Methodology for the investigation of threading dislocations as a source of vertical leakage in AlGaN/GaN-HEMT heterostructures for power devices

Due to its outstanding properties, GaN is one of the most promising materials for power electronics and high frequency applications and therefore has been a subject of intense research over the last few years.^1–4 Most commonly used devices are AlGaN/GaN high electron mobility transistors (HEMTs), making use of the piezoelectrical nature of GaN and the formation of an electrically highly conductive two dimensional electron gas (2DEG) located right below the heterointerface.⁵ Electron sheet densities up to the order of 10¹³ cm⁻² are obtained by using AlGaN-barriers with an Al-content of typically 20%–25% without any further extrinsic doping being necessary.^5,6

Especially when grown on Si, a large limiting factor of AlGaN/GaN-HEMTs is the reduced reliability compared with pure Si-devices, which becomes evident in both permanent and reversible degradation mechanisms.^7–10 Charge carrier trapping by point defects^11,12 and the presence of leakage current paths through extended defects like threading dislocations (TDs) are big issues.^13–19 A huge effort has been spent on reducing the dislocation density on the one hand and on providing a vertically good electrically insulating behavior of the whole layer stack on the other hand. Today, this is achieved by growing thick multi-layered buffer structures to adapt the mismatched crystal lattices and thermal expansion coefficients of Si and GaN and by growing C-doped GaN-layers for compensating the intrinsic n-type behavior of GaN to a semi-insulating one.^20–22 However, many questions regarding leakage and breakdown are still under discussion. What role do specific types of dislocations play? What is the relationship between electrical properties of dislocations, the growth regime of their host material, and the presence of impurities therein? How does vertical leakage current penetrate through the highly defective and microstructurally inhomogeneous layer sequence between the active device region and the Si-substrate and what is its relation to vertical breakdown?

Pure screw and mixed type dislocations have been identified to act as leakage current paths and to directly increase reverse bias leakage of Schottky diodes.^13–19 It is also known that dislocations can getter impurities like, for example, carbon, in the C-doped GaN-buffer and hence induce n-type carbon depletion zones, which may degrade the breakdown performance of a transistor²³ and influence its dynamic behavior, which was recently confirmed by theoretical calculations.²² Conductive interfaces or layers have been identified to be critical with regard to leakage too.²⁴ The above mentioned questions are especially difficult to access if direct evidence of how material defects influence device characteristics is desired. A direct correlation of those electrical properties with specific structural disturbances on a sub-micrometer scale is necessary. However, the dislocation density of AlGaN/GaN-HEMT structures grown on Si is typically in the order of 10⁹ cm⁻², i.e., many dislocations are covered by a conventional device.²⁵ This makes it generally difficult to perform such sophisticated investigations.

In this work, we present a strictly place-to-place correlative approach from a pure material’s point of view. Direct evidence is given as to what dislocation types are most critical with respect to vertical leakage in our investigated AlGaN/GaN-HEMT heterostructure grown on Si, and the distribution of those dislocations in the device relevant region around the 2DEG is analyzed. The presented work-flow and results are important for upcoming studies involving test structures and for directly drawing conclusions from the underlying material about the device characteristics.

In our study, we investigated an AlGaN/GaN-HEMT heterostructure for power devices grown on p-doped Si(111) by metal organic chemical vapor deposition (MOCVD). The buffer is built up of unintentionally doped step-graded AlGaN strain-relief layers with a uniformly C-doped GaN-layer for charge carrier compensation on top, where the carbon concentration is 4 × 10¹⁹ cm⁻³. The GaN-channel layer is capped by an AlGaN-barrier with 25% Al and a thickness of 20 nm. The surface is passivated with a 4 nm GaN-cap. Two further samples with a similar buffer structure, but with a lower C-doping in the range of 10¹⁷ cm⁻³ were investigated: one sample was grown with AlGaN-barrier, the other one without.

To address the above-mentioned questions, we applied several characterization methods, namely atomic force microscopy (AFM) in tapping mode, conductive AFM (C-AFM), cathodoluminescence (CL) imaging, energy dispersive X-ray (EDX) analysis, defect selective etching (DSE), and scanning electron microscopy (SEM). All methods were used in direct correlation on the same area of the sample. Before the investigation was started, the as-grown sample was wet chemically cleaned. Reference markers for orientation were set by carefully performed nano-indentation in a distance of several 100 μm to the region of interest in order to prevent the influence of extrinsically induced dislocations on the results. First Tapping-AFM was measured with a Bruker Dimension Icon as a method to obtain highly resolved topography images. After that, C-AFM was carried out using boron doped diamond tips. A bias of several volts was applied between an Al-contact on the Si-substrate and the grounded AFM-tip in order to map differences in vertical leakage current or vertical leakage current paths, respectively. Thereafter, CL-measurements were done with a Gatan MonoCL3 system attached to a Jeol JSM-7500F for mapping of non-radiative recombination centers on a sub-micrometer scale at room-temperature. For selective elemental determinations, EDX was performed with an X-Max Detector from Oxford Instruments also attached to the Jeol JSM-7500F. Defect selective etching was realized in KOH/NaOH eutectic melt at 450 °C for 4 min with sample backside protection. The etched sample was analyzed by SEM.

The result from topography measurements by Tapping-AFM is depicted in Fig. 1 and is typical for GaN-capped AlGaN/GaN-HEMT heterostructures: flat terraces separated by atomic steps are indicative for step flow growth. Two types of depressions in terms of diameter and depth can be observed on the surface. The larger ones, located at terminating step edges, have a diameter of 50–70 nm and a depth of typically around 0.5 nm, which corresponds to the height of a full GaN unit cell. The smaller ones, mostly aligned on top of terraces, have a diameter of around 30 nm and a depth of 0.25 nm or half a unit cell. It is known from theoretical calculations that dislocations which meet the crystal surface can form depressions. The geometry of the pits depends on the absolute value of the corresponding Burgers vector.²⁶ Small depressions are expected to form at locations of threading edge dislocations (TEDs), whereas medium ones are related to threading screw dislocations (TSDs) and large ones to mixed types. From the Burgers vector analysis, it can be estimated that the ratio of diameters of mixed type dislocations to TSDs is around 1.4 and that of TSDs to TEDs 2.7.²⁶ The fact that our observed small depressions appear on flat terraces but not at terminating step edges supports the assumption of them corresponding to TEDs and all observed large depressions to dislocations with a screw component. The variation in diameter within the group of large depressions is around 1.4 and therefore fits well to the theoretically expected difference between the depression diameter of mixed type dislocations and TSDs. However, we cannot distinguish between them which is likely due to the following facts: (i) considering that there is always a convolution of surface topography and tip geometry, the expected difference in diameter might be too small to be resolved; (ii) the absolute screw and edge portion of a mixed type dislocation is not fixed; and (iii) natural deviations of the depressions’ diameters from theoretical values. The density of large depressions, i.e., the density of dislocations with a screw component, was determined to a value of $ρTSD+mix=1.50×109cm−2$ on an area of 20 × 20 μm² around the region shown in Fig. 1. The surface density of small depressions or TEDs cannot be determined reliably, as their geometry is close to the resolution limit of the measurement setup and eventually not all of them can be detected.

In Fig. 2(a), a representative result from C-AFM measurements with a vertically applied bias of −3 V on the Si-substrate relative to the grounded AFM-tip can be seen: clearly defined discrete spots of locally high vertical current flow are observed with a density of around $ρspot=4×108cm−2$⁠. The peak currents are in the range of several 100 pA, whereas the background current in the surrounding of the spots is near to zero and only fluctuating by some pA. The diameter of the spots varies between 100 nm and 300 nm. It has to be mentioned that the observed absolute values of the current can strongly depend on the parameters used like deflection set point as well as on hardly controllable facts like locally different surface oxidation, water adhesion, contamination, specific tip geometry on an atomic scale, contact area, and pressure, all of which additionally may change over time. A part of the region investigated in Fig. 2(a) is enlarged in Fig. 2(b) and corresponds to the same area as analyzed by Tapping-AFM in Fig. 1.

A direct comparison of the results from C-AFM with those from Tapping-AFM shows that leakage spots are always located at large depressions, i.e., at terminating step edges, but not vice versa as can be also noticed by comparing $ρTSD+mix=1.50×109cm−2$ to the density of leakage spots of $ρspot=4×108cm−2$ (see Fig. 1). Many depressions at terminating step edges exhibit no electrical activity at all. This indicates that the observed current is not an artefact caused by the topography of the depression itself, but by the local electronic properties of the underlying dislocations with screw component. TEDs remain electrically inactive within the whole observable range of ±10 V in the vertical setup.

An experimental verification of the predicted relation between depressions in the surface and dislocations themselves or types of dislocations, respectively, can be performed in two ways, namely by CL and DSE.

In Fig. 3, a panchromatic CL-mapping of the investigated region can be seen. The observed contrast is a measure for the difference in local light-emission of the sample as a result of the electron beam-material interaction. Dark areas correspond to regions of a locally higher ratio of non-radiative recombination and therefore to a higher density of deep-level defects. An acceleration voltage of 3 kV was used and the information depth was simulated in order to verify that mainly near-surface deep-level defects are observed.²⁷ The results show that all resolved surface depressions correspond to dark regions in the CL image, which is in good agreement with the findings from Rosner et al.²⁸ However, the CL-contrast reveals mainly the distribution of deep-level defects like dislocations. It is not possible to distinguish between defects like the dislocation types. The total defect density is difficult to determine or subject to large error under the given high amount of dark spots. The dislocation density in the order of 10⁹ cm⁻² in combination with the diffusion length of the injected excess charge carriers leads to extensive overlapping of the dark spots and to the formation of dark lines instead of single spots. The advantage of panchromatic CL is, however, that it is possible to map the distribution of deep level defects non-destructively also on much larger scales as shown in Fig. 3, i.e., one can easily determine the size of a statistically relevant area that is representative for the whole sample. One finds a value of typically around 20 × 20 μm² in our case.

The result from DSE observed by SEM is shown in Fig. 4. Hexagonal etch pits of roughly three groups in terms of diameter have been found. In previous observations, it was clarified by transmission electron microscopy (TEM) calibration that large pits form at TSDs, medium pits at mixed type dislocations, and small pits at TEDs.^29,30 The density of the different types of etch pits is only easy to determine for the large ones, i.e., TSDs, as the diameter of etch pits corresponding to TEDs and mixed type dislocations often differs not much and is therefore not completely distinguishable in a reliable way. On a 20 × 20 μm²-scale around the area in Fig. 4, one can find a value of $ρTSD=7.39×107cm−2$⁠, which allows to calculate $ρmix=1.43×109cm−2$ using the results from Tapping-AFM. It is noteworthy that there are flat plateau-like structures remaining between the etch pits, i.e., in dislocation free areas. Selective EDX with 3 kV shows an Al-signal unique to the location of these plateaus. Together with Tapping-AFM on the etched surface, it was confirmed that these plateaus are un-etched remnants of the GaN-cap and AlGaN-barrier. The plateaus exhibit a height difference to their flat surrounding of around 20 nm, which is near to the expected thickness of the AlGaN-barrier and GaN-cap together. An overlay of the etch pit locations and the CL contrast of the region before etching gives an excellent agreement which can be seen in Fig. 5. Hence, the observed CL-contrast of the un-etched sample indeed reveals the dislocation structure of the GaN-channel layer directly below the AlGaN-barrier. This is also expected because of the generally lower luminescence of the barrier compared to GaN. The used etching conditions, therefore, were suitable to visualize all dislocations within the GaN-channel layer. Due to the fact that all etch pits exhibit six tilted side planes, they always terminate in a little dark spot at the bottom of the pit when observed by SEM with an in-lens detector. This is also the case in the presence of strong etch pit overlapping like especially observed for TEDs, which typically appear closely aligned chain-like with a distance of often only some nm. Therefore, it was possible to determine the total dislocation density to $ρtot=5.74×109cm−2$ by analyzing 20 × 20 μm² around the area shown in Fig. 4. Hence, the density of TEDs can be concluded to a value of $ρTED=4.24×109cm−2$⁠. A value of 1%–2% of all dislocations being TSDs and 25% being mixed type dislocations is realistic for GaN-on-Si grown by MOCVD. A summary of all obtained values is depicted in Table I.

A comparison of the results from DSE and the topography before etching shows that all former observed little depressions on atomically flat terraces transformed into small etch pits or TEDs, respectively, and that there is a one-to-one correspondence between etch pits of medium and large size to former depressions at terminating step edges. This shows that no detectable difference in the amount of TSDs and mixed type dislocations between the regions below and above the 2DEG is present. As not all TEDs reaching the GaN-cap surface could be resolved by little depressions in the topography before etching, no direct statement can be derived about the difference of their density between GaN-channel layer and the layers above, but it is assumed to be very similar as well.

The key question of which dislocation types are most critical regarding vertical leakage can be addressed when looking at Fig. 4. Prominent leakage spots correlate with large as well as with medium sized etch pits, i.e., with TSDs and mixed type dislocations. However, there are some TSDs and many mixed type dislocations showing hardly or even no leakage at all. The fact of dislocations with screw component being critical regarding serving as leakage paths is generally consistent with former studies.^13–19 However, our result shows that even within one group of dislocation types, there are distinct differences in the electrical properties with the exception of TEDs. These results were reproducibly obtained on several different regions of the wafer and are mainly possible to be achieved only by good statistics and the direct correlative application of the various methods, which is a main difference compared to former studies.

The electrical activity of a dislocation is a widely discussed topic.^31–35 Decisive for the electrical properties of a dislocation is its energy level within the bandgap and the atomic configuration of the dislocation core. No material is pure, i.e., impurities or dopant atoms attracted by the strain field of a dislocation can decorate its core, change its atomic configuration and the electrical properties. Thus, the electrical nature of a dislocation can vary regarding its host material in general, the specific growth regime, and local inhomogeneities. Leakage through TSDs and mixed type dislocations was shown to be strongly influenced by growth stoichiometry¹³ and dependent on local chemical and structural changes.¹⁴ Moreover, TSDs in MOCVD-grown GaN can, e.g., exhibit a closed and an open core, and mixed type dislocations can differ significantly in terms of the screw and edge portion as discussed earlier. Also, the absolute value of the Burgers vector can vary significantly within one dislocation type. Furthermore, the point of origin of a dislocation within the layer stack is important. Dislocations forming at the AlN nucleation layer and meeting the sample surface might form a vertically conductive path through the whole buffer, whereas dislocations originating in the C-doped GaN-layer do not. The fact of dislocation type transformation might also play an important role. All this shows the difficulty of a theoretical understanding of dislocations’ electrical properties and also addresses the occurring differences even within one certain type of dislocation in our investigated material. We could not distinguish between mixed type dislocations with different Burgers vector as DSE on a highly defective material is not an easy task to be performed that selectively. The same holds true for the difference between closed and open core screw dislocations. Due to our results, it is likely that the atomic configuration of the dislocation cores varies significantly within the group of TSDs as well as within that of mixed type dislocations.

It has been shown that reverse bias leakage of Schottky diodes based on AlGaN/GaN-heterostructures or gate leakage, respectively, can be correlated with leakage current through dislocations between gate contact and 2DEG.^14,36 Hence, the path of leakage current seems to be clear in such a configuration. However, in our measurements, we did not measure the leakage current between AFM-tip and an Ohmic contact to the 2DEG, but to an Ohmic contact to the Si-substrate, i.e., to the backside of the sample. The observed leakage current is therefore influenced by both barrier and buffer leakage. An investigation of the leakage path between AFM-tip and Si-substrate requires clarifying the influence of interfaces, especially of conductive ones like the 2DEG. Addressing this, two tailored samples with similar buffer structure than the investigated sample above, but with a lower C-doping in the range of 10¹⁷ cm⁻³ were additionally investigated: one sample was grown up to the GaN-channel layer only, and the other one with the same growth conditions, but additionally with an AlGaN-barrier on top. Vertical leakage current mappings were performed by C-AFM on both samples. The sample with the AlGaN-barrier on top and therefore with a 2DEG right below the AlGaN/GaN-heterointerface shows clearly defined dislocation related leakage spots for an applied bias of −4 V on the Si-substrate relative to the grounded AFM-tip. In terms of the obtained leakage contrast, the result is fully consistent to that above and is depicted in Fig. 6(a). If there is no AlGaN-barrier grown on top, the C-AFM mapping exhibits no detectable vertical leakage current at all, which is shown in Fig. 6(b). This finding is independent of the C-doping within the buffer. Therefore, we assume that the locally different leakage current obtained by C-AFM on the sample with AlGaN-barrier is only due to locally different leakage between AFM-tip and 2DEG, which is apparently totally dominated by certain dislocation types. This means that a C-AFM measurement in a vertical setup is a suitable method to perform qualitative mappings of local differences in barrier or gate leakage without the necessity to process an Ohmic contact. This is only required for quantitative mappings. As observed, the measured absolute values of vertical leakage current depend strongly on the presence of a 2DEG and are a sum of barrier and buffer leakage. The 2DEG is a laterally highly conductive sheet of free electrons, which generally has the ability to promote a lateral spreading of vertical leakage current and a parallel circuiting of underlying electrically conductive defects, which results in a degradation of the insulating behavior of the buffer. In this sense, the observation that no leakage could be detected by C-AFM without the presence of a 2DEG is plausible since a single dislocation through the whole buffer may exhibit a too low conductivity to be sensed. In this manner, our observed absolute values of vertical leakage current at dislocations do not only depend on the specific properties of the dislocations within the barrier but also on the distribution and nature of dislocations within in the buffer in the surrounding. A schematic drawing of that understanding is shown in Fig. 7. That clusters or arrangements of dislocations may be more harmful than isolated ones was also discussed by Knetzger et al.²³ with regard to the dependence of vertical breakdown on the presence of deep carbon depletion zones in the vicinity of dislocation clusters in the C-doped GaN-buffer.

We have shown the feasibility of a strictly correlational methodology with significant statistics on a typical AlGaN/GaN-HEMT heterostructure on Si with a dislocation density in the range of 10⁹ cm⁻², which is applicable for GaN-on-Si with a lower dislocation density as well and allows to gain equivalent results. A successful way to get to directly proven relations between the structural and electronic properties of AlGaN/GaN-HEMT heterostructures was systematically developed. The critical role of dislocations with a screw component with regard to vertical leakage current could be directly shown, and the potential reasons for differences within the dislocation types were discussed. The obtained vertical leakage current contrast could be correlated with barrier or gate leakage contrast. Moreover, the density of different types of dislocations was determined on a large scale with high preciseness compared to conventional methods like TEM or rocking curve analysis, and the content of electrically critical dislocations directly above and below the device relevant 2DEG was analyzed. A potential model of dislocation related vertical leakage was developed with the help of two tailored samples and will be further investigated by a systematic study in a future work. The method of Electron Beam Induced Current (EBIC) may give additional insights into the electrical nature of dislocations, especially when performed in correlation with described methods. The results obtained in our work are important inputs for the development of device layer stacks, as it is possible to tune the ratio of certain dislocation types and their arrangement by growth conditions and layer sequence design to a certain extent, therefore to reduce the influence of dislocations on the electrical behavior of the material and so to optimize device reliability.

The research activities were carried out within the project “InRel-NPower,” which has received funding from the European Union’s Horizon 2020 research and innovation programme under Grant Agreement No. 720527 and partly within the project “ZuGaNG” (No. 16ES0084) funded by the German Federal Ministry for Education and Research (BMBF).