Analysis of Discharge Failure Mechanism of IGBT Power Modules

Abstract

IGBT power modules are usually used as circuit-breaking components in power systems, and are widely used in solid-state DC circuit breakers, hybrid DC circuit breakers, all-electric aircraft, high-speed railways, new energy vehicles, and power transmission systems. In these systems, IGBT power modules are usually faced with extremely harsh working conditions and there is a failure risk. Insulation degradation should be a cause for concern as a potential path of power module failure. In this paper, the discharge phenomena of the IGBT power module were observed based on Intensified Charge Coupled Devices (ICCD), and the triple junctions composed of copper–ceramic–silicone gel inside IGBT were found as the discharge points. Furthermore, the directed bonded copper (DBC) ceramic filled with silicone gel was used as a test sample to study the discharge failure process, including the partial discharge (PD), surface charges, and electric trees. The mechanism of discharge failure is discussed and analyzed. The insulation degradation process is accompanied by phenomena such as severe partial discharge and rapid electric tree growth. This research provides support for the analysis idea and guidance of the research method for the cause of power module failure.

1. Introduction

As an efficient energy technology, DC power grid has broad application prospects in data centers, subway traction systems, ship power supply systems, and all-electric aircraft power supply systems [1]. The fault current of the DC power grid rises quickly and has a high amplitude, and has no natural zero crossing point, which requires the DC circuit breaker to have the ability to quickly and reliably cut off the fault current [2].

Solid state circuit breaker (SSCB) has received extensive attention as an effective DC protection device [3,4]. Figure 1 shows a typical topology diagram of SSCB [5], in which the solid-state switch is connected in series by two insulated gate bipolar transistors (IGBTs), which can realize bidirectional power flow. A snubber circuit is used to limit the transient overvoltage generated when the solid-state switch turns off. The fault detection circuit can quickly identify the fault and transmit the fault signal to the drive and control circuit, and then the control circuit sends a command to the solid-state switch according to the type of fault, thereby quickly isolating the fault.

However, the working conditions of the SSCB determine that the IGBT module needs to withstand a high blocking voltage in a short time (including AC and unipolar square wave electric stress), which may cause the failure of the IGBT module. Figure 2 depicts the IGBT module, which implements a direct bonded copper (DBC) ceramic substrate. The chips and diodes are soldered to the upper surface of the DBC. The encapsulation structure, which is composed of silicone gel and ceramic substrates, prevents electrical discharges. The encapsulation structure sustains high electric stress between the upper and lower copper layers of DBC when the IGBT chip is in a blocked state. The distribution of electric field is extremely abnormal at the triple junctions, which have been recorded in red in Figure 2. They are defined as the junctions of the ceramic, copper layer, and silicone gel. The partial discharges (PDs), which are a risk to the module [6], may initiate at the triple junctions when they are overstressed, spread on the interface of ceramic–silicone gel, and ultimately lead to failure [7,8].

Some studies have been conducted on the insulation failure in IGBT modules. When Si-based IGBTs first emerged, the problem of creeping discharge within the module was regarded as a potential hidden danger and attracted the attention of researchers. Mitic et al. was the first to determine that the triple junctions were the primary sources of partial discharge using optical measurement methods [9]. Fabian et al. further captured the electroluminescence before the generation of partial discharge using a charge-coupled device (CCD) [10]. With the development of high-voltage high-power Si-based IGBTs, the phase-resolved partial discharge (PRPD) [11] and the origin of PD [12] were analyzed. In terms of the subsequent failure process of PD, Kumada et al. studied the relationship between the charge density and discharge at the glass/ceramic–gel interface [13,14]. However, the electrode used was a needle, not the DBC copper, which cannot be equivalent to the actual situation in IGBT modules.

So far, the failure mechanism of the IGBT module has not been clarified. There are few reports on the insulation failure process of the package structure in the actual module. Therefore, in this paper, the PD detection and location platforms were established, and the discharge phenomena of IGBT power module were observed. Furthermore, the interface charges and electric trees were measured in DBC ceramic filled with silicone gel. The insulation degradation process is accompanied by phenomena such as severe partial discharge and rapid electric tree growth. We believe this research will provide support for the analysis idea and guidance of the research method for the cause of power module failure.

2. Experiment Platform

2.1. Partial Discharge Location

Although related studies [9,10,11,12,13,14,15] have pointed out that the triple junction is the risk point of insulation, the discharge location is limited to the package structure. Other insulation risk points that may exist in the actual IGBT module have not been reported yet. Moreover, there are many circuit components and complex structures in the IGBT module, and the potential failure locations may not be of a single type. Therefore, in order to analyze the insulation failure mechanism of the IGBT module, the risk point of the insulation inside the module should be found first.

Discharge location for actual IGBT modules requires high-precision detection methods. The usual discharge location methods include ultrasound (UA), ultra-high frequency (UHF) and optical methods, which can be divided into ultraviolet and fluorescence detection methods. The spatial accuracy of UA and UHF can meet the discharge location at the equipment level (such as SSCB), but it is difficult to realize the discharge positioning inside the module [16,17]. In addition, the amplitude sensitivity of UA and UHF is poor, and struggle to detect weak discharges, making them unsuitable for the IGBT module. Since the strong discharge mostly originates from the triple junction, it usually overwhelms weak discharges elsewhere. The ultraviolet detection method realizes image localization through the ultraviolet signal generated by the discharge. This method has a fatal flaw in the application of IGBT modules: the discharge inside the silicone gel cannot be observed because the strong ultraviolet signal mostly comes from the gas corona discharge.

The fluorescence method has extremely high sensitivity and can realize image localization with high spatial precision. However, it has high requirements for the test environment: one needs a darkroom, and the other requires the sample to be transparent.

In summary, we have built a test platform for IGBT module discharge luminescence localization based on intensified charge coupled devices (ICCD). And the IGBT module is encapsulated in a glass container, so as to ensure the transparency of the sample and realize the observation of all positions inside the module. This operation just replaces the outer frame in the IGBT module, which is separate from the electrical stress inside the module and is no risk of discharge.

The accuracy test of the discharge location platform was carried out before the actual module test. The test discharge sample was a “silicone gel-needle electrode” sample: the needle electrode was inserted into the prepared silicone gel and voltage was applied to the needle tip to produce partial discharge. The ICCD was used to record the luminous intensity of the needle tip.

The test results are shown in

Table 1. PD amplitude detection accuracy of ICCD.

Amplitude (pC)	~30	~10	~5	~3
Single mode	√	×	×	×
Accumulation mode 1	√	√	×	×
Accumulation mode 2	√	√	√	√

‘√’: This symbol refers to the amplitude (in row 1) that can be responded to in this mode (in column 1); ‘×’: This symbol refers to the amplitude (in row 1) that can’t be responded to in this mode (in column 1).

. Note that the discharge amplitude listed in the table are the multiple observations in the shooting period, which means that for the detection accuracy defined as ~3 pC, there will also be a short-term transient higher than 5 pC discharge pulses, but the entire shooting time remains at the 3 pC level as a whole. The corresponding settings of the three modes are shown in

Table 2. Mode parameters for ICCD.

Model	Number	TTL Width	Exposure Time	MCP Gain
Single	1	10 s	600 s	4095
Accumulation 1	10	1 s	1 s	4095
Accumulation 2	100	0.1 s	0.1 s	4095

. Combining Table 1 and Table 2, it can be seen that ICCD has the highest detection accuracy in the high-gain short-exposure multiple accumulation mode, and the highest accuracy can reach 3 pC level.

Considering the upper limit of each ICCD parameter (for example, the exposure limit is 9.1 h), theoretically, the detection of partial discharge below 3 pC can be realized, but due to the high randomness of the discharge behavior of the sample, it is difficult to maintain the discharge of 1 pC level at the needle tip for a long time in actual testin, so as to create a stable discharge source for the test. Therefore, the shooting mode with 1 pC precision cannot be obtained.

2.2. Partial Discharge Detection

Although ICCD can find out the location of the discharge, the severity of the discharge can only be expressed by the light intensity, which makes it difficult to conduct quantitative research on the discharge characteristics. Therefore, in order to carry out research on the partial discharge characteristics of IGBT modules, a partial discharge detection platform is designed and built.

The platform consists of a high-voltage AC/DC power supply, a coupling capacitor, detection impedance CPL1 and a partial discharge measurement and analysis system MPD. Partial discharge signals are susceptible to electromagnetic noise interference, so the entire test system is placed in an electromagnetic shielding room. The high-voltage source includes a voltage console, a PD-free transformer, and a protective resistor with a capacity of 5 kVA, an output voltage effective value of 0∼50 kV, and a current of 100 mA. The IGBT partial discharge measurement and analysis platform is shown in Figure 3. The background noise test results show that the background noise level of the built partial discharge test platform is maintained at 100 fC. In order to focus on the failure research, the packaging structure composed of DBC and silicone gel is used as the test sample to ensure the stability of the discharge.

2.3. Interface Charge Measurement

After locating the insulation risk point in the IGBT module, the propagation and dissipation of charge in IGBT power module much be elucidated, as they give rise to PDs and electric trees. Here, the simplified IGBT module is a DBC liner with silicone gel, and a silicone gel-DBC substrate interface charge dynamics measurement platform is built. The measurement principle and method refer to the existing research of the research group [18], which we only briefly introduce here.

The measurement system consists of three main components, namely the optical, electrical, and mechanical parts. The optical component serves as the centerpiece of the measurement system, enabling the measurement of interfacial charge. The electrical components, on the other hand, facilitate the functions of power sourcing, timing signal control, and collection of voltage, current, and partial discharge signals. Lastly, the mechanical components interconnect the IGBT module sample with both the optical and electrical components. Figure 4 illustrates a schematic diagram of the measurement system.

The principle of this method is the primary electro-optic effect, also known as the Pockels effect. It describes the relationship between the electric field and the optical phase difference, as shown in Equation (1).

$$\Delta \theta =\frac{2\pi }{\lambda }{n}_0^3\gamma {\displaystyle {\int }_0^l{E}_{\mathrm{z}}}dz$$

(1)

where:

• λ—wavelength of light in vacuum [m].
• n₀—refractive index without electric field.
• l—length of the pockels crystal in the z-direction of electric field, [m].
• γ—electro-optic coefficient, [m/V⁻¹].

This relationship is realized on the pockels crystal in Figure 4 in this manuscript, which is placed next to the sample. This arrangement allows the measurement of electric field distribution caused by discharged charges on the sample surface based on the phase difference of optical signal passing through the pockels crystal. Furthermore, optical elements such as a polarizing beam splitter (PBS) and 1/8 wave plate convert the light phase difference into light intensity, as shown in Equation (2).

$$\frac{I}{I_0}={\sin }^2(\frac{\Delta \theta +\pi /2}{2})$$

(2)

where:

• I—outgoing light intensity.
• I₀—incident light intensity.

Based on this measurement system, the electric field can be calculated from the light intensity. However, the electric field is not equivalent to the charge density, which requires an inversion algorithm, as shown in Figure 5 in the manuscript. According to the electric field simulation of the charge, the electric field distribution of the unit charge is calculated, and the transfer function is constructed with a Wiener filter. Finally, the charge density is obtained based on the electric field distribution monitored by the optical measurement system. Simulation calculations between unit charge and electric field distribution are carried out in the time domain. The construction of the transfer function and the Wiener filter are realized in the frequency domain based on the two-dimensional Fourier transform (2D-FT). It should be noted that this frequency relationship is applicable to cases in which the relationship between the charge and the electric field is not affected by the position. Here is an approximation. The final solution of the charge density distribution is the time domain solution.

It should be noted that the theoretical formula derivation, parameter information, range discussion, algorithm accuracy analysis, and validity verification of this method system can refer to our other research [18,19,20], and will not be introduced in detail here.

2.4. Electric Tree Observation

At a late stage of insulation failure, the observation of electric tree is helpful for understanding insulation failure in IGBT modules. Here, the simplified IGBT module is a DBC substrate–silicone gel structure, and the electric tree observation platform of the encapsulation structure shown in Figure 6 is established. The electric tree static experiment is carried out based on a microscope, and the shape of the electric tree is collected at a fixed time interval to achieve long-term failure process records. The self-healing observation of electric tree of the package structure is also based on the change in the morphology of the electric tree recorded by the microscope. The synchronization between devices in the experimental platform is as follows. The pulse signals of DG645 are divided into three groups. One is used to trigger the function generator and the high voltage amplifier outputs high voltage. One is used to trigger the oscilloscope (OSC) to record voltage and current signals, which are collected by high-voltage probes and HFCT, respectively. One is used to trigger the microscope to record electrical tree images.

3. Initial Failure Location

3.1. Sample Division

There are six ceramics in the IGBT module selected in this paper, and epoxy resin is encapsulated on the top. Therefore, the method of side area division is adopted to locate the discharge. The side observation area division and observation sequence are shown in Figure 7. The side of the module is divided into 10 observation areas according to the number of ceramics in the figure. It should be noted that in the process of observing one by one in accordance with the sequence of Figure 7, due to the continuous insulation degradation inside the module caused by long-term electric stress, the module broke down during the observation process in area 8, and as a result, the remaining three areas (S3-Left, S4-Left, S4-Right) could not be observed.

3.2. Discharge Location

The discharge area is shown in

Table 3. Discharge location in IGBT power module.

No.	Area	Discharge	Position	Points	Type
①	S1-Left	59.3 pC	Ceramic-PCB bonding wire	P1	Gel discharge
			Ceramic/PCB edge	P2	Interface discharge
②	S1-Middle	40.9 pC	Ceramic bonding wire	P3	Gel discharge
			PCB solder joints	P4	Solder discharge
③	S1-Right	61.6 pC	Ceramic bonding wire	P3	Gel discharge
			PCB edge	P2	Interface discharge
④	S2-Left	90.8 pC	Ceramic bonding wire	P3	Gel discharge
			PCB edge	P2	Interface discharge
⑤	S2-Right	62.3 pC	Ceramic bonding wire	P3	Gel discharge
		/	Ceramic edge	B0	Interface discharge
⑥	S3-Right	/	/	/	/
⑦	S3-Middle	63.2 pC	Ceramic-PCB bonding wire	P1	Gel discharge
			Ceramic edge	P2	Interface discharge

. The partial discharge measurement and analysis system, MPD, which has been introduced in Section 2.2 is applied here to assist in monitoring the discharge amplitude. The discharge amplitudes monitored by this system have been listed in Table 3 as a reference. Overall, no discharge was observed in area 6 (S3-Right), and discharges were observed at different locations in the rest of the area. In addition to the reported triple junctions (labeled as ceramic edge P2 in Table 3, the discharges of the bond wire and the solder joint were observed for the first time in this experiment. The bonding wire can be divided into the bonding wire (P3) on the ceramic and the bonding wire (P1) between the ceramic and the PCB, and the solder joint (P4) that generates the discharge is on the PCB. There is no doubt that the discharge at the edge of the PCB also belongs to a triple junction (P2). The discharge points are classified into three types: gel discharge, interface discharge, and solder joint discharge.

These discharges were observed in different areas, as shown in Figure 8, which lists the discharges of several typical areas, namely S1-Middle, S2-Left, and S3-Middle. In order to remove light noise, each area shows images of a normal environment, a darkroom with 0 kV, a darkroom with 10.2 kV, and discharge point marks. What needs to be explained is that ICCD records the number of photons, so the unit on the scale bars refers to the counts. This is related to the acquisition settings, specifically referring to the mode, number, TTL width, exposure time, and gain of the MCP in Table 2. This makes the scale bar of the normal environment inconsistent with that of the darkroom. This is because the former is used to clearly show the internal structure of the module, and is not used as a background for discharge. Therefore, the acquisition parameters are set with the highest resolution as the goal. This does not interfere with the discussion and analysis, because it is certain that the acquisition settings of no discharge (0 kV) and discharge (10.2 kV) in the darkroom are the same, which provides a premise for comparison of discharge analysis.

It can be seen that these areas cover all types of discharge points. Further, the partial discharge amplitude was recorded by the system shown in Figure 3. The discharge amplitude of P3 and P4 is about 40.9 pC, the discharge amplitude of P2 and P3 reaches 90.8 pC, and the discharge amplitude of P1 and P2 is 63.2 pC. This shows that the discharge amplitude of P2 is the highest; that is, the triple junction discharge is the most intense.

Although the new discharge points have been revealed, the final breakdown still occurs at the triple junction (B0 in Table 3), which indicates that the risk of triple junction failure is the highest. Therefore, in order to focus on research and control variables, the follow-up mechanism analysis no longer revolves around the IGBT module, but directly constructs a ceramic–silicone gel packaging structure with triple junctions and analyzes the interface charge and electric tree.

4. Interface Charge Kinetics

An interface charge kinetics experiment was performed on the IGBT module with a DBC substrate and silicone gel. The sample used for the experiment consisted of a 0.5 mm-thick Al₂O₃ ceramic layer and a 0.3 mm-thick copper layer on the DBC substrate, as depicted in Figure 4. An AC voltage with an effective value of 6 kVrms and a frequency of 50 Hz was applied to the sample. The temporal resolution of the experiment was set to 1000 frames per second with an exposure time of 1 ms, while the spatial resolution was determined by the object–image distance on the optical path and set at 1:2. Consequently, each pixel size was 6.75 μm. The system was triggered by the digital pulse delay generator, which produced effective signals for 100 cycles and triggered the power source, high-speed camera, and oscilloscope at t = 0 s. The results of the experiment are shown in Figure 9, which illustrates the interfacial charge densities at the four phases of 0°, 90°, 180°, and 270°.

As seen in Figure 9, the interfacial charges at the substrate–silicone gel interface are generated by the high electrical field applied to the electrode. These charges form a leakage current, which is detected by the current sensor on the ground wire. However, these signals are usually weak (at μA level in the early stage of failure) and appear as narrow pulses lasting only a few microseconds. These pulses represent the PD signal of interest in the insulation condition assessment of power modules and other devices. PD follows the charge movements: Charge injection from the electrode to the interface leads to rapid charge drift and diffusion, resulting in a violent discharge at peak voltage. The recombination of positive and negative charges near zero voltage eliminates the accumulated charge and activates the discharge simultaneously. The simultaneity of PD and charge movement suggests that PD reflects the one-dimensional time information of the insulation state in the IGBT module, while the monitored charge movement reflects the two-dimensional spatial–temporal information of the insulation state, revealing the failure state in the IGBT module.

5. Electric Tree Extend

Electrochemical deterioration, known as the electric tree, results from the interaction between the charged particles generated by partial discharges (PDs) and the material. An exemplary electric tree, induced by PDs, is presented in Figure 10, composed of filamentary channels and bubble-like cavities. This transparent and lighter-hued hollow channel is formed due to gas-filled cavities in the electric tree’s growth region, which are created by PDs.

Application of an AC voltage on the copper electrode triggers carrier injection and extraction, resulting in free charges and space charges formed from carriers that were trapped by the material’s traps. Free charges, propelled by the electric field, collide with the molecular chains of silicone gel, fracturing them into smaller molecules and releasing heat. At high temperatures, these smaller molecules undergo vaporization, aided by the combined air pressure and electrical stress, resulting in the formation of hollow, cracked channels that compose the filamentary channels of electric trees. Filamentary channels facilitate the movement of free charges, promoting diffusion and accumulation at the end. End expansion and vibration occur through the repulsive interaction of like charges, gas pressure, and elastic stress, ultimately forming bubble-like cavities at the end [21].

Figure 11 captures the growth of the electric tree over a 30 min period at an 8.5 kV AC voltage. The growth process of the electric tree involves rapid multiplication of filamentary channels, creating a tree-like structure with numerous branches. Accumulated charges gather at the branch tip, creating bubble-like cavities that extend over one or more branches, promoting branch channel development. The ends of the branches continue to form bubbles, reciprocating further development. The growth of the electric tree is a continuous process of accumulation and expansion.

6. Conclusions

In this paper, the discharge phenomena of the IGBT power module were observed based on ICCD. Furthermore, DBC ceramic filled with silicone gel was used as a test sample to study the discharge failure process, including the PD, interface charges, and electric trees. We believe this research will provide support for the analysis idea and guidance of the research method for the cause of power module failure. The mechanism of discharge failure was discussed and analyzed. The main conclusions are as follows:

(1)

Interface discharge, gel discharge and solder joint discharge occur in the IGBT module. The interfacial discharge originating from the triple junction is the most dangerous failure. The discharge of bond wires and solder joints is reported here for the first time.

(2)

The kinetics of charge propagation from triple junctions along the ceramic–gel interface is shared. It reflects the spatial distribution of the discharge current.

(3)

The electric tree consists of filamentary channels and bubble-like cavities. The electric tree is lighter in color and is a transparent hollow channel.