Spatial correlation of the E_C-0.57 eV trap state with edge dislocations in epitaxial n-type gallium nitride Free

GaN is a near-ideal material for use in high electron mobility transistors (HEMTs) designed for radio frequency (RF) and high power applications due to the inherently large carrier densities and high AlGaN/GaN two-dimensional electron gas (2DEG) mobilities. However, the reliability of these devices can be affected by populations of electrically active intrinsic and extrinsic crystal defects that can create trap states within the semiconductor bandgap. One particularly important trap state that has been associated with degraded performance and poor operational lifetimes in GaN HEMTs subjected to RF and DC stressing,^1–3 is an electron trap with an activation energy reported to be in the range of about 0.54–0.60 eV relative to the conduction band energy E_C.^1–18 We refer to this trap hereafter as the “E_C–0.57 eV trap,” but note that in some studies, it is labeled as the “E₂” trap. This trap is thought to be due to the same physical defect in all these studies because of its similar location in “Arrhenius space,” i.e., very similar emission time constants over a range of temperatures.¹⁷ The small variations in the reported activation energy of this trap are likely due to differences in equipment or measurement procedures and/or differences in the physical environment around the traps (e.g., doping, electric field, density of point defects or dislocations, etc.) due to variance in sample growth conditions. While the exact physical origin is not yet clear, it is now widely agreed that E_C–0.57 eV traps lie within the GaN layer, as opposed to the AlGaN layer, of these HEMTs.^{12,13,15–17} Previous studies had found similar trap energy and cross section on Schottky samples with no barrier layer (e.g., see Ref. 17), on HEMTs with either AlGaN or InAlN barriers,¹³ and on regions of an AlGaN HEMT where the AlGaN barrier was removed.^12,15

The E_C–0.57 eV trap was first identified in 1994 in unintentionally doped (background n-type) GaN grown by hydride vapor-phase epitaxy (HVPE).⁴ Since then, the E_C–0.57 eV trap has been observed in both intentionally doped and unintentionally doped n-GaN samples grown by different techniques, and its physical origin has been the source of much speculation (see, for instance, Refs. 1–18 and references therein). More recent work to explore how high-energy particle irradiation impacts the presence of trap states revealed the E_C–0.57 eV trap to be effectively insensitive to radiation damage,¹⁴ suggesting that it is likely not a simple intrinsic point defect, but is more likely associated with dislocations, dislocation-point-defect complexes, and/or extrinsic sources. Trap capture kinetics studies can be used to learn about the clustering behavior (or lack thereof) for specific traps.^19–21 Detailed capture kinetics studies in GaN layers grown by several methods^6,8,11,17 suggest that the E_C–0.57 eV trap has similar trapping kinetics as had been previously associated with electron-trapping by dislocation-point-defect complexes, defect clusters, and “Cottrell atmospheres” in other semiconductors.^20,22 In fact, the E_C–0.57 eV trap concentration was observed to increase with the threading dislocation density (TDD), further supporting the notion of a dislocation-related state.¹¹ One report suggested that the E_C–0.57 eV trap could be associated with edge dislocations, based on a comparison of average trap concentrations in different samples with average concentrations of different types of threading dislocations measured by transmission electron microscopy (TEM).⁹ In several studies, the E_C–0.57 eV trap concentration has also been found to be correlated with the concentration of Fe impurities,^10,12,17 which are often used to create semi-insulating buffer layers in GaN-based HEMTs. However, as discussed in several of the above references (see Refs. 6, 8, 11, and 14), this trap has also been observed in GaN samples grown by Molecular Beam Epitaxy (MBE) where there was no intentional Fe doping. Hence, the precise relation between Fe and the E_C–0.57 eV trap is not understood at this time.

Ultimately, determining the spatial location and distribution of this very detrimental trap within the HEMT structure will greatly aid in resolving the question of its specific source. Moreover, this kind of elucidation will have a significant impact on the general understanding of the complex HEMT degradation behavior associated with this trap, which in turn should enable an evidence-based approach toward its mitigation. In fact, the same can be said for defect levels within functional materials and devices, in general. To this end, the work presented here employs a combination of two complementary nm-resolution measurements applied to the same sample region of functional Ni/GaN Schottky contact structures. Scanning probe-deep level transient spectroscopy (SP-DLTS, or scanning-DLTS)^{12,15,23–26} is used to provide high-resolution spatial mapping of energy-resolved trap distributions, while electron channeling contrast imaging (ECCI)^27–31 is used to provide visualization and characterization of the underlying physical defect microstructure. This combination enables the direct correlation of the electrically active traps with specific types of physical defects within the functional device structures. E_C–0.57 eV traps were found to be highly localized (i.e., non-uniformly distributed) into compact “trap concentrations,” which were strongly correlated (co-located) with pure edge type dislocations in the GaN, but showed no appreciable correlation with screw or mixed dislocations.

The novel SP-DLTS technique was established in recent years and has been described in detail in prior publications.^{12,15,23–26} Key to this nm-scale technique is the ability to relate surface potential transients (SPTs), as measured via a scanning Kelvin probe system, to the kinetics of deep levels, similar to that done with capacitance transient-based conventional DLTS. Detailed descriptions of the equipment, the measurement techniques to record temperature-dependent surface potential transients (SPTs), and the data analysis procedures to convert SPTs into quantitative deep level defect spectroscopic parameters (trap energies, cross-sections, and concentrations) with high spatial resolution, can be found within the noted references. We note that the SP-DLTS technique used in Refs. 12, 15, and 23–26, and in the current work, is entirely different from the scanning transmission electron microscope (STEM)-based S-DLTS method of Petroff and Lang,³² who used a pulsed electron beam STEM to perform localized trap filling, with subsequent measurements of rapid macroscopic current transients when the beam was turned off.

ECCI is a non-destructive SEM-based technique that uses diffraction to form imaging contrast, allowing dislocations in crystalline materials to be observed and characterized in a manner similar to dark-field/two-beam transmission electron microscopy (TEM).^27–31 ECCI is an established technique for rapid characterization of threading dislocations in GaN,^33–37 and the present authors have demonstrated its application to an even wider range of heteroepitaxial materials and structures in recent years.^31,38–41

Si-doped n-type GaN layers were grown via NH₃-based MBE on GaN/sapphire templates (Lumilog), which contain a nominal threading dislocation density (TDD) of 5 × 10⁸ cm⁻². The epitaxial structure is comprised of a 300 nm-thick layer of Si-doped (N_D = 3 × 10¹⁸ cm⁻³) n⁺-GaN with a subsequent 500 nm-thick layer of Si-doped (N_D = 3 × 10¹⁶ cm⁻³) n-GaN. The epitaxial structure was processed into Schottky diodes using Cl-based reactive ion etching to provide isolated mesas of n-GaN surrounded by the exposed n⁺-GaN underlayer. An 8 nm-thick Ni film was deposited to cover most of the area of the mesas to form Schottky contacts, and a Ti/Al/Ni/Au stack was used to form Ohmic contacts to the exposed n+ layer. The Schottky contacts studied here were 290 μm × 290 μm in size, and had additional 75 μm × 75 μm Ni/Au pads deposited at the corners of the thin Ni film to make more robust contact pads to minimize probe damage. Figure 1(a) shows a schematic of the corner of one of these Schottky structures, and Fig. 1(b) shows a corresponding 50 μm × 50 μm plan-view atomic force microscopy (AFM) topographic image. Additional details of the device structure can be found in Ref. 14.

The experimental SP-DLTS setup used in this work is illustrated in Fig. 1(a). As in conventional DLTS,⁴² a voltage pulse sequence [top curve in Fig. 1(a)] is used to successively fill and empty traps at a particular range of depths called the modulation region [see Fig. 1(c)]. For an n-type Schottky contact, the bias is low during the fill pulse condition allowing traps in the modulation region to capture electrons from the substrate. During the empty pulse condition, the Schottky contact is reverse-biased and these same traps emit electrons because their energy is raised above the electron quasi-Fermi level. As shown in Fig. 1(c), beyond the edge of the metal, we approximate the modulation region to have a quarter-annulus cross-sectional shape with an inner radius of r_in and an outer radius of r_out, which are defined as the transition-depth (i.e., where the trap level crosses the electron quasi-Fermi level) during the fill and empty pulses, respectively. For a particular trap energy, the transition-depth at a given bias can be determined from knowledge of the voltage-dependent depletion depth and doping concentration,⁴³ both of which can be obtained from a capacitance-voltage curve. For the temperature-dependent measurements, the biases applied to the Ni contact were 0 V and −4 V for the fill- and empty-pulse conditions, respectively, and the modulation region extended from r_in ≈ 50 nm to r_out ≈ 220 nm below and on the side of the Ni film. In conventional DLTS,⁴² the total trapped charge under a Schottky contact or a metal-insulator-semiconductor structure is monitored by measuring the sample capacitance, since trapped charge alters the depletion region (and hence the capacitance) of a Schottky contact. Conventional DLTS can monitor the average trap concentration as a function of depth below the contact by varying the biases used for the fill and empty pulses (and hence r_in and r_out) but cannot measure lateral trap distributions.

In SP-DLTS, the scanning probe microscope (SPM) is operated in the Kelvin probe mode,^{12,15,23–26} and the local surface potential below the tip apex can be measured as the tip bias required to minimize the electrostatic force on a conducting probe tip. Trapped charge in the vicinity under the probe tip will modify the local surface potential, and hence trapped charge transients can be monitored by recording and analyzing local SPTs. As illustrated in Fig. 1(a), the measured SPTs are induced by and synchronized with bias pulses applied to the metal contact. Once time constants (τ) of SPTs measured at several temperatures (T) have been determined, the activation energy and capture cross-section of traps on the local scale can be extracted from an Arrhenius plot of ln(τT²) vs 1/k_BT.⁴² In this study, the Kelvin probe feedback was operated with sufficient bandwidth to measure transients with time constants greater than a few ms. Since the Ni Schottky contact will screen traps located well beneath the Ni film from the SPM tip [e.g., traps near TD4 in Fig. 1(c)], only modulated traps located on the side of the metal [e.g., traps near TD2 and TD3 in Fig. 1(c)] can be directly probed by SP-DLTS. Traps located beyond the modulation region [further than r_out ≈ 220 nm from the edge of the Ni film, such as those near TD1 in Fig. 1(c)] are not detectable because their occupancy is not modulated.

Spatial variations in local trap density can be observed from an SP-DLTS image, or “map,” by plotting the change in the surface potential signal, ΔSPT, during a specific time window as a function of the tip position. This time window is referred to as a “rate window” and it makes a map sensitive to trap emission close to a particular time constant.²⁴ If the trap population exhibits strong spatial localization, then the SP-DLTS map will show well-defined localized peaks. From here on, we use the term “trap concentrations” to refer to these regions with higher local trap concentrations. The areal density of trap concentrations, ρ_T, can be estimated from ρ_T ≈ λ_T/r_out, where λ_T is the measured linear density of trap concentrations in the direction parallel to the metal edge.

Figure 2(a) shows an atomic force microscope (AFM) topographic image measured near a corner of a device mesa, with the bare GaN on the upper part and the ∼8 nm thick Ni film on the bottom part. This Schottky contact was located roughly 3.5 mm from the wafer edge. Figure 2(b) shows an SP-DLTS map measured close to room temperature for a 14 μm × 1.2 μm region along the edge of the Ni film using an inclusive rate window of [6 ms, 125 ms]. This rate window gives peak sensitivity to transients with a time constant of ∼40 ms, which provides a high sensitivity to the E_C–0.57 eV traps at room temperature. The bottom panel shows the SP-DLTS map data for ΔSPT ≥ 9 mV superimposed on the topographic image. Transients due to traps near the edge of the Ni film were observed and found to exist in a highly non-uniform spatial distribution, with multiple relatively isolated spot-concentrations, (i.e., trap concentrations). If the traps were homogeneously distributed in the GaN film, the SP-DLTS map would appear as a uniform band along the edge of the Ni film with a width related to that of the modulation region. The trap concentrations in this SP-DLTS map had peak ΔSPT values of up to ∼44 mV, with an average value of ∼30 mV. We note that a simple electrostatic calculation indicates that room-temperature emission of 13 trapped electrons distributed in a narrow cylinder in GaN extending from just beneath the surface to a depth of (r_out – r_in) ≈ 170 nm would produce a ΔSPT of ∼30 mV at a location ∼40 nm directly above the cylinder, which is an estimate of the tip height. However, we also note that this simple calculation neglects effects related to the tip geometry and partial screening from the Ni film, and hence 13 is an underestimate of the number of trapped electrons needed to produce a measured ΔSPT of ∼30 mV.

From SP-DLTS maps measured along 36 μm of the Ni/GaN boundary, the linear density of trap concentrations was estimated to be λ_T ≈ 4000 cm⁻¹. This yields an estimated areal density of trap concentrations of ρ_T ≈ λ_T/r_out ≈ 2 × 10⁸ cm⁻², which is comparable to the ∼5 × 10⁸ cm⁻² total TDD in these GaN films. We note that the dislocation population numbers for GaN grown on sapphire typically place pure edge dislocations at about 50% and pure screw dislocations anywhere from 1% to 10%, with mixed dislocations making up the balance.^44,45 Hence, the areal density of trap concentrations measured here is actually very close to the expected density of either pure edge or mixed dislocations. This will be discussed further below.

To determine the activation energies and cross-sections for specific trap concentrations, the tip was positioned over a trap concentration and multiple transients were measured and averaged for different temperatures between 26 °C and 51 °C. For example, Fig. 3(a) shows average transients and 3-parameter single-exponential fits at various temperatures when the tip was positioned at the location indicated by the magenta arrows in Fig. 2(c). About 3000 transients were averaged at each temperature at this location. We see that the time constant of this trap concentration decreased systematically with increasing temperature, as expected for thermally-activated trap-emission. Figure 3(b) shows an Arrhenius plot of (τT²), illustrating the very good Arrhenius-space linearity for this particular trap concentration, with an extracted activation energy of 0.56 eV and a capture cross-section of 9 × 10⁻¹⁶ cm². These values are close to those for the well-known E_C–0.57 eV trap in GaN as measured by conventional DLTS.¹⁷ Temperature-dependent SPTs were also measured at the locations of the seven other trap concentrations shown in Fig. 2, and at another location at the edge of a different contact located about 1.4 mm away from the contact shown in Fig. 2. Arrhenius plots measured at these nine separate locations each showed very good linearity with squared Pearson product moment correlation coefficients of R² > 0.99. Figure 3(c) shows the measured SP-DLTS data from all nine locations superimposed on macroscopic conventional DLTS data¹⁴ that was measured on a similar device. Extremely close proximity in “Arrhenius space” is seen for all the SP-DLTS data with respect to the conventional DLTS data¹⁴ in Fig. 3(c) for a trap with an activation energy of ∼0.6 eV, which was identified as the E_C–0.57 eV trap due to its location in Arrhenius space compared to other studies.^{1–13,15–18} This close proximity in Arrhenius space with prior observations indicates that all of the trap concentrations measured here consist of E_C–0.57 eV traps. The best-fit lines for the SP-DLTS data measured at the 9 locations included here yield an average value for the trap energy of E_C–0.53 eV, with a standard deviation of 0.05 eV.

To determine whether the trap concentrations are in fact located near, and thus potentially associated with, threading dislocations, the sample was characterized using electron channeling contrast imaging (ECCI). As previously noted, ECCI is a non-destructive SEM-based technique that uses diffraction to form imaging contrast, which allows for the visualization and geometric characterization of dislocations—in this case, surface-penetrating threading dislocations—in crystalline materials. The resulting images are similar to plan-view dark-field/two-beam TEM micrographs.^27–31 Because of the non-destructive nature and accessible SEM magnifications, it was possible to easily identify and image the exact same area used in the SP-DLTS measurements, allowing for direct comparison between the two techniques.

Briefly, ECCI is performed by establishing a scanning geometry in which the in-going electron wavefront is at a target diffraction (channeling) condition. Any displacement field, or perturbation within the diffracting planes, such as extended crystal defects (like dislocations), serve as scattering elements. Depending on the nature of the defect or feature, the scattered electrons can either be backscattered and detected with a backscattered electron detector (BSED) or forescattered (i.e., channeled into the sample) and lost. The net resulting signal provides imaging contrast (intensity variation versus the background) related to the specific geometry of the displacement field.

To obtain an ECCI image, the sample is first oriented with respect to the beam-sample tilt angle to establish a fixed channeling condition at a specific Bragg angle; low-index crystal planes are typically chosen to obtain a strong contrast. A diffraction condition (“$g→$ vector”) can be selected through use of a low-magnification (large area) BSED image wherein an electron channeling pattern (ECP) is visible.⁴⁶ To establish a diffraction condition, a Kikuchi line is centered onto the optic axis by tilting and rotating the sample. At this point, increasing the imaging magnification effectively locks in the diffraction condition enabling the observation of dislocations in the form of local contrast variations. Further details regarding ECCI measurements similar to those discussed herein can be found in prior reports.^31,38–41

An experimental ECP, along with a wide angular range simulated pattern,³⁰ are shown in Fig. 4. The six-fold symmetry of the GaN wurtzite crystal structure can clearly be observed in the simulated ECP. We note here that the large-area Ni film contacts, which cover the majority of the processed sample, were sufficiently thin (∼8 nm) as to be semi-transparent to the SEM electron beam, making it possible to obtain the low-magnification ECP images needed for crystallographic orientation of the processed sample. The Au coated regions, seen in Fig. 4(a) as the bright white lines and geometric shapes, are too thick to be electron-transparent here, and result in large backscatter signals (aided in part due to the enhanced Z-contrast).

Figure 5 shows high magnification ECCI images of the same region as the Schottky contact from Fig. 2, with the bare n-GaN on the upper part of the images, and the thin Ni film on the bottom. These images show sharp contrast features that are in agreement with prior reports of surface-penetrating threading dislocations in GaN as observed via ECCI.^33–37 The average areal density of these features is approximately 4 × 10⁸ cm⁻², consistent with the substrate template manufacturer (Lumilog) specified TDD value of 5 × 10⁸ cm⁻². The dislocation features in Fig. 5 can be generally classified based upon their appearance: (1) large, strong contrast dipoles (dark on one side, bright on the other) or (2) small, faint dipoles or monopoles (either bright or dark compared to the background).

In addition to these clearly observable threading dislocations within the exposed n-GaN, we note that high-magnification ECCI images also revealed dislocations within the etched n⁺-GaN regions (not shown) and underneath the 8 nm-thick Ni contact layers; several of these latter features are visible in Fig. 5. However, the contrast of the dislocations under the Ni is strongly reduced compared to those in the uncoated areas due to additional scattering and attenuation by the Ni, making detailed characterization difficult. Nonetheless, this unexpected result, in addition to the ECP visibility in the same areas, indicates that ECCI is likely of use within a wider variety of different types of processed device structures than has been previously demonstrated, even when they are mostly covered with a thin metal overlayer.

Threading dislocations in (0001)-oriented GaN are typically of three different geometries or types (i.e., Burgers vectors, $b→$⁠)⁴⁷: pure edge, $b→=13112¯0$⁠; pure screw, $b→=0001$⁠; or mixed, $b→=13112¯3$⁠. As previously noted, relative dislocation population numbers for GaN grown on sapphire indicate a pure edge fraction of about 50% and pure screw anywhere from 1% to 10%, with the remainder possessing mixed character.^44,45 Based upon the ECCI contrast criteria established in the literature,^35–37 multiple diffraction conditions, as shown in Figs. 5(a)–5(c), were used to characterize and catalog the geometries of the various dislocations. Screw and mixed dislocations tend to show a strong dipole (bright-dark) contrast due to the large degree of induced surface relaxation. Additionally, upon rotation of the $g→$ (diffraction) vector, the directions of the screw and mixed dislocation contrast dipoles also rotate. Edge dislocations, on the other hand, exhibit much weaker contrast. Further, upon rotation of the $g→$ vector, edge dislocation appearance either remains unchanged or the contrast dipole is reversed. The resultant identifications are given in Fig. 5(d).

Finally, a direct comparison of the SP-DLTS and ECCI data over the same sample region is presented in Fig. 5(e). The metal edge was used as the alignment reference for the overlay. The key finding here is that each trap concentration has at least one edge dislocation within its bounds, and all edge dislocations within the modulation range (11 out of 11) are found to lie within high SP-DLTS signal regions (i.e., ΔSPT ≥ 9 mV for an E_C–0.57 eV trap concentration). As discussed in the supplementary material, the probability that all 11 pure edge dislocations would happen to appear by chance within the high-amplitude contours is less than 10⁻⁴, which we consider to be strong evidence that there is a strong spatial correlation between edge dislocations and trap concentrations. Furthermore, there are several mixed and/or screw type dislocations within the modulation region that lie outside any trap concentration. Where mixed/screw dislocations do coincide with some trap concentration, an edge dislocation can be found within the immediate vicinity. Therefore, the clear conclusion is that the SP-DLTS detected E_C–0.57 eV trap concentrations within the Si-doped NH₃-MBE GaN are directly associated with only pure edge dislocations, while the combined set of mixed and screw type dislocations do not show a significant spatial correlation to this trap.

The exact reason behind the strong correlation between the E_C–0.57 eV trap and edge dislocations is not currently known. Previous capture kinetics studies have suggested that the E_C–0.57 eV trap may result from the clustering of point defects at dislocations rather than being related to the native dislocation core itself.^6,8,11 The dilational strain fields surrounding edge dislocations are known to effectively getter impurities (depending upon their relative size),⁴⁸ producing a “Cottrell atmosphere” of impurities around the dislocations.⁴⁹ If these impurities or impurity clusters contribute to electronic defect states, then one would indeed expect to find higher trap concentrations around edge dislocations. However, based on this simple picture, both the pure edge (⁠$b→=a→$⁠) and mixed (⁠$b→=a→+c→$⁠) dislocations, which possess identical edge components, would be expected to exhibit equally high local trap concentrations; the total elastic fields of mixed dislocations are merely the superposition of the fields associated with the individual components. Nevertheless, the core reconstructions of the pure edge, mixed, and screw dislocations are all significantly different, as determined by recent high-resolution STEM studies.^50,51 Because we observe that E_C–0.57 eV traps are only correlated with pure edge dislocations, it stands to reason that more complicated effects are at play. A likely explanation is that the interaction of some impurity (or impurities) with the edge dislocation core structure produces a unique local chemistry or bond configuration that induces this particular electronic state within the bandgap. The distinctly different core structures possessed by the mixed/screw dislocations must then produce sufficiently different local chemistries such that this particular trap state is not created. Determination of the exact nature of the interactions leading to these trap states will require further detailed investigation involving high-resolution electron microscopy and/or accurate first-principles modeling, both of which are beyond the scope of this present work.

In conclusion, E_C–0.57 eV traps were found to exhibit strong spatial localization in Si-doped NH₃-MBE GaN, and are spatially correlated with pure edge threading dislocations, but not with pure screw or mixed threading dislocations. It was also demonstrated that ECCI can be implemented on functional Schottky contact devices even though most of the surface area was covered by a thin metal film, indicating that ECCI can be used to study dislocations and other defects in a much broader range of processed device structures than previously thought possible, without significant sample preparation.

See supplementary material for additional details regarding the experimental techniques and the probability calculations used in this work.

K.G. was supported by The Ohio State University Graduate School and an external fellowship. This work was supported in part by The Ohio State University Department of Physics, the Office of Naval Research under Grants Nos. N00014-16-1-2641, N00014-15-1-2077, and N00014-16-1-2932, and the Air Force Office of Scientific Research under Grant No. FA9550-17RXCOR434.

The authors are grateful to D. W. Cardwell (OSU) for help with the SP-DLTS experimental setup and for the valuable discussions.