Partial Discharge Inception in Ceramic Substrates Embedded in Silicone Liquid, Silicone Gel, and Mineral Oil at Fast Voltage Rise and Sinusoidal Voltage

This article summarizes experiments that were carried out on insulating ceramic substrates and a needle-plane arrangement with the purpose of characterizing the inception of partial discharges (PDs) in the insulation of high-voltage semiconductors. PD measurement results obtained on substrates are correlated with the high-field preinception currents measured in a needle-plane electrode arrangement in liquids and gel. The tested substrates were embedded in either silicone liquid, silicone gel, or mineral oil. The PD inception was measured at sinusoidal and switched voltage of different polarities and rise times. For all test objects, the PD inception voltage (PDIV) is significantly lower under switched voltage compared to the sinusoidal voltage. Silicone gel and silicone liquid as insulating media surrounding the substrate are similar in terms of the PDIV. The PDIV is lower for substrates placed in mineral oil, particularly under switched voltage. Silicone gel and liquid are characterized by very low preinception currents measured at a high electric field. In contrast, a higher preinception current is observed in mineral oil. These results are consistent with the hypothesis that the high predischarge current in the more conductive mineral oil leads to space charge build-up in the high-field region that, by deforming the electric field, affects the PD inception probability.

I N ORDER to utilize the improved electrical and thermal properties of wide bandgap semiconductors such as SiC and GaN, a reliable packaging technology for power modules must be developed for higher electric and thermal stresses.Currently, the high-voltage insulation of power modules consists of three main components: insulating ceramic substrates, silicone gel filling, and the insulating housing [1], [2].Additionally, polymeric passivation materials can be applied [3].The use of nonlinear field grading materials in the triple point has been actively discussed [4], [5].Aluminum nitride is used for insulated gate bipolar transistors (IGBTs) with the highest power rating.The production of the substrates includes a step where parts of the copper conductors are removed by chemical etching.This procedure creates very sharp metallic protrusions along the contour of the copper conductors.During operation, electrical discharges may start at such protrusions due to very high local electric fields [6], [7].Such discharges may propagate deeper into the insulation along the interface between the ceramic and silicone gel.
Questions like the discharge inception in silicone gel and its degradation under partial discharge (PD) activity have been drawing the attention of scientists in recent years [8], [9], [10], [11], [12].Earlier investigations focused on the discharge inception properties in silicone gels in generic electrode arrangements such as needle-plane [13].The results of these tests are of high relevance in the context of power module insulation with razor-sharp protrusions as shown in the micrograph of a cross section of the triple point of the substrate in Fig. 1.Several experiments aimed at describing the inception and propagation of discharges along the interface between the silicone gel and ceramic materials [6], [14], [15].These investigations considered ac and switched voltage pulses with very high dV/dt values.It is seen that the PD inception voltage (PDIV) decreases with increasing voltage rise time for PDs originating in the trench of printed circuit boards (PCBs) [16].Characterization of electrical failure modes under different types of voltages is on the one hand important to understand phenomena where the presence of space charge in high-field regions plays a role.On the other hand, it is important to draw practical conclusions relevant for designing packaging for power electronic components.Furthermore, the results of studies can be used for reviewing the qualification test that currently specifies PD measurement and a high voltage withstand test under sinusoidal ac voltage of power frequency [17].
In this article, PD measurements were conducted on aluminum nitride substrate mockups embedded in silicone liquid, in silicone gel, and in mineral oil.With the sharp edges of the copper islands, electric fields in the insulations will be very high.Conduction mechanisms that may result in field grading and charge relaxation after PD has occurred should be considered.High field conduction in liquids and gel were therefore investigated.One issue that needs clarification is how well a substrate insulation model using silicone liquid represents real gel-insulation.
Conditions for discharge inception in gel/liquids are expected to depend on the shape of the applied voltage.With slow-rising voltages space charge limitation of the field may suppress discharge inception.Differences in discharge occurrence under sinusoidal and switched voltages are investigated, as well as the effects of varying slew rates for the switched voltages.
While power electronic modules are tested with bipolar sinusoidal voltages, the internal insulation on a substrate is stressed with unipolar switched voltages.Space charge formation and relaxation can be very different in the two cases.Hence, the effects of both bipolar and unipolar stresses are investigated.
Under sustained voltage stress the insulation system may become charged from conduction currents and from discharges, thus, reducing local fields.At voltage switch off these space charges may result in high electric fields that may initiate discharges.Varying the duration of the applied voltage (i.e., duty-cycle) is a tool for investigating this.Also, by changing the conductivity of the insulation one can reveal the relevance of such a charging mechanism.

II. MATERIAL AND SAMPLE PREPARATION
This study used model aluminum nitride substrates covered with either silicone gel; used in high voltage IGBTs, silicone liquid; Xiameter PMX-200 50 cSt, or mineral oil; Nytro 10XN.The substrate covered with silicone gel is a typical configuration, the silicone liquid is easier to apply in laboratory testing, and the mineral oil has a much higher current injection than the silicone gel and silicone liquid, which is assumed to influence the PD occurrence.The micrograph in Fig. 1 shows that a triple point was present.Here, field enhancement can trigger high-field phenomena, such as preinception currents.Highfield currents will introduce charges that alter the electric field and hence influence PD inception.High field currents in the selected liquids and gel were measured in a dedicated set-up.
The liquids were treated before use.The silicone liquid was filtered by a 20 µm glass filter, and the mineral oil was filtered by a 1 µm filter, whose pore size was not applicable to the silicone liquid.The silicone gel was mixed with liquids right from new bottles.The mineral oil was initially dry (measured initially by Karl Fischer titration to be 5 ppm), whereas the silicone liquid and gel were in equilibrium with the ambient humidity (∼20%RH), making an estimated water content of about 40 ppm in the silicone liquid [18].
The tested substrate mockups made of aluminum nitride were provided by a substrate manufacturer.The geometry of the substrates was specifically designed for investigating PDs.The ceramic plate with dimensions 24 × 50 × 1 mm was metalized on both sides by using the active metal brazing method.The copper plate on the upper side 15 × 15 × 0.3 mm has rounded corners.The profile of the metallization edge was shaped according to the design applied for real substrates.The substrates were soldered onto a copper base plate as shown in Fig. 2. The copper base plate was screwed on a metallic electrode in the bottom of the test vessel.A brass conductor lug was soldered on top of the upper conductor onto which an isolated copper wire was ball-soldered.The substrates were washed in isopropanol in an ultrasonic bath at 40 • C to remove the soldering paste and fat residuals.
The first group of the substrates was covered with silicone liquid under vacuum.The second group was covered with mineral oil in the same way, and the third group was molded in a silicone gel.The silicone gel was made by mixing two components in a 1:1 ratio.Samples covered with silicone gel were placed in a vacuum chamber at room temperature for 1 h before finalizing the curing at room temperature and atmospheric pressure for more than 12 h.

A. High Field Current Measurement Setup
The experimental setup consisted of a high-voltage step pulse generator, a point-plane test cell, a differential amplifier, and an oscilloscope.A detailed description of the experimental setup used can be found in [19].
The chosen electrode configuration for high-field current measurements of the selected liquids and gel consisted of a sharp (4.5 µm radius) needle and a spherical probe (D ∼ 1 mm) in parallel over a plane electrode (D = 9 cm) with a needle plane distance of 7 mm.The sharpness of the needle was comparable to the sharpness of the copper edge seen in Fig. 1 (3-4 µm).The needle electrode was made by electrochemical etching of a 100 µm diameter tungsten wire.The needle was controlled using a microscope before and after the experiment to ensure that it was not eroded during the experiments, see Fig. 3.
The impulse generator created step voltages with a rise time of ∼50 ns and a duration of 50 µs.Capacitors were used to integrate the current from the needle and the probe, as described in [19] and [20].The capacitors used were 480 pF, resulting in a total capacitance of 530 pF including the cables.A differential amplifier was used to subtract the displacement current from the conduction current.The setup was calibrated at low voltages where the conduction current can be neglected both at the probe and the needle electrode.This was done by adjusting the distances between the electrodes and the plane to get the same displacement current from the point and the probe; then the displacement current can be subtracted by the differential amplifier.The sensitivity of the setup is better than 0.2 pC.
The test cell was first rinsed with mineral oil to dilute and remove impurities from previous tests with different liquids.Then, 2 dL of the sample liquid was filled in the test cell.After the mineral oil tests, the same was done for silicone liquid.For the silicone gel, the test cell was rinsed by isopropanol and toluene before filling with the mixed liquid gel before curing at room temperature.

B. High Field Current Measurement Procedure
The negative pulses were increased until the first PD occurrence in the range from 5 to 10 kV, whereas the positive pulses were applied in the range of 7-16 kV.Three repetitions at each voltage level and a waiting time of 10 min between impulses were used to allow sufficient time for any space charge to dissipate.This time between impulses was based on previous experience [19].

C. PD Measurement Setup
The voltage was either sinusoidal or bipolar and unipolar switched voltage.Bipolar switched voltage pulses were generated by using a solid-state switch Behlke HTS 651-03 GSM push-pull amplifier, commutating between two capacitor banks charged to high dc voltage of positive and negative Unipolar switched voltages at 30 Hz repetition frequency and varying duty cycles were generated by using the same setup with one of the HV terminations grounded.The applied voltage could in all cases be varied in preprogrammed sequences.
The 30 Hz sinusoidal voltage was generated by a resonant circuit and the rest of the setup was similar to the switched voltage, as described in [3].

D. PD Measurement Procedure
The electrical PD detection at switched voltage pulses was complicated by the high displacement current flowing through the capacitive test object at the voltage switch on and off for each voltage pulse.The solid-state switch produced high-frequency current pulses from gate operation that also interfered with PD current signals.Although the displacement current could be eliminated by using a balancing capacitor as described in [3], the PD currents were too small to be measured with an acceptable signal-to-noise ratio using the high-frequency current transformer.Light detection of surface discharges was therefore chosen in this study.
PMTs have dark current pulses, which interferes with the smallest PDs.By using an oscilloscope to count PMT pulses within a narrow time frame, the probability of counting dark current pulses is low.This method fits well to switched voltages where the PDs occur close to or at the switching events.However, this method does not apply to sinusoidal voltage as the PDs occur during a much wider time span in the voltage period.Different approaches were therefore chosen to investigate PD occurrence and the influence of space charges.Fast switched voltages and slowly varying sinusoidal voltage both of 30 Hz were used to clearly show any difference for different voltage rise time on the PD occurrence.Only the steepest voltage raises were studied (45 ns-1 µs rise time) for the switched voltage to limit the number of parameters and focus on the difference between sinusoidal and switched voltage with respect to voltage raise time.
The high-sensitivity measurements were obtained by placing the PMT directly above the substrate.To obtain a high sensitivity of visible PDs on the substrate, a PMT Philips 56 UVP was placed vertically above and close to the sample.Detecting the low-intensity light from small discharges is problematic as these occur in the same region as the stochastic dark-current pulses from the PMT during continuous stresses.
Both bipolar and unipolar switched voltages were investigated.The unipolar voltage pulses used had a 50% duty cycle.The waiting time between voltage application was in the order of minutes.A 30 Hz switched voltage with varying rise time was applied to detect the smallest PDs.Additionally, measurements with 30 Hz sinusoidal voltage were performed allowing calibrated electrical measurements as a reference.The voltage was increased in steps of 200 V until a sustained PD activity was reached.Measurements were done on an oscilloscope within a short time window during voltage switch on (0 to applied voltage) and voltage switch off (applied voltage to 0).The time window was ten times the rise time to cover the part of the period where discharges are most likely to occur.Manual counting was done of 50 accumulated successive triggering events of the oscilloscope.For determining reproducibility, measurements were repeated for five different samples.
The sensitivity was then slightly reduced by studying the signals above the dark noise limit.The duty cycle of negative unipolar voltage was then varied from 5% to 95%.The voltage steps were increased to 2 kV to reduce the time the object was exposed to high voltage stress.This was done to reduce potential erosion of the sharp edges.Omicron MPD 600 was used to automatically record the optical signals from the PDs.This would detect whether space charges influence the PD on voltage switched off.
For the reduced sensitivity measurements, a PMT EMI 6255A was used together with a high-speed camera HiCam 500 with the purpose of comparing images and electrical PD measurements.The introduction of the camera required moving the PMT further away from the object, thus reducing its sensitivity.When setting the PD threshold above the dark current level commercial PD detection system (Omicron MPD600) could be used to detect the signals from the PMT, both for switched and sinusoidal measurements.At sinusoidal voltages conventional PD measurement could also be facilitated, showing that the sensitivity of the optical detection in this case was equivalent to 0.5-1 pC, also implying that the most sensitive measurements mainly measured PDs with magnitude well below 1 pC.
Both the 30 Hz sinusoidal and switched voltage PD detection setups were the same and the voltage was increased in steps of 500 V peak with a duration of 60 s.The measurements were repeated for five different samples.This enabled direct comparison between PD measurements performed at respectively sinusoidal and switched voltage.Omicron MPD 600 was used to automatically record both optical signals and Average measured current during the first 50-µs voltage application period as a function of applied voltage.
conventionally measured PDs.The camera was synchronized to the voltage and recorded pictures at 60 Hz, so one could differentiate between PD occurring at respectively positive and negative polarity.

IV. RESULTS
Throughout the study, PDs were mainly detected optically, thus only visible PDs originating in liquid or gel at the sharp metallization edges on ceramic substrates were recorded.The PDIV was measured using the cumulative PD occurrence probability as function of voltage.The 100% is when there is at least one discharge occurrence in each period.

A. High-Field Preinception Currents
The accumulated charge increased almost linearly with time for all voltages in mineral oil, indicating constant currents.The reproducibility of the recorded charge is good.The average preinception currents in mineral oil increased nonlinearly with voltage as is seen in Fig. 5, which indicates a nonlinear field dependence.The preinception currents for silicone gel are similar to silicone liquid, with hardly any preinception current measured.The measured currents for mineral oil are first reported and thoroughly discussed in [21].

B. PD Occurrence of High Sensitivity Measurements in the Voltage Range of 2-10 kV
The data points obtained from tests in this part on five different substrates with the high-sensitive setup showed significant scatter (Fig. 6).This is mainly due to the variation of curvature of the sharp edge of the metallization created by etching of the copper conductor.The standard deviation for each curve is sometimes larger than differences between the curves.However, the averaged curves identify systematic tendencies consistent with theoretical assumptions on which the following discussion will be based.
The PD inception probability as a function of peak voltage for negative unipolar voltage pulses and bipolar pulses are compared in Fig. 6 for substrates embedded in silicone liquid with voltage rise time of 400 ns.It is seen that the bipolar voltage reduces inception to about half of what was found for unipolar switched voltages.The big difference seen between Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.bipolar and unipolar voltages suggests that space charges are present close to the electrodes.The average 50% PDIV at negative polarity of the sharp metal edge is lower than at positive polarity by approximately 1 kV.
The inception probability of PDs on the rising edge of unipolar negative voltage pulses with varying rise times is shown in Fig. 7. Here, there is no significant difference between the different rise times in the rise time range from 45 to 400 ns.However, the mean values show a small tendency of higher inception voltage when increasing the voltage rise time.
There was a high probability for PDs when switching the unipolar pulses off with silicone liquids seen in Fig. 8 with voltage rise time of 400 ns.This indicates the presence of space charges.
The results in Fig. 9 show the PD inception probability as a function of voltage for unipolar negative voltage pulses for voltage switch on and voltage switch off for substrates covered in silicone gel.The main observations are that there is no difference when applying different voltage rise/fall time in the range of 45 ns-1 µs, similar to the silicone liquid (Fig. 7), and that there is a relatively high PD probability for voltage switch off, similar to what was seen with the silicone liquid.
There is no difference in the PD inception probability of small PDs for substrates embedded in silicone liquid from that of silicone gel, as seen in the direct comparison between the gel and liquid in Fig. 10 based on curves in Figs. 8 and 9.

C. Influence of Duty Cycle
During bipolar voltage application, the space charges of different polarities are injected with the varying electric field.During unipolar voltage, this is not the case, as there is no opposite polarity.Heterocharge field enhancement can consequently be disregarded.The results from different duty cycles for negative unipolar voltage pulses for both mineral oil and silicone liquid are shown in Fig. 11.The main observation in this figure is that the PD probability is higher for mineral oil than for silicone liquid during voltage switch on.The probability of having a PD at voltage switch off, given that there was a PD on voltage switch on the previous voltage change, was equal to without considering the condition: P(A|B) = P(A).Next, the probability for PDs during switch off for silicone liquid is low (10%-20%).For mineral oil, the PD inception probability increases when duty cycle is increased to 95%, while not for silicone liquid.This suggests that at a 95% duty cycle with a high conductivity liquid, there is time for conductive charging of the electrode before switching the voltage off.This establishes an electric field in the opposite direction that during switch off is sufficiently high for PD inception.Results in Fig. 8 were achieved with a sensitivity below the dark current floor, while for the duty cycle experiments in Fig. 11, sensitivity of the measurements was reduced to suppress dark current pulses.The probability for having PD on voltage switch off is smaller when having the sensitivity above the dark current floor.Hence, it is deduced that it is more probable to have small PDs than larger PDs at voltage switch off.

D. PD Occurrence of Reduced Sensitivity Measurements in the Voltage Range of 7-16 kV
The PD probability for having large (>1 pC) PDs in a period for substrates covered with either silicone liquid (low conductivity) or mineral oil (high conductivity) with 30 Hz sinusoidal, or 30 Hz bipolar switched voltage is shown in Fig. 12.The plots show that there are large variations within each test voltage type.The PD inception is much lower at switched voltages than at sinusoidal voltages.For mineral oil the PD inception at switched voltage is lower than for silicone liquid.There is a slightly higher PD probability for silicone liquid around 5 kV.However, this occurred for only one out of the five samples.The activity stopped when increasing the voltage.The root cause for this might be some impurities in the liquid, making a field enhancement when coming close to the electrode.The PD activity stops when the impurity has, e.g., vanished.This singularity should therefore be discarded when analyzing the results.
The large PDs are in general located at the electrode corners, as illustrated in Fig. 13 by the images of selected representatives of both positive and negative PDs out of 1000 successive polarity changes during 500 cycles in substrates covered with silicone liquid.These measurements were correlated to calibrated PD recordings.The PD amplitudes at switched voltage and negative sinusoidal are smaller in distribution and intensity than the positive sinusoidal PDs.
The images obtained at switched voltage were used to correlate the PD probability of the occurrence of the next PDs on the N following half cycles.The result is shown in Fig. 14 and illustrates that there is a high probability of having a PD during one of the next half cycles, indicating that presence of residual space charges from PDs increase the probability for further PD activity.As only single frames were used, no influence of weak electroluminescence was observed, contrary to when using a long exposure time and detecting electroluminescence as reported in [8].

V. DISCUSSION
The PD and high-field preinception current measurements provided a consistent basis for the interpretation of the physical phenomena that were observed in substrates embedded in silicone liquid/gel and mineral oil.The observed effects are mainly related to the preinception of high-field conductivity in the insulating liquid or gel in the high-field region around the triple point.The preinception current flow forms a space charge in the high-field region before the first discharges occur.The field limiting or field enhancing effect of the space charge, Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.depending on the voltage polarity, determines the probability of the discharge inception.This effect was more pronounced in mineral oil with higher conductivity than in silicone liquid or gel.

A. High-Field Currents and Field Change in Liquids and Gel
The high-field preinception currents were very low and about the same in the silicone liquid and silicone gel, while they were substantially higher in the mineral oil.Earlier studies have demonstrated that the preinception currents can vary [19], [20], [21].In general, currents will be higher for negative than for positive polarity.
The measured currents in Fig. 5 represent charge injection.For slow rising sinusoidal voltages, they will produce a homocharge that limits the electrical field around sharp edges.At the falling edge of a slow sine, these charges will have drifted away and will have little influence on the field in the following half cycle.For a bipolar switched voltage, the space charges formed in the first half cycle will not have had time to drift away when voltage polarity changes and then act as field enhancing heterocharges.These effects are expected to be larger for the mineral oil than for silicone oil and gel due to the higher conductivity.

B. PD Inception at Sharp Edges on Substrates
The study aimed to elucidate the effects of space charge and the influence of conductivity on PD behavior in the insulation system of models of converter substrates.Using optical detection limits observation to discharges at the edges of the copper islands.The camera images from Fig. 13 show that the PDs are predominately located at the copper electrode corner edges, where the electric field is highest.The exposure time of the camera was a little shorter than a period of 60 Hz, thus studying the half cycles of the applied 30 Hz voltage.This means that weak electroluminescence prior to the first discharges was not detected as it was in [8] with 40 s equivalent exposure time.
Comparing PD occurrence at sinusoidal and switched voltage is roughly a comparison between behavior at a very slow rise time (∼10 ms) to that at a fast rise time (∼100 ns).With a slow rise time, there is time for homocharges to be created during the voltage rise that limits the field at sharp edges that otherwise would result in high local fields.Contrary, at switched voltage, the homocharges have no time to drift away before polarity reversal, thus charge with the opposite polarity of the electrode (charges are then renamed to heterocharges), leading to field enhancement and lowering of PDIV.This explains the much higher PDIV for sinusoidal voltages than for switched voltages shown in Fig. 12.
The lower observed PDIV at switched bipolar voltages for mineral oil than silicone oil seen in Fig. 12, together with the higher measured preinception current for mineral oil support the hypotheses that the mineral oil enables more space charges to be injected and thus, a larger field enhancement after the fast voltage polarity reversal.This is also supported from the observation that for unipolar switched voltages the PD probability at voltage switch off for mineral oil with a duty cycle of 95% is much higher than for silicone liquid as shown in Fig. 11.The only possibility for triggering PDs when removing the external electrical field is that remaining space charges generate a field of opposite polarity.A duty cycle of 5% is probably too short to establish sufficient charge for a counter-field high enough to initiate PDs at voltage switch off.
When varying the rise time of switched voltages applied on silicone insulated substrate in the range 45-400 ns, there were no significant changes in inception probability as seen from Fig. 7. Silicone liquid shows a very low preinception current, so apparently all these risetimes are too short for sufficient space charge to limit the field to be formed.However, based on the results from sinusoidal versus switched voltage (Fig. 12), one would expect that there should be a transition rise time for which the inception voltage changes.A wider range of voltage rise times is required to throw light on this effect.
It is interesting to observe that the PD inception probability in Fig. 9 for small PDs on voltage switch-off is significantly higher (e.g., about 75% at 8 kV) than in Fig. 11 with brighter PDs (e.g., <10% at 8 kV).The main difference between the two figures is the sensitivity of the PMT measurement, with measurements shown in Fig. 9 having the higher sensitivity.This means that it is more probable to have small PDs for voltage switch off than larger PDs.In the tests shown in Fig. 8 with small PDs, it was observed that PDs at the voltage switch off occurred only as a result of occurring PDs at preceding the voltage switch on, leaving charges in the system that could lead to field enhancements when voltage was switched off.
The results in Fig. 6 are somewhat counter-intuitive when expecting the preinception currents to be close to nothing based on the current measurements shown in Fig. 5.If that conductivity can be neglected, then the only injection of charge can come from PDs, that deploy homocharges, as the normal charge conduction process will do.However, the figure shows that compared to unipolar voltages PD inception is reduced to when a bipolar voltage is applied.This indicates field enhancement from heterocharges that amplifies the field during polarity reversal.Apparently charge injection cannot be neglected.
Compared with the more conductive mineral oil, effects on the PD behavior related to the space charge formation were small in silicone liquid and gel.The results are consistent with observations made by other authors that silicone gel and silicone liquid as insulating media have similar PD characteristics [8].From a practical point of view, this result confirms that the silicone gel produces an adequate adhesion with the porous ceramic surface and fills well the microscopic pores around the sharp metallization edge.

VI. CONCLUSION
PD inception under switched voltage pulses on ceramic insulating substrates used in high voltage IGBTs were investigated.Silicone gel and silicone liquid used as encapsulating materials were examined for their preinception currents.Both experiments indicated very low preinception high-field currents in silicone gel and silicone liquid.For PDs in substrates embedded in these media, this result suggests that there is a very low influence of field-limiting space charge on the PD inception.With careful material and sample preparation, the PD inception levels are identical for substrates under silicone gel and silicone liquid.The PDIV was significantly higher at sinusoidal than at switched voltage.This gives an idea that the industrial standard test using power frequency ac voltage could lead to overestimation of the insulation performance of high voltage semiconductors.The results of tests at ac voltage should be considered with safety margins accounting for the physical effects of, e.g., space charges that directly influence the PD behavior at the triple points.

Partial
Discharge Inception in Ceramic Substrates Embedded in Silicone Liquid, Silicone Gel, and Mineral Oil at Fast Voltage Rise and Sinusoidal Voltage I. INTRODUCTION

Fig. 1 .
Fig. 1.Electron micrograph of the triple point on a ceramic substrate molded in epoxy before cutting showing the edge of the metal brazing and micropores in the ceramic.The gap between epoxy and substrate is artificial and due to sample preparation for microscopy investigations.

Fig. 2 .
Fig. 2. Photograph of an aluminum nitride substrate soldered on a copper plate.

Fig. 3 .
Fig. 3. Micrograph of the used needle.The point circle has a radius of 4.5 µm.

Fig. 4 .
Fig. 4. Schematic of the PD detection system for measurements at switching voltage pulses.

Fig. 5 .
Fig. 5.Average measured current during the first 50-µs voltage application period as a function of applied voltage.

Fig. 6 .
Fig. 6.PD inception probability as a function of applied peak voltage for substrate embedded in silicone liquid.The voltage rise time was 400 ns.The curves show mean values and standard deviation from five different samples.The sinusoidal was only tested until the first observed PDs, whose value is given with standard deviation.

Fig. 7 .
Fig. 7.PD inception as a function of peak voltage amplitude for substrates embedded in silicone liquid, measured at the rising edge of negative pulses with different rise times.The curves show the mean values from five different samples.

Fig. 8 .Fig. 9 .
Fig. 8. PD inception probability as a function of applied peak voltage for substrate embedded in silicone liquid for negative voltage pulses with 400 ns rise time.The curves show mean values from five different samples.

Fig. 10 .Fig. 11 .
Fig. 10.PD inception probability as a function of applied peak voltage for substrate embedded in silicone liquid and gel for negative unipolar voltage pulses.The curves show mean values from five different samples.

Fig. 12 .
Fig. 12. PD inception probability for sinusoidal and square voltage with rise time of 400 ns on substrates covered with either mineral oil or silicone liquid.The peak voltage was used.

Fig. 13 .
Fig. 13.Images of selected representative PDs in silicone liquid and sinusoidal voltage (17.7 kV) over a 500-period long recording.The 25 positive cycles and 50 negative cycles had PDs.The yellow lines represent the electrode surface.Images obtained with switched voltage appeared as light dots.

Fig. 14 .
Fig. 14.Probability for the next PD N half cycles later.At a voltage 9 kV for mineral oil (400/1000 half cycles with PD) and at a voltage of 11 kV for silicone liquid (100/1000 half cycles with PD) for the switched voltage.