Workbench Study Concerning the Highest Reliability Outcome for PoL Converters with Different Output Capacitor Technologies

Abstract

The last decade’s studies show that the PoL (point-of-load) converter’s output capacitor is an important component for reliability, implying that its careful selection may determine the overall converter’s failure rate and lifetime. PoL converters are commonly found in many electronic systems, usually as part of the Intermediate Bus Architecture (IBA). Their important requirements are a stable output voltage at load current variation, good temperature stability, a low output ripple voltage, high efficiency, and reliability. If the electronic system is portable, a small footprint and low volume are also important considerations. These were recently well accomplished with eGaN (enhancement gallium nitride) transistor technology, whose VUFoM (vertical unipolar figure of merit) is 1.48 compared to 1.00 for silicon. This ensures a higher converter power density (watts/area). This paper reviews the most-used capacitor technologies, highlighting the reliability of these components as part of the converter’s output filter by presenting original data related to their best performance. The test was set up with EPC’s eGaN FET transistor, which was enclosed within a 9059/30 V evaluation board with a 12 V input and 1.2 V output. Different output capacitor technologies were evaluated, and reliability was calculated based on measurements of the ripple of the output voltage and thermal scanning.

1. Introduction

Due to their rapid advancement in the last decades, processor device technologies have imposed considerable challenges on power supply designers related to their low voltage and high current buck converters [1,2]. As an example, a device with a 7 nm process and 4 GHz for the clock requires 1 V and about 140 W peak power to supply the core [1].

Due to the specifics of the digital load (CPUs, FPGAs, SoC, ASICs), the PoL converter must match the load requirements as closely as possible to ensure excellent efficiency, high reliability, and a small form factor. With its dual role as an energy reservoir at the output of the power converter while simultaneously smoothing the output voltage, the output capacitor bank in the PoL DC–DC converter is increasing in importance. Consequently, the selection of an output capacitor bank became a very important step in the design of switching converters [1].

The PoL synchronous buck converter was herein selected for experimental measurements because it is the most commonly used topology nowadays in the supply of microprocessors, FPGAs, ASICs, memory banks, and alike digital circuits within notebooks, laptops, telecom equipment, servers, and data storage systems. PoL converters are typically used with high output current demands, associated with low internal voltage buses.

There has lately been a solid trend of progressively replacing silicon components within power converters with GaN devices [3]. The applications range from industrial motor drives, automotive and transportation systems (including hybrid and electric vehicles), photovoltaic inverters, railways, and wind turbines, to digital power equipment.

Gallium nitride-WBG (wide band gap) devices are semiconductors that have bandgaps that are considerably wider than silicon: 3.4 eV versus 1.1 eV. These new power transistors are able to switch at higher frequencies with lower losses, and they are characterized [4] by a high thermal conductivity, low ON resistance, low parasitic capacitance, a very small footprint, a lack of a body diode (so no reverse-recovery loss), robustness in harsh environments, and high breakdown voltages.

The most frequently highlighted feature of eGaN-based PoL converters relates to their impressive reliability [3]. This pushes an accompanying research effort on output capacitor technologies featuring better reliability and smaller form factors. The association of eGaN transistors with modern capacitors enables more efficient and compact power converter designs while improving reliability [5,6].

This paper presents the comparative results of tests on six different capacitor technologies used in the output filter of DC–DC converters and is an extended version of a conference paper [7]. The comparison is carried out with respect to the reliability of the entire power supply and its association with eGaN-type transistors. After the problem is stated and explained in Section 1, Section 2 presents the current stage of capacitor technology, Section 3 introduces the calculation of reliability for PoL power converters and outlines the effect of capacitor technology, Section 4 presents the experimental setup used for tests, and Section 5 describes the experimental measurement results that are able to illustrate the thermal effects and the reliability implications. Finally, the conclusions are stated in Section 6.

2. General Aspects of Capacitor Technologies and Reliability

2.1. Capacitor Technologies

This paper investigates various capacitor technologies suitable for the output capacitor bank within eGaN-based PoL converters [8,9,10,11,12,13,14]. This comprehensive comparison addresses the following types of capacitors:

• Aluminum electrolytic capacitors with non-solid electrolyte, can-type through hole mounting (radial).
• Aluminum polymer can-type V-type SMD capacitors.
• Polymer electrolytic multilayered capacitor in an SMD case.
• Multi-layer ceramic (MLCC) in an SMD case.
• Tantalum MnO₂ in an SMD case.
• Tantalum polymer capacitor in an SMD case.

2.2. Brief on the Reliability of Electronic Components

The reliability prediction for electronic components is commonly based on the Failure Rate λ which is assumed to have a quasi-constant value during the Lifetime (useful life) of the system. The Lifetime (useful life) [15] is defined with a bathtub curve, as seen in Figure 1.

The reliability of a component or system is defined in the following form:

R(t) = e^−λt

(1)

Herein, λ signifies the intrinsic failure rate and is measured in F/10⁶ h or FIT, excluding early failures and wear-out failures. The mean time between failures (MTBF, measured in hours) is the reversed value of the failure rate.

MTBF = λ⁻¹

(2)

For any system, the failure rate is the sum of the failure rates of all components.

λ_system = λ_MOSFET + λ_control + λ_diode + λ_inductor + λ_{output capacitor} + λ_{PCB+connectors}

(3)

In the case of a point-of-load converter, as solely used in telecom and computer applications, the most important failure rate component comes from the output capacitor bank [16], as exemplified in Figure 2. Hence, it is very important to reduce the failure rate of the output capacitor bank with the introduction of new technologies.

3. The Reliability Standard Used in Paper and the Specifications for Capacitors

In 2004, the International Electrotechnical Commission released the IEC-TR-62380 standard [17]. This standard includes features of newer components such as polymer capacitors that do not appear in older standards such as MIL-HDBK-217 [18]. This section outlines the calculation of the failure rate λ and the MTBF (mean time between failures) based on the mathematical model for the capacitors provided in the IEC-TR-62380 standard. The other parameters used in the calculation correspond to the use of the capacitor as the output capacitor for a PoL buck converter.

Both the IEC-TR-62380 and MIL-HDBK-217 standards use reliability data derived mainly from the field data of electronic equipment operating in an environment defined as ground, stationary, and weather-protected. This means equipment for stationary use, on the ground, in weather-protected locations, operating permanently or otherwise with controlled temperature and humidity, and under the benefit of good maintenance. This category also applies to telecom equipment and computer hardware, which is the topic of eGaN-based PoL converters.

In addition to the environmental conditions, experience has shown that component reliability is similarly influenced by mechanical conditions. In this paper, reliability calculations have to be performed according to a mission profile [17,18,19,20,21,22].

A mission profile defines a table that includes all the details on the ambient temperature cycles during the lifetime of the device, on/off state durations, device loading, numbers of operation cycles, and so on [18,19,20,21,22].

3.1. The Mathematical Model Used by the IEC-TR-62380 Reliability Standard

Reliability is evaluated by the calculation of the failure rate of the capacitor. The mathematical model for the failure rate stated by the IEC-TR-62380 standard is defined as:

$$\mathsf{\lambda }=\mathsf{\alpha }\times \left(\left[\frac{{\displaystyle {\sum }_{i=1}^y}{\left({\mathsf{\pi }}_T\right)}_i\times {\mathsf{\tau }}_i}{{\mathsf{\tau }}_{on}+{\mathsf{\tau }}_{off}}\right]\times {\mathsf{\pi }}_A+\mathsf{\beta }\times {10}^{-3}\times \left[{\displaystyle {\sum }_{i=1}^j}{\left({\mathsf{\pi }}_n\right)}_i\times {\left(\Delta T_i\right)}^{0.68}\right]\right)\times \frac{\left({10}^{-9}\right)}{\mathrm{h}}$$

(4)

where the parameter π_T has the following expression:

$${\mathsf{\pi }}_T=e^{\mathsf{\gamma }\left(\frac{1}{\delta }-\frac{1}{273+T_A}\right)}$$

(5)

and in the case of aluminum electrolytic non-solid electrolyte capacitors, this becomes:

$${\mathsf{\pi }}_T=e^{\mathsf{\gamma }\left(\frac{1}{\delta }-\frac{1}{273+T_R}\right)}$$

(6)

where T_A is ambient temperature and T_R is capacitor temperature.

The terms within the model are meticulously explained in the IEC-TR-62380 standard. The temperature factor π_T was derived for each capacitor type and is shown in Figure 3, where an ambient temperature of ~25 °C was considered for experimental measurements using the nomograms provided by the IEC-TR-62380 standard. The resulting values for π_T are shown in the first row of

Table 1. Calculated parameters π_T and ΔT_i function of capacitor’s technology.

Parameter	MLCC	Tantalum MnO2	Tantalum Polymer	Polymer Multi-Layered	Al-Polymer Can-Type	Al Electro Lytic Wet
πT	0.93	0.99	0.99	0.7565	0.7565	0.48
TR (ΔTi)	32 °C	32 °C	31 °C	30 °C	31 °C	32 °C

. The measured temperatures of the capacitor’s capsule are in the second row. For a better view and understanding of calculation particularities, Appendix A is provided at the end of the paper, in addition to notes and data from reliability standards.

3.2. Calculation of the Failure Rate

The calculation is next performed for six possible capacitor technologies (Table 1). In each case, FIT stands for failure in time (1 FIT = one failure in 10⁹ h) and F/10⁶ h is failures in one million hours (the last variant is used in the MIL-HDBK-217 standard).

The calculation of the failure rate considers the following assumptions:

• τ_i = 365 days × duty cycle = 365 × 0.15 = 54.75.
• π_A parameter exists only for electrolytic wet capacitors.
• π_T = is derived with the nomograms shown in Figure 3.
• n_i = 365, (π_n)_i = (n_i^0.60) × 1.
• τ_on= 1 and τ_off = 0.
• Σ(π_n)_i = n_i, Σ(π_T)_i = π_T.
• ΔT_i = according to Table 1.

The temperature of the capacitor capsule T_R within Table 1 is provided by thermal scanning with an infrared photography device (no air cooling, natural convection). These values and their corresponding photo are shown in Figures 8–13, Section 5. More details concerning these assumptions are provided in Subchapter 8: mission profiles from TR-62380 standard.

• Failure rate for MLCC-SMD case capacitors:

λ_MLCC = 0.05 × {[(0.930291912 × 365 × 54.75)/(1 + 0)] + 3.3 × (10⁻³) × [(1.7 ×
365^0.6) × 32^0.68]} × (10⁻⁹)/h = 929.53605 FIT = 0.92953605 F/10⁶ h

(7)

MTBF_MLCC = 1,076,020 h or 122.80 years

(8)

2.

Failure rate for tantalum MnO₂ SMD case capacitors:

λ_{tantalum MnO2} = 0.4 × {[(0.99 × 365 × 54.75)/(1 + 0)] + 3.8 × (10⁻³) × [(1.7 × 365^0.6) × 32^0.68]}
× (10⁻⁹)/h = 6050.018 FIT = 6.050018 F/10⁶ h

(9)

MTBF_{tantalum MnO2} = 165,288.8 h or 18.86 years

(10)

3.

Failure rate for tantalum polymer capacitors SMD case:

λ_{tantalum polymer} = 0.4 × {[(0.99 × 365 × 54.75)/(1 + 0)] + 3.8 ×
(10⁻³) × [(1.7 × 365^0.6) × 31^0.68]} × (10⁻⁹)/h = 6049.966 FIT = 6.049966 F/10⁶ h

(11)

MTBF_{tantalum_polimer} = 165,290.2 h or 18.86 years

(12)

4.

Failure rate for polymer electrolytic multilayer capacitors SMD case capacitors:

λ_{polymer electrolytic multilayer} = 0.6 × {[(0.756565 × 365 × 54.75)/(1 + 0)] + 1.4 × (10⁻³) × [(1.7 ×
365^0.6) × 30^0.68]} × (10⁻⁹)/h = 6049.966 FIT = 6.049966 F/10⁶ h

(13)

MTBF_{polymer multilayer} = 165,290.2 h or 18.86874 years

(14)

5.

Failure rate for aluminum polymer can-type V-type SMD capacitors:

λ_{polimer can type} = 0.6 × {[(0.756565 × 365 × 54.75)/(1 + 0)] + 1.4 × (10⁻³) × [(1.7 × 365^0.6) ×
31^0.68]} × (10⁻⁹)/h = 9072.71 FIT = 9.072 F/10⁶ h

(15)

MTBF_{Al-polimer can type} = 110,000 h or 12.6 years

(16)

6.

Failure rate for aluminum electrolytic non-solid electrolyte (wet) can-type through hole mounting (radial) capacitors:

λ_{Al-electrolytic wet} = 1.3 × {[(0.678 × 365 ×54.75)/(1 + 0)] × 1 + 1.4 × (10⁻³) × [(1.7 ×
365^0.6) × 32^0.68]} × (10⁻⁹)/h = 17633.44 FIT = 17.63344 F/10⁶ h

(17)

MTBF_{Al-electrolytic wet} = 56,710.43 h or 6.4 years

(18)

It is necessary to proceed with extreme caution when dealing with the MTBF number and its definition because confusion frequently occurs between the MTBF and the lifetime [15]. For better clarification of this issue, the life expectancy for the aluminum electrolytic non-solid electrolyte capacitor is calculated as an example in the next subparagraph, according to TR62380.

3.3. Life Expectancy of Aluminum Electrolytic Non-Solid Electrolyte Capacitors

Because, in the literature, the aluminum electrolytic non-solid electrolyte capacitor is often subject to critical issues due to its proneness to early failure, a lifetime calculation was performed for this capacitor only [23]. The main factor that affects the lifetime of a capacitor is well-known as being the increased operating temperature. The rule of thumb would be that, for each increase of 10 °C in operating temperature, the lifetime will be reduced by half (m). The failure rate of non-solid electrolyte (wet) aluminum capacitors is assumed to be constant, but only within the life expectancy limits.

In order to calculate life expectancy, the IEC-TR-62380 standard provides a nomogram, as in Figure 4, which is based on qualification tests, as explained in

Table 2. Qualification test condition according to capacitor types.

Qualification Test	Duration (Hours)	Temperature (°C)
1	1000	85
2	2000	85
3	5000	85
4	2000	105
5	10,000	85
6	2000	125

. Therein, T_C represents the temperature of the capacitor with a default value, i.e.,

T_C = T_A + 5 °C

(19)

The IEC-TR-62380 standard also states that it is necessary to take account of two facts:

(1)

Life expectancy depends on the operating temperature, qualification, and test conditions (i.e., technology type);

(2)

Ambient temperature for a non-solid electrolyte aluminum capacitor is taken at its immediate vicinity, meaning that overheating of the other components has to be considered. The mathematical formula for the life expectancy is:

$$\mathit{Life} \mathit{expectancy} [\mathrm{hours}] = \mathit{qualification} \mathit{test} \mathit{duration} \times 2^{\left[\frac{\left(T_M+5\right)-T_C }{10}\right]}.$$

(20)

When this was applied to the example considered herein, the T_M represents the maximum temperature of the climatic category (here 85 °C). Hence, the life expectancy was 55,715 h or about 10 years.

4. Experimental Setup for Measurements

The calculation of the failure rate requires certain data collected from the experiment. In this respect, an EPC9059/30V development board [24] containing a buck converter with cutting-edge discrete transistor eGaN technology inside (parameters and characteristics in

Table 3. Parameters of the tested PoL buck converter.

Parameter	Value
Load resistor	Rload = 1 Ω/50 W
Load current	Iout = 1.2 A and 12 A
Input current	Iin = 1.2 A
Input voltage	Vin = 12 V
Output voltage	Vout = 1.2 V (Duty cycle = 7.65%)
Switching frequency	fsw = 250 khz
Inductor	L = 0.03 mH
Ambient temperature	~25 °C
Duty cycle	15%

) was used. The two EPC2100 enhancement-mode gallium nitride FET transistors (eGaN) [6,24] have the following high-performance features (which have a particular impact on switching behavior):

• R_ds(ON) = 8.2 mΩ, Q_g = 3.6 nC and 10 A for Q1.
• R_ds(ON) = 2.1 mΩ, Q_g = 0.6 nC and 40 A for Q2.

The gate control was ensured with a Texas Instruments LM5113 gate driver [24,25]. This experimental setup uses the measurement connection diagram depicted in Figure 5. The six different types of capacitor technology were successively connected as output capacitor banks. An infrared thermal camera was used to scan the actual temperature of the capacitor’s surface. This value of the temperature was required by the temperature factor π_T in the reliability calculation with the IEC TR 62380 standard. Figure 6 shows the appearance of the detected temperatures for Murata’s SMD polymer electrolytic can type. The equivalent series resistance (ESR) is a very important parameter for both the great impact on I²R losses [4] and the ripple capability. The theoretical ESR was extracted from the catalog datasheet of each capacitor. A switching frequency of 250 kHz was adopted in this experiment. For the validation of the CCM, the operation of the converter is depicted in Figure 7, along with the waveforms for the tested converter (switching point voltage and coil current).

5. Results

After the reliability calculation, the equivalent series resistance (ESR) measurement is done. This is a very important parameter for both I²R losses and ripple capability. The theoretical ESR was extracted from the catalog datasheet of each capacitor.

5.1. Experimental Results

Considering the reliability calculation performed in Section 3.2, ESR measurements (using an electronic RLC bridge), and the collection of the ripple for the output voltage using the oscilloscope, the experimental results are grouped in Figure 8, Figure 9, Figure 10, Figure 11, Figure 12 and Figure 13, reporting:

• The temperature at the surface of the capacitor;
• The component ID from the manufacturer’s catalog;
• The nominal capacitance value and rated voltage;
• The measured actual capacitance value;
• A photo of each capacitor tested;
• The capsule’s temperature T_R for each capacitor;
• MTBF number in years;
• The peak-to-peak ripple voltage;
• The failure mode (%) taken from the standard.

Obviously, the results indicate that the ESR, ripple voltage, and thermal behavior are different for each type of capacitor [26]. Nevertheless, the MTBF number tends to increase with the newest capacitor technologies. With the information contained in Figure 8, Figure 9, Figure 10, Figure 11, Figure 12 and Figure 13, a synthesis of the behavior of the six types of capacitors is shown in

Table 4. Capacitors’ comprehensive characterization by their technology after measurements and calculation.

Capacitor Technology	ESR	Reliability	Footprint	Ripple	Price	Failure Mode	Stable over Time	Stable over Voltage	Optimal Solution
Ta-MnO2 H-chip	very high	••	••	°	↑ (relatively high)	mostly short	•	•	to avoid
MLCC SMD	very low	•••	•••	•••	↓ (low)	mostly short	° (aging !)	°	very good
Aluminum electrolytic non-solid electrolyte	ultrahigh	° °	°	°	↕ (evenly)	short↔open	°	°	to avoid
Polymer electrolytic can-type, V-chip	low	••	•	••	↑↑ (high)	mostly open	°	•	very good
Ta-polymer H-chip(flat)	low	•••	••	•	↑↑ (high)	mostly short	•	•	very good
Polymer multilayered H-chip(flat)	very low	•••	••	•	↑↑↑ (costly)	mostly open	•	•	the best

• is a plus, ° is a minus.

. In the last column, an attempt was made to decide on their use within the converter. This table could serve as a benchmark for designing reliable buck converters with a low output voltage and high current.

5.2. Experimental: First Extension

Although they are not known to be often used as output capacitors for DC–DC converters, capacitors with a dielectric film (polyester) were also considered in this experiment, as a point of curiosity. Thus, a 25 μF/250 V capacitor film of 10% type PMP0305-β IPRS-Baneasa was tested under the same conditions as the other six typologies. A temperature of 29 °C was found on the package (case), which resulted in a failure rate of 1.59878 × 10⁻⁶ (and a factor π_T = 0.8), leading to an MTBF of 71.4013 years. A fairly high ripple value of 2.55 V_pp) was measured (details are shown in Figure 14), which disqualified it from use within the converter’s output filter.

5.3. Experimental: Second Extension

To further highlight the role of capacitor reliability in the overall reliability of the converter and because the results obtained are adequate for conventional mode operation, a new test at a much harsher stress level was conducted. Thus, the parameters of the converter were modified. A suggested comparison of the two stress situations for the tested capacitors is presented in Figure 15. Compared to the parameters in Table 3, the load current differed in particular, being increased 10 times in this new stress situation, and the values obtained (including package temperature) are summarized in

Table 5. Actual ESR in working environment within mission profile.

Capacitor Technology	ESR at 25 °C (mΩ)	Tactual (°C)	ESRactual (mΩ)
MnO2-tantalum	100	32	90.751916
Ta-polymer	50	31	46.009383
Polymer multilayer	26	30	24.258858

.

The failure rate and MTBF were calculated for this new stress situation, and the values obtained (including package temperature) are summarized in

Table 6. Indices of reliability yield at load current of 1.2 A and 12 A.

Capacitor Technology	TR (°C) at 1.2 A	TR (°C) at 12 A	πT	MTBF at 1.2 A (yr)	MTBF at 12 A (yr)
Ta-MnO2	32	40	1.02	18.86	13.99
MLCC	32	35	1.02	122.80	122.83
Al-lytic wet	32	45	0.80	6.40	5.49
Polymer can-type	31	40	0.98	12.60	4.48
Ta-polymer	31	35	1.03	18.86	13.86
Polymer multilayered	31	35	0.90	18.86	10.57
Film	29	31	0.83	71.39	71.40

.

5.4. Experimental: Third Extension

Because ESR has a very complicated relationship with temperature and frequency variation, especially for tantalum and polymer capacitors, a discussion about ESR must be addressed. Catalogs and datasheets provide scarce data about the maximum ripple characteristics without detailing the frequency and temperature variation. The fact is that the ESR of tantalum and polymer capacitors varies with the changes in these two variables [26,27,28,29].

In the case of tantalum capacitors containing MnO₂ (manganese dioxide) electrodes, there is a very strong dependence on temperature due to the dominance of the cathodes on ESR. A relationship that approximates this behavior [18] is:

$$ES{R}_{actual }=ES{R}_{@25°\mathrm{C}}\times 4^{\frac{25°\mathrm{C}-T_{actual}}{100°\mathrm{C}}}$$

(21)

where “actual” means the actual package temperature for the capacitor in the working environment within the mission profile. This equation shows that a value four times lower than that of the ESR at 25 °C is obtained when the working temperature increases by 100 °C above ambient temperature. The calculation provides the values for ESR that are shown in Table 5. A histogram of the percentage decrease in actual ESR determined at the working temperature seen by the capacitor is illustrated in Figure 16.

6. Conclusions

The output capacitor bank from a low-voltage, high-current PoL converter is more important for the converter’s reliability than the semiconductor devices. The performance required from the output capacitor bank makes selection difficult.

This paper proposes an investigation of several capacitor technologies in order to find the best match for the eGaN FET transistor-based PoL converters. This includes the kinds of parameters that are less prominent in a quick specification overview, such as reliability, modes of failure, and behavior with time or voltage. The selection of the right capacitor technology has afforded more consideration in the case of low-voltage, high-current applications.

Taking into account the results derived within this paper, it is suggested that a low output ripple voltage and high reliability for the eGaN transistor-based DC–DC converter can be achieved using capacitors with a lower ESR at the switching frequency. In this study, polymer electrolytic can-type SMD, tantalum polymer SMD, and polymer multilayered SMD technologies have been found to provide good ripple absorption and optimal reliability performance, with the downside of a higher cost. MLCCs can be used as an addition in order to form a parallel bank along with the technologies mentioned above, with an intended role at high frequencies. However, MLCCs suffer from an actual capacitance change with a DC voltage component, a risk of cracking when voltage exceeds ratings, and the generation of acoustic noise. Tantalum capacitors have good company-specific design guidelines and a capacity for self-healing but are prone to starting fires on the printed circuit board.

Due to their higher ESR values, the aluminum electrolytic wet capacitors result in high output ripple voltage and poor filtering. In addition, they are much larger in volume and require a bigger footprint on the printed circuit board. Despite many articles that indicate that this type of capacitor is the most prone to failures within a printed circuit board, their low cost and high capacitance density make aluminum electrolytic wet capacitors the first choice for many designers. Replacing them with polymer capacitors offers a better but more expensive solution.

The major and unique contributions of the paper include:

• It provides a comprehensive review of capacitor technology options for low-voltage, high-current PoL converters;
• It provides a guideline for the selection of the proper reliability standard for the calculation of the failure rate for modern capacitor technologies;
• It performs calculations for failure rates and MTBFs for various capacitor technologies when used in PoL converters equipped with eGaN transistors;
• It performs a calculation of the life expectancy for electrolytic-type capacitors based on component qualifications;
• It uses a thermal infrared camera for capacitors’ temperature measurements (used in π_T factor calculation) and uses a precise in-circuit measurement of the actual ESR of the capacitors, which helps thermal modeling;
• It uses a precise in-circuit ripple measurement for various capacitor technologies;
• It provides a comparative investigation of the possible failures.