], [38] measure the fifth the output current. Any changes in the

Case Temperature Monitoring

Temperature
Loss Variations
Nonuniformity

35
Case Temperature Ambient Temperature
30 Measurement (Tc) Measurement (Ta)
Temperature (°C)

25
20 Before Aging
15 After Aging Thermal Model of the Heat Sink

10
5
Estimation of the Power Loss
0
1 2 3 4
Temperature Sensor Number
Evaluation of the Health of the
Power Module

FIGURE 6 – The case temperature measurements [34], [35].

28  IEEE industrial electronics magazine  ■  December 2015

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 characteristics of the circuit compo-
nents compared to their original condi-
The forward voltage is consistently measured
tion will cause the output of the circuit
to deviate from the values estimated by to estimate the degradation of the wire bonding
the network. This deviation serves as
a signal to the user that part of the cir-
and upper metallization of the chip.
cuit is not operating as expected. In that
sense, artificial intelligence techniques In the 1990s and 2000s, power and/ interruption to converter operation,
can be very useful for CM. The authors or thermal cycling test benches were which leads to a relatively complex
in this study suggest that this concept not fully representative of real elec- monitoring system for this degrada-
can be extended to distinguish between trical operating conditions and did tion parameter [47].
different components in the converter, not include device switching. Instead,
providing that microlevel monitoring they attempted to solely simulate the Determining the Critical
was implemented with measurement thermomechanical stress, which led Level of Degradation
points installed in various locations to power module damage, particular- Arguably the most complex task in a
within the system. The measured and ly, railway applications were of partic- CM system is the determination of the
estimated signals could then be fed to ular interest [41]. This approach has critical levels of degradation. These
a neural network and, by proper train- since expanded to become standard are very dependent on the application
ing, the health of the system could be in evaluating the thermomechanical and the power module technology.
monitored. robustness of power modules [11]. The determination of these levels is
In recent years, however, many test challenging and requires a large num-
Implementation Issues benches have begun incorporating ber of aging tests for given technology
of CM Systems more realistic electrical conditions. and mission profile.
CM necessitates the measurement of Specifically, the inclusion of real-de- For example, bondwire degradation
one or more damage indicators dur- vice switching during power-cycling is often determined by an increase of
ing the life of the converter. However, tests is now reasonably common and the forward voltage Vfwd in an IGBT or
simple knowledge of these indicators used by several research teams [42]– diode. However, several investigations
is not sufficient to form a CM strategy. [45]. These test benches are generally have resulted in differing proposals for
In fact, a damage threshold must be de- based on the opposition method [46]. a critical level of Vfwd. In fact, increas-
fined to warn the user of a critical deg- es of 5% [48], 10% [12], or 20% [16] of
radation level. There are three stages Establishing and Measuring the device forward voltage under high
in the implementation of a CM system: Damage Indicators current have been specified. In other
■■ specification of the key degrada- Typically, a damage indicator exists for literature, more complex criteria are
tion mechanisms each known degradation mechanism given, i.e., 1,500%/Ic,rated, where Ic,rated
■■ establishing and measuring the of standard power module technology. is the rated current of the device [49],
correct damage indicators for each Nevertheless, one damage indicator or 20%/ncp, where ncp is the number of
degradation mechanism can be influenced by several types of chips in parallel [18]. This is further
■■ determining when a damage indicator degradation (Table 1). For example, the complicated since Vfwd has also been
displays a critical level of degradation. variation of the forward voltage is af- seen to decrease during aging [22].
fected by temperature (corresponding This decrease is perhaps due to other
Specifying the Most Key to heat path degradation), gate oxide failure mechanisms being dominant in
Degradation Mechanisms integrity, the quality of electrical con- the device during these aging tests.
The specification of degradation nections (including bondwire lift off), As a consequence, without a thor-
mechanisms is particularly dependent and metallization reconstruction. ough knowledge of the device technol-
on the technology and the operating A further issue is the difficulty in ogy and application, there is no general
conditions. For example, gate oxide measuring these indicators in power rule that can be applied to design a CM
degradation is a major concern in SiC converters. For example, the measure- strategy for a power electronic con-
metal-oxide semiconductor field-effect ment of the thermal resistance in a verter. As such, there is a large amount
transistors (MOSFETs) operating at power module requires an estimation of recent literature that deals with the
high temperature [40], whereas it is of- of the chip temperature. This is gen- implementation of CM systems in pow-
ten ignored in IGBT devices in railway erally made with a thermosensitive er electronic converters.
applications [10]. To specify the main electrical parameter (TSEP), specifi- Since the end of the 2000s, a signifi-
realistic degradation mechanisms, ac- cally by an injection of a low current in cant number of articles discussing the
celerated aging tests that apply realis- the device and a measurement of the CM of actual inverters with damage
tic mission profile-based stresses to the subsequent voltage drop. In a func- indicators has been written. The num-
device should generally be used (ide- tional system, this low-current source ber of measured indicators, however,
ally in conjunction with real field data). and the measurement system require is generally quite limited. For instance,

December 2015  ■  IEEE industrial electronics magazine  29

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 benches [50]–[54]. This setup is pre-
The measurement of the thermal resistance sented in Figure 7(a). With this tech-
nique, they estimate the degradation
in a power module requires an estimation state of both the wire bonding and the
solder joints in power modules. The
of the chip temperature. health of wire bonding is estimated by
injecting a high current in the devices
with the related current source and
the forward voltage is consistently measurements of damage indicators. measuring the forward voltage. This
measured to estimate the degradation In this case, the measurements are current source is also used to increase
of the wire bonding and upper metal- carried out during the operation of the junction temperature of the devic-
lization of the chip. In a few cases, the the converter. The following sections es in view of estimating thermal im-
thermal resistance or impedance are will present the main offline and on- pedances. The junction temperature
measured to estimate the integrity of line solutions, which were proposed is measured via a TSEP using a low-
the thermal path. These two damage during the last decade. sensing current source. The thermal
indicators generally form the bulk of impedance is therefore estimated by
the research, which is mainly dedicat- Offline CM Systems injecting a high current in the devices
ed to IGBT power modules. Two main methods are proposed in and, at different time steps, interrupt-
However, many different method- the literature to carry out CM in offline ing the high current to measure the
ologies exist today. In fact, several conditions (Figure 7). The first is pro- junction temperature using the low-
researchers propose to simplify the posed by Newcastle University in the current source. In this experimental
measurement of health status by char- United Kingdom and is characterized setup, because of the high value of the
acterizing the devices in nonfunction- by the presence of a relay network dc bus voltage VDC, the forward volt-
al or “offline” conditions. Therefore, that allows the injection of current in age measurements must be isolated.
the power converter has to be halted each device to test the devices under Because the electrical conditions
to carry out these measurements. conditions that are identical to those are close to those presented in classi-
Other researchers study “online” obtained in traditional aging test cal aging test benches, this method is

Condition Monitoring of Inverters in Offline Conditions

External Electrical Stimuli Modified Operation of the Converter

T1 D1 T3 D3 T5 D5

Vdc M
3
T1 D1 T3 D3
T2 D2 T4 D4 T6 D6 L1
Vdc Relay
IL
Forward
T2 D2 T4 D4
Voltage
Isolation Selector
Relay Network

T1-T6 Measurements Relays

Control Control 100 mA D2-T4 On:
Switch High Current
Calculation Unit Control IL Vce (T4) Measurement

High Current: Forward Voltage Measurement D2-D3 On

One Device
Current in

T1-T4 On

Time

Time
100 mA: Junction Temperature Estimation

FIGURE 7 – An example of condition monitoring in offline conditions [13], [54].

30  IEEE industrial electronics magazine  ■  December 2015

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 able to give a very good estimation of
the IGBT modules health status. How-
ever, the presence of the relay net-
The presence of the relay network
work and the high-current sources to and the high-current sources to conduct the tests
conduct the tests has non-negligible
consequences on the global cost of
has non-negligible consequences on the
the system. global cost of the system.
To simplify the setup, the Univer-
sity of Aalborg presents an offline CM
test, which does not include addition- example, Nasrin et al. [56] propose to are linked to aging test benches in
al high current sources. Due et al. [13] use spread spectrum time domain re- laboratories, which are able to moni-
propose a specific current injection flectometry when the converter is not tor different degradation indicators
sequence in the devices to measure operating. This method is implement- during the life of the tested power
the forward voltage under high cur- ed by Li et Khan [57] in a PV inverter. modules. Smet et al. [44] and Rashed
rent and estimate the degradations et al. [47] from Montpellier Univer-
of the bondwires [Figure 7(b)]. This Online CM Systems sity (France), and Ghimire et al. [59],
injection is quite simple: for the char- Online CM systems should be able to [60] from Aalborg University propose
acterization of IGBTDUT-Low, the current measure aging indicators when the measurement systems to evaluate the
is increased applying the dc bus volt- power converter is working. However, forward voltage and the junction tem-
age to the inductive load (IGBTDUT-High the main issues for these systems are perature in switching conditions. The
and IGBTDUT-Low are turned-on) and, the varying electrical and thermal knowledge of this temperature is im-
at a given current level, IGBTDUT-High conditions: the current and junction portant because the forward voltage
is turned off—the forward voltage temperature levels are linked to the is temperature sensitive. With a good
measurement is then done during actual mission profile. Therefore, the and robust loss model, it can also be
the free-wheeling phase. The main conditions are not repetitive, which used to estimate the thermal resis-
problem of this setup is the use of a complicates the health monitoring. tance. However, there are many issues
relay. This is necessary to protect the This is the reason why authors like to carry out these measurements in a
measurement circuit against high volt- Xiong et al. [58] propose monitoring power converter.
ages when IGBTDUT-Low is in off state. As the indicators just before or after the First, the measurement of the for-
relays are generally slow in response operation of the inverter. In the case of ward voltage must have high resolu-
time, this relay is replaced by a more a hybrid vehicle, they present (but not tion to observe voltage variations
adequate monitoring circuit in follow- validate) a strategy that inserts a mea- in some cases of only several per-
ing work [42]. Choi et al [55] general- surement subroutine to the vehicle centage points, at which point large
ize this method and characterize the control system immediately after the degradation can already be present
forward voltage of IGBT and diodes key-on and/or key-off of each vehicle- [44]. Technically, the measurement is
in a three-phase power module. This use period. Even if the varying environ- also difficult to perform. The voltage
test bench is not used by the authors mental temperature is still an issue, the across the device in real conditions
to estimate the thermal impedance or monitoring is largely simplified. is equal to the dc bus voltage (several
resistance of power modules to esti- We have thus separated online hundreds of volts) when it is in the off
mate the thermal path integrity. How- measurements into two families: state and is then only a few volts when
ever, it seems to be possible to insert ■■ Quasi-online measurements: the it is in on state. The realization of an
a low-current source in the setup and current in the devices is controlled accurate measurement tool therefore
measuring the junction temperature and is the same for each monitor- requires the circuit to be protected
at different time steps as presented by ing step. against a very large voltage, while
Newcastle University. ■■ Real-online measurements: the de- having sufficient sensitivity when
Because of the inherent voltage vice is used and characterized in monitoring a low voltage several or-
drop (junction voltage) in the cur- real operation with conditions set ders of magnitude smaller during the
rent–voltage characteristics of diodes by the real mission profile. on state [61], [62]. Generally, the mea-
and IGBTs, variations of the forward surement also requires the use of an
voltage due to bondwire degradations Quasi-Online Measurements isolation circuit. As a consequence,
have to be observed when these de- These measurements are conducted the trade-off between the cost, the
vices are crossed by high currents (to during actual switching [42]; however, bandwidth, and the accuracy of the
include the ohmic resistive parts of the current level is controlled to ob- measurement circuit is of great im-
the bondwires). In the case of MOSFET tain repetitive conditions. These mea- portance in its design.
devices, the measurement is simpler, surements can be done when there The classical composition of a for-
as the resistive part can be mea- is a routine pause in the application. ward voltage measurement circuit is
sured without current injection. For The most advanced works in this area described in Figure 8. For more details

December 2015  ■  IEEE industrial electronics magazine  31

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 O’Connor et al. [65] propose a com-
The realization of an accurate measurement tool plete CM system for an IGBT three-
phase inverter where the problems of
therefore requires the circuit to be protected against noise and acquisition speed are treat-
ed. They measure the forward voltage
a very large voltage. VCE, the collector current IC, the control
voltage VGE, the gate current IG, the am-
bient and case temperatures in online
or examples, Wagenitz et al. [45] pro- estimate the dynamic resistance of IG- conditions to estimate the degrada-
pose a description of what should be BTs, which is considerably dependent tion of the solder and of the gate oxide
included is such measurement sys- on bondwire integrity. through a physical model. However,
tems, and Ghimire et al. [63] present they do not measure the junction tem-
a literature review on voltage limita- Real-Online Measurements perature. This is a main drawback of
tion and/or clipping circuits. Other Real-online measurements are the their method because almost all of the
solutions are presented by Davis and most difficult to manage. They necessi- estimated parameters are temperature
Sprenger [64] and O’Connor et al. [65]. tate the use of high-performance elec- dependent. Therefore, this experimen-
Finally, it should be noted that trothermal and loss models [43] or the tal setup has to be classified as a quasi-
it could be possible to improve the use of junction temperature measure- online method because of the lack of
sensitivity of bondwire degradation ment tools. Furthermore, measure- junction temperature measurement.
monitoring. For example, by measur- ments are typically very noisy and the Regarding junction temperature
ing the forward voltage along one use of complex data processing is gen- estimation, different solutions are
low-frequency period, Babel et al. [66] erally needed [45], [67]–[69]. proposed by a variety of authors. For

MOS1 Koenig et al. [62]

Voltage Measurement
Vce C1
R1 Vout Several V
DUT
Voltage Limitation or Clipping
Circuit

Amplifier
R R
Possible A /D Conversion D2 –
+
Output Isolation Vout
D1
Vce
Possible A /D Conversion
Beczkowski et al. [61]
DUT
Sampling (Holding)

Possible A /D Conversion

1,000
i (A)

0
–1,000
0 0.05 0.1 0.15 0.2
2
vCE (V)

1
0
0 0.05 0.1 0.15 0.2

FIGURE 8 – The measurement principle of the forward voltage [61], [62]. DUT: Device Under Test

32  IEEE industrial electronics magazine  ■  December 2015

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 example, Rashed et al. [47] interrupt
the operation of the converter during
The embodiment of a quasi-online and cost-effective
several tens of microseconds and in-
ject a low current in the device to cor- system in a real application seems to be attainable
relate the saturation voltage with this
temperature. However, in this case, ele-
in the near future.
ments are placed in series with the load,
which induce an increase of converter
losses. Other authors [59], [70] use the [71]. The direct use of the forward volt- voltage and therefore inherently induc-
forward voltage directly to estimate the age at high current as both a tempera- es an error in the temperature measure-
chip temperature, but using the forward ture sensor and a damage indicator is ment as the device ages. Ghimire et al.
voltage directly for temperature mea- counterintuitive, since bondwire degra- [59] propose a method to cancel the
surement is particularly prone to error dation causes a variation in the forward influence of bondwires on temperature

TABLE 2 – a COMPARISON OF THE DIFFERENT DAMAGE INDICATORS.

DAMAGE INDICATORS ADVANTAGES SHORTCOMINGS
DEGRADED PARTS
CLASSICAL INDICATORS
Forward voltage (high current) —Mature in laboratory conditions —Affected by numerous degradation mechanisms
Metallization, wire bonding, gate oxide —Usable in offline conditions [57], [54], [13] —Difficult measurements under switching conditions
—Usable in quasi-online conditions [47], [80] (measurement time, accuracy, noise) [65]
Thermal impedance or resistance —Easy to carry out in laboratory conditions —Difficult measurements under switching conditions:
Thermal path —Usable in offline conditions [53] variable loss and ambient temperature [69]
—Usable in quasi-online conditions [47]
Threshold voltage —Mature in laboratory conditions —Difficult measurements under real working conditions
Gate oxide —Can be measured under switching (low switching speed) [81]
Gate leak current —Mature in laboratory conditions —Difficult measurements under real working conditions
Gate oxide
C(V) characteristic —Easy to carry out in laboratory conditions —Not mature
Gate oxide
SWITCHING TRANSIENTS (NOT MATURE)
Gate current —Possible integration in the gate driver —Influenced by the semiconductor and the driver parts
Wire bonding [24] —Limited to devices with several chips in parallel
Device voltage —Possible integration in the gate driver —Influenced by the semiconductor part (gate oxide)
during switching —Influenced by the gate driver and the power circuit
Wire bonding [25], [27]
Current or voltage ringing —Usable with only phase current measurement —Need a powerful system for achieving the measurement
Gate oxide? [31]
CASE TEMPERATURE MEASUREMENTS (NOT MATURE)
Temperature imbalance —Easy measurements in offline and online conditions —Multidamages indicator (solder…)
Wire bonding [34] —Sensor number: implementation, cost
—Limited to devices with several chips in parallel
Electrothermal model of the heat sink —Easy measurements —Multidamages indicator (wire bonding…)
Thermal path [35] —Need an accurate electrothermal model
—Need accurate temperature sensors
Global methods (not mature)
Control of the converter efficiency —No additional sensor —Multidamages indicator
All parts [36] —Need an accurate power loss estimation
Current harmonics amplitude —No additional sensor —Relatively low sensitivity compensated by resonance
All parts [37] out circuit
—Specific tools have to be developed
—Perhaps cannot distinguish which specific component in
the system is damaged
Artificial intelligence —Global view of the converter —Need training
All parts [39] —Low variation of electrical quantities with aging

—All indicators are temperature dependent measurements in real-online conditions are not possible without junction temperature estimation.
—T he criterions have to be defined in correlation with the power module technology (semiconductor type and/or assembly technology) and operating conditions
(mission profile).

December 2015  ■  IEEE industrial electronics magazine  33

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 usually done in military applications)
Primary monitored degradation mechanisms are extends the device life drastically. In
that case, a lifespan of 20 years (like
linked to power module packaging and generally PV cells) can be easily attained (but
the cost is increased).
include bondwire aging and thermal resistance/ It should also be noted that other de-
impedance degradation. vices are subject to degradation in power
converters. As mentioned earlier, capaci-
tors (specifically electrolytic ones), con-
nectors and gate drives are also very
measurement, but no experimental time until they demonstrate accept- fragile components. Therefore, CM sys-
verification is proposed yet. Generally, able characterization and robustness. tems should estimate their health too.
the measurement of the junction tem- For these newly proposed indicators,
perature in actual operating conditions a large area of potential future scien- Acknowledgment
is still under study, and a universal solu- tific research will be defining the level at The authors thank the French Nation-
tion does not exist. A literature review which the indicator displays that critical al Research Agency (MEMPHIS ANR-
on junction temperature measurement damage is present. Thorough sensitiv- 13-PRGE-0005-01 PROGELEC project)
during converter operation was con- ity analysis on these indicators toward for the grants allocated for this re-
ducted by Baker et al. [72] in 2014, and a variety of parameters is also needed search topic.
a large amount of research was concur- to avoid erroneous interpretation of the
rently published since the publication of measurements. Similarly, artificial intel- Biographies
these articles [71], [73]–[79]. ligence techniques seem to be very inter- Yvan Avenas (Yvan.Avenas@g2elab.
At the current state of CM develop- esting but are not yet mature. grenoble-inp.fr) received his Ph.D.
ment, it appears that the most eco- At the current state of CM develop- degree from the National Polytech-
nomically and technically feasible ment, it appears that the most eco- nic Institute of Grenoble, France, in
solution is to use quasi-online mea- nomically and technically feasible 2002. He is currently an associate
surements. In fact, it can be well fitted solution is to use quasi-online mea- professor at the Grenoble Institute of
to use in some real applications. For surements. These can be well suited Technology. His first research areas
example, in transportation or PV sys- to use in several real applications. were power electronics cooling and
tems, opportunities to introduce inter- For example, in transportation or PV three-dimensional packaging, now he
ruptions exist regularly along the life systems, opportunities to introduce principally works on thermal charac-
of the converters, for example, during measurement windows exist regu- terization of power modules. He has
the night for PV systems and before larly along the life of the converters, worked on French collaborative and
the key-on in electrical vehicles. for example, during the night for PV industrial projects. Currently, he is
systems and before the key-on in elec- the coordinator of the health monitor-
Conclusions trical vehicles. The embodiment of a ing of power semiconductor modules
This article overviews a large number quasi-online and cost-effective system in photovoltaic inverters (MEMPHIS)
of propositions which have been made in a real application seems to be at- project supported by the French Na-
to monitor the health of power devices, tainable in the near future. tional Research Agency project that
especially in case of silicon IGBTs, which With the advent of the wide-band- concerns the health monitoring of
are synthesized in Table 2. Primary gap semiconductor devices and new power semiconductor modules in
monitored degradation mechanisms are packaging technologies, research photovoltaic inverters.
linked to power module packaging and still needs to be done to propose Laurent Dupont (Laurent.Du-
generally include bondwire aging and complete monitoring systems. Partic- [email protected]) received his electri-
thermal resistance/impedance degrada- ularly, new degradation mechanisms cal engineering degree in 2002 and
tion. The most advanced methods to de- and indicators will likely have to be his Ph.D. degree in electrical engi-
tect these degradation modes in terms defined through aging studies. This neering from the École Normale Su-
of robustness are measurements of clas- will be a big challenge especially for périeure de Cachan, France, in 2006.
sical indicators such as forward voltage nonmature technologies like GaN de- After ten years of experience in in-
drop in controlled offline conditions. vices, which will likely operate in con- dustry, he works as research scien-
However, the implementation cost likely verters that have markedly different tist in the Systèmes et Applications
remains high and infeasible for use in applications to silicon-based devices. des Technologies de l’Information
real operating converters. On the other Of course, the CM system is not et de l’Energie (SATIE) Laboratory
hand, the use of unconventional indica- sufficient. Good design practice in the Institut français des scienc-
tors such as switching transients may be within the specifications is essential es et technologies des transports,
more appealing for use in operating con- to extend reliability. For example, the de l’aménagement et des réseaux
verters, but have a longer development reduction of junction temperature (as (IFSTTAR), Versailles, France, since

34  IEEE industrial electronics magazine  ■  December 2015

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 2007. His research interest is the ro- References [18] P. Jacob, M. Held, P. Scacco, and W. Wu, “Re-
[1] A. Ristow, M. Begovic, A. Pregelj, and liability testing and analysis of IGBT power
bustness evaluation of power semi- semiconductor modules,” in Proc. IEE Collo-
A. Rohatgi, “Development of a methodology
conductor modules. In the French for improving photovoltaic inverter reliabil- quium on IGBT Propulsion Drives, 1995, pp. 1–5.
ity,” IEEE Trans. Ind. Electron., vol. 55, no. 7, [19] P. Cova and F. Fantini, “On the effect of power
project MEMPHIS, supported by the cycling stress on IGBT modules,” Microelec-
pp. 2581–2592, July 2008.
National Research Agency, his re- [2] S. Yang, A. Bryant, P. Mawby, D. Xiang, L. Ran, and tron. Reliab., vol. 38, no. 6, pp. 1347–1352, 1998.
P. Tavner, “An industry-based survey of reliabil- [20] D. C. Katsis and J. D. Van Wyk, “Experimental
search activities are focused on the measurement and simulation of thermal perfor-
ity in power electronic converters,” IEEE Trans.
evaluation of functional and aging in- Ind. Appl., vol. 47, no. 3, pp. 1441–1451, May 2011. mance due to aging in power semiconductor de-
[3] S. Yang, D. Xiang, A. Bryant, P. Mawby, L. Ran, vices,” in Proc. Industry Applications Conf. 37th
dicators usable to monitor the health IAS Annu. Meeting, 2002, vol. 3, pp. 1746–1751.
and P. Tavner, “Condition monitoring for de-
status of power components. vice reliability in power electronic converters: [21] S. Saha, J. R. Celaya, V. Vashchenko, S. Mahiuddin,
a review,” IEEE Trans. Power Electron., vol. 25, and K. F. Goebel, “Accelerated aging with electri-
Nick Baker ([email protected]) re- cal overstress and prognostics for power MOS-
no. 11, pp. 2734–2752, Nov. 2010.
ceived his M.Eng. degree in electri- [4] Smet, Vanessa, “Aging and failure modes of FETs,” in Proc. 2011 IEEE Energytech, 2011, pp. 1–6.
IGBT power modules undergoing power cy- [22] N. Patil, J. Celaya, D. Das, K. Goebel, and M.
cal and electronic engineering from Pecht, “Precursor parameter identification
cling in high temperature environments,” Ph.D.
Loughborough University, United dissertation, Montpellier 2, France, 2010. for insulated gate bipolar transistor (IGBT)
[5] T. T. L. Pham, F. Richardeau, and G. Gateau, prognostics,” IEEE Trans. Reliab., vol. 58, no. 2,
Kingdom, in 2011. In 2013, he began pp. 271–276, June 2009.
“Diagnosis strategies and reconfiguration of
pursuing his Ph.D. degree at Aal- a 5-level double-boost PFC with fault-tolerant [23] Y. Avenas, L. Dupont, and Z. Khatir, “Tempera-
capability,” in Proc. 2011 IEEE Int. Symp. Indus- ture measurement of power semiconductor
borg University, Denmark, in tem- devices by thermo-sensitive electrical param-
trial Electronics (ISIE), 2011, pp. 1857–1862.
perature measurements of power [6] M. Musallam, C. M. Johnson, C. Yin, C. Bailey, eters ; a review,” IEEE Trans. Power Electron.,
and M. Mermet-Guyennet, “Real-time life con- vol. 27, no. 6, pp. 3081–3092, June 2012.
semiconductor devices using elec- [24] S. Zhou, L. Zhou, and P. Sun, “Monitoring poten-
sumption power modules prognosis using on-
trical parameters. He was awarded line rainflow algorithm in metro applications,” tial defects in an IGBT module based on dynamic
in Proc. 2010 IEEE Energy Conversion Congress changes of the gate current,” IEEE Trans. Power
the European Power Electronics As- Electron., vol. 28, no. 3, pp. 1479–1487, Mar. 2013.
and Exposition (ECCE), 2010, pp. 970–977.
sociation Young Member Award in [7] H. Oh, B. Han, P. McCluskey, C. Han, and [25] W. Kexin, D. Mingxing, X. Linlin, and L. Jian,
B. D. Youn, “Physics-of-failure, condition moni- “Study of bonding wire failure effects on ex-
2015. His research interests include ternal measurable signals of IGBT module,”
toring, and prognostics of insulated gate bipolar
online monitoring of fault or degra- transistor modules: a review,” IEEE Trans. Power IEEE Trans. Device Mater. Reliab., vol. 14, no. 1,
Electron., vol. 30, no. 5, pp. 2413–2426, May 2015. pp. 83–89, Mar. 2014.
dation parameters in power semi- [26] D. Mingxing, W. Kexin, L. Jian, and X. Linlin,
[8] M. Ciappa, “Selected failure mechanisms of
conductor devices. modern power modules,” Microelectron. Re- “Condition monitoring IGBT module bond wire
liab., vol. 42, no. 4, pp. 653–667, 2002. lift-off using measurable signals,” in Proc. 7th
Henri Zara ([email protected]) Int. Power Electronics and Motion Control Conf.
[9] R. Wu, F. Blaabjerg, H. Wang, M. Liserre, and F. Ian-
studied general sciences oriented on nuzzo, “Catastrophic failure and fault-tolerant de- (IPEMC), 2012, vol. 2, pp. 1492–1496.
sign of IGBT power electronic converters - an over- [27] D. W. Brown, M. Abbas, A. Ginart, I. N. Ali,
electronics in Lyon (France) and re- P. W. Kalgren, and G. J. Vachtsevanos, “Turn-
view,” in Proc. IECON 2013 – 39th Annu. Conf. IEEE
ceived his Ph.D. degree in electronics Industrial Electronics Society, 2013, pp. 507–513. off time as an early indicator of insulated gate
[10] X. Perpiñà, L. Navarro, X. Jordà, M. Vellvehi, bipolar transistor latch-up,” IEEE Trans. Power
in 1997 at Institut National des Sciences Electron., vol. 27, no. 2, pp. 479–489, Feb. 2012.
J. F. Serviere and M. Mermet-Guyennet. (2012).
Appliquées. Since 2009, he as worked Reliability and Lifetime Prediction for IGBT Mod- [28] M. J. Ryan, “Method and apparatus for power
ules in Railway Traction Chains, Reliability and electronics health monitoring,” U.S. patent
in photovoltaic (PV) systems in a labo- 6,405,154, 2002.
Safety in Railway, X. Perpinya, Ed. InTech. [On-
ratory of Commissariat à l’énergie At- line]. Available: http://www.intechopen.com/ [29] A. Bryant, S. Yang, P. Mawby, D. Xiang, L. Ran,
books/reliabilityandsafetyinrailway/­reliability P. Tavner, and P. Palmer, “Investigation into
omique et aux énergies Alternatives on IGBT dV/dt during turn-off and its tempera-
andlifetimepredictionforigbtmodulesinrailwa
the subject of power conversion of PV ytractionchains ture dependence,” IEEE Trans. Power Electron.,
[11] M. Bouarroudj, Z. Khatir, J. P. Ousten, F. Badel, vol. 26, no. 10, pp. 3019–3031, Oct. 2011.
power plants as well as autonomous [30] A. E. Ginart, D. W. Brown, P. W. Kalgren, and
L. Dupont, and S. Lefebvre, “Degradation be-
systems. Reliability of power electronic havior of 600V–200A IGBT modules under pow- M. J. Roemer, “Online ringing characterization
er cycling and high temperature environment as a diagnostic technique for IGBTs in power
components is a critical subject in PV drives,” IEEE Trans. Instrum. Meas., vol. 58,
conditions,” Microelectron. Reliab., vol. 47,
inverters, and he focuses his studies no. 9–11, pp. 1719–1724, Sept. 2007. no. 7, pp. 2290–2299, July 2009.
[12] A. Hamidi, N. Beck, K. Thomas, and E. Herr, [31] A. Ginart, M. J. Roemer, P. W. Kalgren, and K.
on it across the French project MEM- Goebel, “Modeling aging effects of igbts in
“Reliability and lifetime evaluation of differ-
PHIS, supported by the National Re- ent wire bonding technologies for high power power drives by ringing characterization,” in
IGBT modules,” Microelectron. Reliab., vol. 39, Proc. Int. Conf. Prognostics and Health Manage-
search Agency. ment, 2008, pp. 1–7.
no. 6, pp. 1153–1158, 1999.
Franck Barruel (franck.barruel@ [13] J. Due, S. Munk-Nielsen, and R. Nielsen, “Lifetime [32] A. E. Ginart, I. N. Ali, D. W. Brown, M. Hargrave,
investigation of high power IGBT modules,” in P. W. Kalgren, and M. J. Roemer, “Power elec-
cea.fr) received his electrical engi- tronics health monitoring test platform for
Proc. 2011 14th European Conf. Power Electronics
neering degree in 2002 and his Ph.D. and Applications (EPE 2011), 2011, pp. 1–8. assessment of modern power drives and
[14] R. Bayerer, “Advanced packaging yields higher electric machines with regeneration capabili-
degree in 2005 from the Grenoble ties,” in Proc. 2010 IEEE AUTOTESTCON, 2010,
performance and reliability in power electron-
Institute of Technology. Since 2006, ics,” Microelectron. Reliab., vol. 50, no. 9–11, pp. 1–9.
pp. 1715–1719, Sept. 2010. [33] A. E. Ginart, D. Brown, P. W. Kalgren, and M. J.
he has worked at the Commissariat Roemer, “On-line ringing characterization as a
[15] K. S. Smith, L. Ran, and J. Penman, “Real-time de-
à l’énergie Atomique et aux énergies tection of intermittent misfiring in a voltage-fed PHM technique for power drives and electri-
PWM inverter induction-motor drive,” IEEE Trans. cal machinery,” in Proc. 2007 IEEE Autotestcon,
Alternatives and the National Insti- 2007, pp. 654–659.
Ind. Electron., vol. 44, no. 4, pp. 468–476, Aug. 1997.
tute of Solar Energy. He is now the [16] W. Wu, M. Held, P. Jacob, P. Scacco, and [34] B. Tian, Z. Wang, and W. Qiao, “Study on case
A. Birolini, “Investigation on the long term reli- temperature distribution for condition moni-
head of the Photovoltaic Systems toring of multidie IGBT modules,” in Proc. 29th
ability of power IGBT modules,” in Proc. 7th Int.
Laboratory. He is involved in the Symp. Power Semiconductor Devices and ICs, Annu. IEEE Applied Power Electronics Conf. Ex-
1995, pp. 443–448. position (APEC), 2014, pp. 2564–2568.
MEMPHIS project supported by the [35] D. Xiang, L. Ran, P. Tavner, A. Bryant, S. Yang,
[17] L. Dupont, S. Lefebvre, M. Bouaroudj, Z. Khatir,
National Research Agency, which and J. C. Faugières, “Failure modes on low volt- and P. Mawby, “Monitoring solder fatigue in a
age power MOSFETs under high temperature power module using case-above-ambient tem-
deals with the health monitoring of perature rise,” IEEE Trans. Ind. Appl., vol. 47,
application,” Microelectron. Reliab., vol. 47,
power devices. no. 9–11, pp. 1767–1772, Sept. 2007. no. 6, pp. 2578–2591, Nov. 2011.

December 2015  ■  IEEE industrial electronics magazine  35

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.
 [36] J. Morroni, A. Dolgov, M. Shirazi, R. Zane, and [52] B. Ji, V. Pickert, and B. Zahawi, “In-situ bond wire [68] J. M. Anderson and R. W. Cox, “On-line condi-
D. Maksimovic, “Online health monitoring in and solder layer health monitoring circuit for IGBT tion monitoring for MOSFET and IGBT switch-
digitally controlled power converters,” in Proc. power modules,” in Proc. 7th Int. Conf. Integrated es in digitally controlled drives,” in Proc. 2011
2007 IEEE Power Electronics Specialists Conf., Power Electronics Systems (CIPS), 2012, pp. 1–6. IEEE Energy Conversion Congress and Exposi-
2007, pp. 112–118. [53] B. Ji, X. Song, W. Cao, V. Pickert, Y. Hu, J. W. Mack- tion (ECCE), 2011, pp. 3920–3927.
[37] D. Xiang, S. Yang, L. Ran, P. Tavner, A. Bryant, ersie, and G. Pierce, “In situ diagnostics and [69] M. A. Eleffendi and C. M. Johnson, “Ther-
and P. Mawby, “Change of terminal character- prognostics of solder fatigue in IGBT modules mal path integrity monitoring for IGBT
istics of a voltage-source-inverter (VSI) due to for electric vehicle drives,” IEEE Trans. Power power electronics modules,” in Proc. 8th Int.
semiconductor device degradation,” in Proc. Electron., vol. 30, no. 3, pp. 1535–1543, Mar. 2015. Conf. Integrated Power Systems (CIPS), 2014,
13th European Conf. Power Electronics and Ap- [54] B. Ji, V. Pickert, W. Cao, and B. Zahawi, “In situ pp. 1–7.
plications, 2009, pp. 1–10. diagnostics and prognostics of wire bonding [70] J. Wu, L. Zhou, X. Du, and P. Sun, “Junction
[38] D. Xiang, L. Ran, P. Tavner, S. Yang, A. Bryant, faults in IGBT modules for electric vehicle temperature prediction of IGBT power mod-
and P. Mawby, “Condition monitoring power drives,” IEEE Trans. Power Electron., vol. 28, ule based on BP neural network,” J. Electr.
module solder fatigue using inverter harmonic no. 12, pp. 5568–5577, Dec. 2013. Eng. Technol., vol. 9, no. 3, pp. 970–977, May
identification,” IEEE Trans. Power Electron., [55] U.-M. Choi, I. Trintis, F. Blaabjerg, S. Jorgensen, and 2014.
vol. 27, no. 1, pp. 235–247, Jan. 2012. M. L. Svarre, “Advanced power cycling test for [71] L. Dupont and Y. Avenas, “Evaluation of
[39] S. Mohagheghi, R. G. Harley, T. G. Habetler, power module with on-line on-state VCE measure- thermo-sensitive electrical parameters
and D. Divan, “Condition monitoring of power ment,” in Proc. 2015 IEEE Applied Power Electronics based on the forward voltage for on-line
electronic circuits using artificial neural net- Conf. Exposition (APEC), 2015, pp. 2919–2924. chip temperature measurements of IGBT
works,” IEEE Trans. Power Electron., vol. 24, [56] M. S. Nasrin and F. H. Khan, “Use of spread spec- devices,” in Proc. 2014 IEEE Energy Conver-
no. 10, pp. 2363–2367, Oct. 2009. trum time domain reflectometry to estimate sion Congress and Exposition (ECCE), 2014,
[40] W. Zhou, X. Zhong, and K. Sheng, “High state of health of power converters,” in Proc. pp. 4028–4035.
temperature stability and the perfor- IEEE 13th Workshop on Control and Modeling for [72] N. Baker, M. Liserre, L. Dupont, and Y. Avenas,
mance degradation of SiC MOSFETs,” Power Electronics (COMPEL), 2012, pp. 1–6. “Improved reliability of power modules: a re-
IEEE Trans. Power Electron., vol. 29, no. 5, [57] Q. Li and F. H. Khan, “Identifying natural degra- view of online junction temperature measure-
pp. 2329–2337, May 2014. dation/aging in power MOSFETs in a live grid- ment methods,” IEEE Ind. Electron. Mag., vol. 8,
[41] H. Berg and E. Wolfgang, “Advanced IGBT tied PV inverter using spread spectrum time no. 3, pp. 17–27, Sept. 2014.
modules for railway traction applications: Re- domain reflectometry,” in Proc. 2014 Int. Power [73] M. Denk and M.-M. Bakran, “An IGBT driver
liability testing,” Microelectron. Reliab., vol. 38, Electronics Conf. (IPEC-Hiroshima 2014–ECCE- concept with integrated real-time junction
no. 6–8, pp. 1319–1323, June 1998. ASIA), 2014, pp. 2161–2166. temperature measurement,” in Proc. Int. Exhi-
[42] A. R. de Vega, P. Ghimire, K. B. Pedersen, [58] Y. Xiong, X. Cheng, Z. J. Shen, C. Mi, H. Wu, and bition and Conf. Power Electronics, Intelligent
I. Trintis, S. Beczckowski, S. Munk-Nielsen, V. K. Garg, “Prognostic and warning system for Motion, PCIM, Europe, 2014, pp. 1–8.
B. Rannestad, and P. Thogersen, “Test setup power-electronic modules in electric, hybrid [74] H. Luo, P. Sun, Y. Dong, W. Li, X. He, G. Chen,
for accelerated test of high power IGBT mod- electric, and fuel-cell vehicles,” IEEE Trans. Ind. E. Yang, and Z. Dong, “PiN diode chip tem-
ules with online monitoring of V ce and V f volt- Electron., vol. 55, no. 6, pp. 2268–2276, 2008. perature extraction method by investiga-
age during converter operation,” in Proc. 2014 [59] P. Ghimire, A. R. de Vega, S. Beczkowski, S. tion into maximum recovery current rate
Int. Power Electronics Conf. (IPEC-Hiroshima Munk-Nielsen, B. Rannested, and P. B. Thogers- di/dt,” in Proc. 2014 IEEE Energy Conver-
2014-ECCE-ASIA), 2014, pp. 2547–2553. en, “An online V ce measurement and tempera- sion Congress and Exposition (ECCE), 2014,
[43] M. Musallam and C. M. Johnson, “Real-time ture estimation method for high power IGBT pp. 4022–4024.
compact thermal models for health man- module in normal PWM operation,” in Proc. [75] H. Niu and R. D. Lorenz, “Sensing power
agement of power electronics,” IEEE Trans. 2014 Int. Power Electronics Conf. (IPEC-Hiroshi- MOSFET junction temperature using gate
Power Electron., vol. 25, no. 6, pp. 1416–1425, ma 2014–ECCE-ASIA), 2014, pp. 2850–2855. drive turn-on current transient proper-
June 2010. [60] P. Ghimire, A. R. de Vega, S. Beczkowski, B. ties,” in Proc. 2014 IEEE Energy Conversion
[44] V. Smet, F. Forest, J. Huselstein, A. Rashed, Rannestad, S. Munk-Nielsen, and P. Thogersen, Congress and Exposition (ECCE), 2014, pp.
and F. Richardeau, “Evaluation of monitoring “Improving power converter reliability: online 2909–2916.
as a real-time method to estimate aging of monitoring of high-power IGBT modules,” IEEE [76] P. Sun, H. Luo, Y. Dong, W. Li, X. He, G. Chen,
bond wire-IGBT modules stressed by power Ind. Electron. Mag., vol. 8, no. 3, pp. 40–50, Sept. E. Yang, and Z. Dong, “Online junction temper-
cycling,” IEEE Trans. Ind. Electron., vol. 60, 2014. ature extraction with turn-off delay time for
no. 7, pp. 2760–2770, July 2013. [61] S. Beczkowski, P. Ghimre, A. R. de Vega, S. high power IGBTs,” in Proc. 2014 IEEE Energy
[45] D. Wagenitz, A. Hambrecht, and S. Dieckerhoff, Munk-Nielsen, B. Rannestad, and P. Tho- Conversion Congress and Exposition (ECCE),
“Lifetime evaluation of IGBT power modules gersen, “Online Vce measurement method for 2014, pp. 4016–4021.
applying a nonlinear saturation voltage ob- wear-out monitoring of high power IGBT mod- [77] V. Sundaramoorthy, E. Bianda, R. Bloch,
server,” in Proc. 7th Int. Conf. Integrated Power ules,” in Proc. 15th European Conf. Power Elec- I. Nistor, G. Knapp, and A. Heinemann, “Online
Electronics Systems (CIPS), 2012, pp. 1–5. tronics and Applications (EPE), 2013, pp. 1–7. estimation of IGBT junction temperature (Tj)
[46] F. Forest, J.-J. Huselstein, S. Faucher, M. Elghaz- [62] A. Koenig, T. Plum, P. Fidler, and R. W. De using gate-emitter voltage (Vge) at turn-off,”
ouani, P. Ladoux, T. A. Meynard, F. Richardeau, Doncker, “On-line junction temperature mea- in Proc. 15th European Conf. Power Electronics
and C. Turpin, “Use of opposition method in surement of CoolMOS devices,” in Proc. 7th and Applications (EPE), 2013, pp. 1–10.
the test of high-power electronic convert- Int. Conf. Power Electronics and Drive Systems, [78] N. Baker, S. Munk-Nielsen, M. Liserre, and
ers,” IEEE Trans. Ind. Electron., vol. 53, no. 2, PEDS’07, 2007, pp. 90–95. F. Iannuzzo, “Online junction temperature mea-
pp. 530–541, Apr. 2006. [63] P. Ghimire, S. Beczkowski, S. Munk-Nielsen, surement via internal gate resistance during
[47] A. Rashed, F. Forest, J.-J. Huselstein, T. Martire, B. Rannestad, and P. B. Thogersen, “A review on turn-on,” in Proc. 16th European Conf. Power Elec-
and P. Enrici, “On-line [T J, V ce] monitoring of real time physical measurement techniques and tronics and Applications (EPE’14-ECCE Europe),
IGBTs stressed by fast power cycling tests,” their attempt to predict wear-out status of IGBT,” 2014, pp. 1–10.
in Proc. 15th European Conf. Power Electronics in Proc. 15th European Conf. Proc. Power Electron- [79] V. K. Sundaramoorthy, E. Bianda, R. Bloch, and
and Applications (EPE), 2013, pp. 1–9. ics and Applications (EPE), 2013, pp. 1–10. F. Zurfluh, “Simultaneous online estimation of
[48] L. Dupont, G. Coquery, K. Kriegel, and A. Melk- [64] R. I. Davis and D. J. Sprenger, “Methodology junction temperature and current of IGBTs us-
onyan, “Accelerated active ageing test on SiC JFETs and apparatus for rapid power cycle accumu- ing emitter-auxiliary emitter parasitic induc-
power module with silver joining technology for lation and in-situ incipient failure monitoring tance,” in Proc. Int. Exhibition and Conf. Power
high temperature application,” Microelectron. Re- for power electronic modules,” in Proc. IEEE Electronics, Intelligent Motion, PCIM, Europe
liab., vol. 49, no. 9–11, pp. 1375–1380, Sept. 2009. 64th Electronic Components and Technology 2014, 2014, pp. 1–8.
[49] R. O. Nielsen, J. Due, and S. Munk-Nielsen, “In- Conf. (ECTC), 2014, pp. 1996–2002. [80] P. Ghimire, A. R. de Vega, S. Beczkowski,
novative measuring system for wear-out indi- [65] P. O’Connor, R. W. Cox, and J. M. Anderson, “Near S. Munk-Nielsen, B. Rannested, and P. B.
cation of high power IGBT modules,” in Proc. real-time incipient fault detection in IGBT switch- Thogersen, “An online V ce measurement
2011 IEEE Energy Conversion Congress and Ex- es,” in Proc. 40th Annu. Conf. IEEE Industrial Elec- and temperature estimation method for
position (ECCE), 2011, pp. 1785–1790. tronics Society, IECON, 2014, pp. 4484–4491. high power IGBT module in normal PWM
[50] B. Ji, V. Pickert, B. Zahawi, and M. Zhang, [66] A. Babel, A. Muetze, R. Seebacher, K. Krischan, operation,” in Proc. 2014 Int. Power Electron-
“In-situ bond wire health monitoring circuit and E. G. Strangas, “Condition monitoring and ics Conf. (IPEC-Hiroshima 2014–ECCE-ASIA),
for IGBT power modules,” in Proc. 6th IET Int. failure prognosis of IGBT inverters based on 2014, pp. 2850–2855.
Conf. Power Electronics, Machines and Drives on-line characterization,” in Proc. 2014 IEEE [81] I. Bahun, N. Čobanov, and Ž. Jakopović,
(PEMD), 2012, pp. 1–6. Energy Conversion Congress and Exposition “Real-time measurement of IGBT’s operat-
[51] B. Ji, V. Pickert, W. P. Cao, and L. Xing, “Onboard (ECCE), 2014, pp. 3059–3066. ing temperature,” Autom. Časopis Za Autom.
condition monitoring of solder fatigue in IGBT [67] J. M. Anderson, R. W. Cox, and J. Noppakunka- Mjer. Elektron. Račun. Komun., vol. 52, no. 4,
power modules,” in Proc. 9th IEEE Int. Symp. Di- jorn, “An on-line fault diagnosis method for power pp. 295–305, 2012.
agnostics for Electric Machines, Power Electron- electronic drives,” in Proc. 2011 IEEE Electric Ship
ics and Drives (SDEMPED), 2013, pp. 9–15. Technologies Symp. (ESTS), 2011, pp. 492–497. 

36  IEEE industrial electronics magazine  ■  December 2015

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on February 16,2020 at 20:32:09 UTC from IEEE Xplore. Restrictions apply.