Characterization and TCAD Modeling of the Lateral Space Charge Accumulation in Epoxy Molding Compound in Packaged HV-ICs

The reliability of high-voltage (HV) integrated circuits (ICs) can be significantly affected by space charge accumulation at the interface between the passivation layer and the epoxy molding compound (EMC) which acts as encapsulation material. The incorporation of moisture, which significantly increases the EMC conductivity, can lead to a stronger distortion of the electric field with a consequent breakdown instability. Moreover, the distance of the integrated-circuit active regions from the peripheral bond pads and wire would require a thorough optimization. To investigate the role played by the EMC under such conditions, a dedicated test chip made by an array of charge sensors covering short distances from bon-pads has been manufactured. A novel technique has been used to estimate the amount of space charge in the EMC independent of the bias applied to the bond pads. The outcome of the experiments has been explained by performing 2-D TCAD simulations of the structure under investigation which accurately account for the charge transport mechanisms of the EMC.


Characterization and TCAD Modeling of the Lateral Space Charge Accumulation in Epoxy
Molding Compound in Packaged HV-ICs Luigi Balestra , Member, IEEE, Elena Gnani , Senior Member, IEEE, Mattia Rossetti , Riccardo Depetro , Member, IEEE, and Susanna Reggiani , Senior Member, IEEE Abstract-The reliability of high-voltage (HV) integrated circuits (ICs) can be significantly affected by space charge accumulation at the interface between the passivation layer and the epoxy molding compound (EMC) which acts as encapsulation material.The incorporation of moisture, which significantly increases the EMC conductivity, can lead to a stronger distortion of the electric field with a consequent breakdown instability.Moreover, the distance of the integrated-circuit active regions from the peripheral bond pads and wire would require a thorough optimization.To investigate the role played by the EMC under such conditions, a dedicated test chip made by an array of charge sensors covering short distances from bon-pads has been manufactured.A novel technique has been used to estimate the amount of space charge in the EMC independent of the bias applied to the bond pads.The outcome of the experiments has been explained by performing 2-D TCAD simulations of the structure under investigation which accurately account for the charge transport mechanisms of the EMC.Index Terms-Charge sensors, epoxy-molding compound (EMC), high-voltage (HV) integrated circuits (ICs), TCAD simulations.

I. INTRODUCTION
E POXY molding compounds (EMCs) have become an essential component in the manufacturing process of high-voltage (HV) integrated circuits (ICs).The superior mechanical and electrical properties of epoxy-based materials have made them a popular choice for encapsulating ICs in a protective shell [1].The use of EMCs in HV ICs has gained significant attention in recent years due to the increasing demand for miniaturization and high-speed performance.This has led to the development of new and advanced epoxy-based materials with a large fraction of insulating fillers that match the thermal expansion of the silicon die and suppress the moisture permeability.Such outstanding properties make them suitable to withstand the HV and thermal stress associated with such applications [2], [3].However, the presence of small amounts of free carriers, ions, or moisture can still lead to electric field distortion at the EMC/passivation interface due to space charge accumulation with a consequent breakdown instability of the underlying power electronic devices [4].It is known that the region close to the HV bond pad, due to the presence of a high electric field, could be the most critical from the reliability point of view.Thus, it is fundamental to properly predict the space charge accumulation in the proximity of such regions.Moreover, it has been recently found that charge injection from the electrodes of test capacitors made with wet EMCs gives rise to a transient phenomenon increasing the current during prolonged exposure to HV stresses [5].It is thus extremely important to measure wet EMCs in similar conditions also in test chips where the charge transport in EMCs directly influences devices underneath.In previous works [6], [7], dedicated test structures made by integrated charge sensors have been manufactured to directly measure the amount of space charge accumulated in the EMC.The main interest was to monitor the charge spreading in the center of an IC.Thus, sensors were placed at a distance of a few hundreds of µm from the bond pads, and experiments were assumed to be totally independent of the detailed geometry of the bond ball, the available models clearly predict that the IC malfunctioning due to charge spreading might be limited by placing critical active areas far from the bond pads, or by shading them with properly optimized field plates, leading to large area occupation, and are not suited to thoroughly address the full packaging structure design.In addition, in [6] and [7] only the low-field transport regime has been investigated since no more than 600 V have been applied to the embedded HV metallization placed at a distance of 200 µm from the ground one, thus limiting the maximum electric field experienced by the EMC on top.In this work, structures with the sensors located at a few µm from the electrodes have been manufactured.Moreover, a maximum voltage of 1800 V has been applied to bond wires as close as 23 µm.By doing so, the electric field distribution in the EMC region of the structure under investigation is very similar to the one encountered in integrated HV capacitors for galvanic insulation which shows a high electric field in between the HV pad and the guard ring [8].By collecting the current flowing through the charge sensor, the amount of charge close to the HV electrodes in the high-field regime has been estimated.Furthermore, the transient effects of decreasing the voltage above the transistors and subsequently exposing them to the stress bias might give rise to unexpected residual space charge leading to temporary malfunctions.To investigate the residual amount of space charge when the stress is removed, a stress-measure-stress procedure has been applied to the charge sensors giving the possibility to discover new interesting EMC properties.The explanation of the observed features has been carried out by performing 2-D TCAD simulations which clarify the role played by different physical phenomena affecting the charge transport and the space charge accumulation in the EMC.

II. EXPERIMENTAL SETUP AND TCAD MODELING
The study of charge accumulation effects at the EMC/passivation interface has been carried out by using the test chip shown in Fig. 1 (top).It consists of two charge sensors realized as long-channel silicon gate-less transistors with shallow-trench insulation (STI) in the channel region and the lateral extensions.On top, a nitride-rich silicon nitride layer is grown together with the HV and ground (GND) bond pads embedded in the EMC.To calibrate the simulation setup against experiments on the adopted technology, a third transistor, realized with the same masks of the sensors, has been included below the HV contact, so to characterize it by using the HV contact as a gate (calibration MOSFET).In Fig. 1 (bottom right), a magnified view of MOS1 is shown to explain the charge detection mechanisms of the sensors: the electrostatic potential above the STI region is enhanced by the presence of positive space charge in the EMC leading to a stronger electron channel formation with a consequent higher current i(t) flowing through the source and the drain contact.The different currents of the sensors MOS1 and MOS2 thus allow us to measure the amount of charge accumulated on top of each of them and its dynamics.All devices have been dried at 100 • C for 48 h to remove preexistent humidity.To evaluate the role of moisture on the space charge formation in EMC, some of them have been exposed to 85 • C/85 RH% conditions in a climatic chamber for 408 h.According to [11], a significant water uptake can be assumed.Two different characterization techniques have been employed: 1) the drain current of both sensors is measured versus time (i(t)), while a dc stress voltage is applied to the HV contact and 2) the measure-stress-measure approach is adopted by measuring i(t) when V HV = 0. Fig. 2 schematically shows a stress cycle.The red dot indicates the instant of time at which the sensor currents are measured.In the first case, the charge accumulation on each sensor is monitored in time.In the second case, the residual space charge and its redistribution in time is detected.All i(t) measurements have been performed at room temperature and by fixing V DS = 1 V at each sensor.TCAD simulations of the test chip under investigation have been calibrated by assuming the presence of defects at the Si/SiO 2 interface which have been modeled as acceptor trap levels allowing to properly reproduce the I D -V HV characteristics of the calibration MOSFET [Fig. 1 (bottom left)].Following the literature [5], [7], [9], conduction mechanisms in EMC are described by means of the drift-diffusion transport model with Schottky boundary conditions to allow charge injection from the electrodes into the material bulk.Moreover, the physical effect of the barrier lowering has been included to reproduce the field-driven charge injection, which is a fundamental ingredient to explain the transient phenomenon of the current increase in EMC interdigitated capacitor under wet conditions [5].As far as the transport properties of the EMC material are concerned, the Poole-Frenkel mobility model has been used since it is the most suitable to describe the field dependence of the conductivity in EMC [5], [7].The oxide layer has been modeled as an ideal insulator accounting for a specific value of the dielectric constant.The same considerations can be applied to the nitride layers which, due to the high nitrogen concentration, is characterized by a very large resistivity [10].Thus, as a first approximation, no charge leakage is assumed within the passivation material.
Despite the holes would require a significant amount of time to move through the EMC and be detected from the two charge sensors, large values of current are measured for very low-stress times.Such feature can be partially ascribed Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.

Fig. 2.
Schematic representation of the stress-measure-stress sequence.The red dot shows the instant of time at which the sensor's current is recorded.
to some additional charge build-up in the passivation stack or to polarization effects which have not been accounted for in this work.For this reason, the presence of a positive fixed charge at the EMC/nitride interface has been assumed in TCAD simulations.This would provide the desired value on the sensor current for t ≈ 10 s.
The field dependence of the Schottky barrier and the Poole-Frenkel mobilities modify the electrostatic potential profile at the EMC/passivation interface leading to specific features in the experimental curves as explained in the following.

III. RESULTS AND DISCUSSION
The two different stress-measurement techniques allowed us to deeply study the role played by the space charge accumulation in the test chip.The TCAD analyses of the PF carrier mobility and Schottky-barrier injection inside the EMC lead to the exhaustive prediction of the experimental data.

A. Poole-Frenkel Mobility
According to [13] and [14], the PF mobility is expressed as where µ 0 is the low-field mobility, E a is the activation energy, k B is the Boltzmann constant, T is the absolute temperature, E is the absolute value of the electric field, and β, from a theoretical point of view, is a function of the relative dielectric constant of the material.The activation energy E a has been fixed to 0 eV in this work since the temperature dependence of the EMC conductivity has not been investigated.The same µ 0 is assumed for electrons and holes, while differences are attributed to β.To clarify the role of the EMC mobility, TCAD simulations that emulate the characterization technique (a) have been carried out.Charge sensors are turned on by a net positive charge accumulation in the EMC.The sensor closer to HV (MOS1) turns on faster and with a larger steady-state current than the second one (MOS2).In Fig. 3(a), simulations performed with different values of µ 0 in the EMC show that the low-field mobility affects the time needed to charge the sensors.However, neither the value of the steady-state current nor the slope of the i(t) curves are modified.Fig. 3(b) shows the role of β: it affects the slope of i(t) without significantly changing the value of the saturation current.This can be justified by analyzing the electric field profile along the EMC/passivation interface for different times [Fig.3(c)]: the electric field close to the HV reduces with time due to charge built-up.As a consequence, the PF mobility is very high at the stress onset but decreases with time, leading to a reduction of the current slope.In Fig. 3(d), the i(t) curves of TCAD simulations with different β for electrons and holes are reported.They allow us to clarify the role played by positive-negative charges in the whole time range leading to different features in the i(t) slopes.
The electron and hole concentrations on top of the different sensors are significantly influenced by the choice of β, as explained in Fig. 4.
In Fig. 4(a), coordinates indicated by arrows as x 1 and x 2 show the position at which the free carriers concentration is recorded.In Fig. 4(b), the electron (n(x 1,2 , t)) and hole ( p(x 1,2 , t)) concentrations are reported for β e = 4, β h = 2. Due to the large value of β e , electrons are immediately extracted from the HV electrode and, as a consequence, n(x 1 , t) drops abruptly for 0 < t < 300 s.For larger times, electrons coming from the GND contact give rise to the observed increase of n(x 1 , t).The same considerations apply to n(x 2 , t), however this effect is less pronounced due to the larger distance from the HV contact and the presence of a lower electric field.In the same way, holes are injected from the HV contact leading to the increase of p(x 1,2 , t) with time.Fig. 4(c) shows n(x 1,2 , t) and p(x 1,2 , t) for β e = 2, β h = 2. Due to the reduced electron mobility, the drop of n(x 1,2 , t) at low-stress times is almost negligible.This corresponds to a lower amount of net positive space charge (p-n) above the sensors with a consequent reduction of the drain currents of the drain currents in the first part of the transient behavior, as shown in Fig. 3(d).Finally, Fig. 4(d) shows n(x 1,2 , t) and p(x 1,2 , t) for β e = 4, β h = 0.In this case, the electron mobility is as high as for case (b), leading to a fast extraction of electrons and then a large current for a small stress time.However, due to the low hole mobility, the positive charge built-up is delayed in time leading to a plateau in the sensor current for 300 s< t <3000 s.For very large stress times, the steady-state condition is achieved and the corresponding saturation current reaches approximately the same values.It is worth noting that the large electron concentration observed at t = 1 s is due to the positive fixed charge at the EMC/passivation interface which attracts electrons independent of the stress bias applied to the HV electrode.This is not caused by injection from the GND contact.The parameters µ 0 and β can then be used to nicely reproduce the time shift and slope of the i(t) curves.

B. Schottky Injection
The Schottky barrier at the electrodes is modeled as where 0 is the difference between the metal work function and the electron affinity of the EMC while β S can be used as a fitting BL parameter [15].To clarify its role, TCAD simulations that reproduce the characterization technique (b) have been implemented.During the stress phase (high V HV ), electrons and holes are injected from GND and HV, respectively.On the contrary, when the stress is removed (V H V = 0), the effect of the residual space charge distribution is observed.More specifically, due to the residual space charge distribution [Fig.5(a) and (b)], holes are freely extracted from both electrodes, while electron injection is affected by the Schottky barrier.By accounting for BL, a larger amount of holes and electrons can be injected during stress and measurement, respectively.At high-stress voltage (V HV = 1800 V), a significant accumulation of positive charge during the stress is observed.With BL, a larger amount of electrons is injected with V HV = 0 V, leading to reduced i(t) at large-stress time [Fig.5(c)].For V HV = 900 V, positive space charge in the EMC and electron injection at V HV = 0 V are small thus limited i(t) reduction due to BL is observed at large-stress time.In this case, the enhanced injection of positive charge during the stress phase dominates the electron injection during measurement.For this reason, i(t) is higher when BL is considered [Fig.5(d)].In Fig. 6, simulations are compared with experiments performed on dry samples with both techniques (a) and (b) at V HV = 1800 V. Similarly, experiments and simulations on wet samples for different values of the stress voltage V HV are reported in Figs.7 and 8.
It can be observed that in the presence of humidity, a larger current is detected from the charge sensors.In addition, Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.a reduction of the time needed to achieve the steady-state condition can be observed.These two features are modeled assuming a larger injection from the electrodes and an enhanced EMC mobility, respectively.
It is worth noting that all simulations in Figs.7 and 8 have been performed with the same parameter set which qualitatively reproduces the field dependence of the EMC properties.At the same time, each measurement has been carried out on different samples.This, due to the large variability of the EMC properties, might explain the slight difference between the measured and simulated currents with technique (a).On the contrary, the significant discrepancy observed by comparing measurements and simulations performed with technique (b) at a low electric field can be attributed to an oversimplified approach based on PF mobility.A more sophisticated model based on the presence of traps in the EMC bulk will be addressed in the future.The proposed TCAD approach then reproduces the main features of the characteristics, allowing us to estimate the amount of space charge at the EMC/passivation interface which is fundamental to predicting the reliability issues in encapsulated HV-ICs.

IV. CONCLUSION
In this work, the evolution with time of the lateral space charge accumulation in EMC has been investigated by using a dedicated test chip.Two different characterization techniques have been employed to estimate the residual space charge in the encapsulation material.The outcome of this study is the explanation by TCAD simulations of the role played by the different physical phenomena affecting the EMC charge transport.This work provides a useful tool for the reliability investigation of encapsulated HV-ICs.

Fig. 1 .
Fig. 1. (Top) Schematic representation of the test structure under investigation (not in scale).(Bottom right) Magnified view of the MOS1 charge sensor.Positive charges accumulated in the EMC enhance the electron density in the sensor channel providing a higher current i(t) flowing through the drain and source contacts.The same consideration can be applied to MOS2.On the contrary, the channel formation in the calibration MOSFET is due only to the bias applied to the HV electrode.(Bottom left) I D -V HV curve of the calibration MOSFET.

Fig. 3 .
Fig. 3. (a) Simulated current of MOS1 and MOS2 with different values of the low field mobility µ 0 .(b) Simulated current of MOS1 and MOS2 with or without Poole-Frenkel effect.(c) Corresponding electric field profile close to the EMC/passivation extracted at different stress times.(d) Simulated current of MOS1 and MOS2 with different values of β.V HV = 900 V.

Fig. 7 .
Fig. 7. Simulated and measured current extracted from the charge sensors of wet samples according to the characterization technique (a) as a function of time for different values of the applied stress voltage.

Fig. 8 .
Fig. 8. Simulated and measured current extracted from the charge sensors of wet samples according to the characterization technique (b) as a function of time for different values of the applied stress voltage.