1. Introduction

III-nitrides have gained huge importance for electronic and optoelectronic devices over the past decades. As a prominent example, GaN in combination with ternary InGaN quantum wells builds the basis for blue and white light emitting diodes (LEDs) [1,2]. In addition to conventional chip processing, including photolithography and different types of dry and wet-chemical etching, laser-based processes have developed into a substantial alternative, e.g., by laser direct-writing of mesa structures or laser lift-off [3–7]. This trend is closely linked to an ongoing development in laser technology. In particular, the availability of high power lasers with ultrashort pulses enables structural modification and ablation of wide band gap semiconductors and even insulators with high precision [8,9].

Absorption mechanisms in the ultra-short pulse regime differ substantially from those in the nanosecond regime. While in the latter case semiconductors and insulators with a band gap larger than the photon energy are practically transparent, non-linear absorption processes become relevant for intensities in the TW/cm² range [10]. Such values are commonly reached by pulsed lasers in the picosecond and sub-picosecond regime. As a consequence, even insulators can be brought into a highly excited state by ultrashort laser pulses. In that context, different regimes must be considered, which are governed by tunneling ionization (field-driven) or by multi-photon ionization (photon-driven) [11]. In case of wavelengths from the UV to the near IR, as they are utilized in this study, most light-matter interactions up to the ablation threshold can be regarded as photon-driven. Here, a significant free carrier population is created by interband absorption, which in case of sub-band gap radiation mainly relies on multi-photon excitation [8,9]. Once this seed population is present, intraband absorption and impact ionization further accelerate the excitation process during the later part of a laser pulse [12]. When a critical free carrier density is exceeded, subsequent energy transfer to the lattice results in irreversible damage to the crystal. With ultra-short laser pulses, optical excitation is much faster than typical timescales of electron-phonon interaction, so that thermal diffusion and the corresponding heat affected zone in laser-based material processing are minimized [13].

At high enough fluences, excitation with intense optical pulses results in laser damage and ablation. The material-specific ablation threshold, given in J/cm², is a function of various laser parameters, e.g., wavelength and pulse width, since they influence relevant processes like absorption behavior and the extent of thermal diffusion [10]. In this study, the ablation threshold of unintentionally doped GaN films is determined for different wavelengths from the IR to the UV range, pulse widths from 0.34 to 10 ps and pulse numbers between 1 and 100, expanding the range of values for the wide band gap semiconductor GaN (${E_\textrm{g}}$ = 3.4 eV) given in previous studies [14–18]. Rather thick GaN films of more than 10 µm, grown on c-plane sapphire wafers, were employed in order to minimize impacts of the underlying substrate onto the ablation threshold of the thin film, as it was observed for different material systems [19,20]. Laser ablation occurs in the form of craters centered around the position of the laser beam, as analyzed in a scanning electron microscope (SEM). In addition to that, we show that laser damage manifests itself in the form of µm-sized pits when the ablation threshold of defect-free material is not yet exceeded. Such pits have been reported in a similar form for the nanosecond regime [21]. This form of laser damage has been described for other materials, e.g., silica, and is attributed to a local reduction of the ablation threshold at so called damage precursors, e.g., crystal defects or particles on the surface [22,23]. In case of GaN, we demonstrate that threading dislocations, i.e., intrinsic one-dimensional crystal defects, act as local excitation pathways. This is investigated by cathodoluminescence (CL), wet-etching studies with H₃PO₄, and transmission electron microscopy (TEM). The perturbing activity of threading dislocations is widely known in the reverse direction: Threading dislocations acting as recombination centers for charge carriers, either by direct facilitation of non-radiative recombination [24] or by accumulation of point defects. This also causes yellow luminescence in GaN with photon energies far below the band gap [25]. The average lateral distance of those damage precursors is in the range of a few µm, which is small compared to the utilized laser spot diameters of 10 to 40 µm, so that the determination of the ablation threshold should not considerably be affected by fluctuations in threading dislocation density [26] (see also Supplement 1, S1).

2. Experimental setup

For the experiments presented in this study, GaN films of approximately 11 µm thickness were employed. They were grown on 650 µm thick 4 inch sapphire substrates by metal-organic vapor phase epitaxy (MOVPE) within an Aixtron AIX2600HT G3 planetary reactor. Standard precursors, namely, trimethylgallium (TMGa), and ammonia (NH3) were utilized. The sapphire wafers were thermally cleaned within the MOVPE reactor in hydrogen at up to 1100 °C prior to a nitridation step and the deposition of a low-temperature GaN nucleation layer. Following a recrystallization and coalescence step, the deposition of an 11 µm thick GaN layer occurs at a temperature of 1050 °C, reactor pressures between 125 and 290 mbar and a V/III (NH3/TMGa) ratio of approximately 1000 using hydrogen as the carrier gas. The GaN layers feature low n-type doping with a concentration of about 1 × 10¹⁷cm^-3. The layer thickness was constantly monitored during growth via reflectometry by a Laytec EpiCurve TT in-situ measurement system.

Two different wafers were employed for the experiments. Sample A was mainly used for the determination of the ablation threshold presented in section 3.1, while laser scanning, wet etching and TEM investigations (section 3.2) were primarily performed on sample B. The samples were produced in different growth runs and slightly vary in their threading dislocation density (sample A: 3 × 10⁸cm^-2, sample B: 5.5 × 10⁸cm^-2, deduced from CL dark spot densities).

The GaN films were exposed to laser pulses in a customized laser machining setup (IntelliFab, Spectra Physics), which is installed in a temperature-stabilized laboratory. Prior to each laser treatment, the samples were cleaned in an ultrasonic bath in first acetone and then isopropyl for several minutes, followed by a DI water rinse. A clean surface condition was confirmed with an optical microscope.

The laser machining setup is operated with an ultrashort pulse laser (Spirit, Spectra Physics). The regenerative amplifier contains an Yb-doped crystal and emits at a fundamental wavelength of 1040 nm. It is equipped with a built-in second harmonic generation (SHG) module. The average output power at 1040 nm can be varied up to 16 W and the repetition rate can be set to 100 or 200 kHz, respectively. The pulse width is tunable between 0.35 ps and 10 ps. Using a flip-mirror, the laser beam can be directed to two different machining stages, as illustrated in the schematic in Fig. 1. At stage 1, a galvanometer scanner in combination with a telecentric objective is used. This setup enables fast scanning of the beam across the sample, with a scribing velocity of up to 2 m/s. The galvanometer scanner and the objective are mounted on a z positioner for focus adjustment. The beam path is designed for the fundamental and the second harmonic wavelength of the laser system. In the focus position, the beam diameter is approximately 25 µm wide for 520 nm and around 40 µm for the wavelength of 1040 nm, respectively. The beam diameter was determined experimentally together with the ablation threshold (see also Supplement 1, S4).

 

Fig. 1. Schematic of the experimental setup. Laser pulses of variable energy are generated by the regenerative amplifier at either 1040 or 520 nm wavelength with a repetition rate of 100 or 200 kHz, respectively. The number of pulses per time can be controlled by an integrated pulse picker. By changing the position of flip mirrors, the beam is directed to either stage 1 or 2, where the influence of different wavelengths on the ablation behavior can be tested. At stage 1, a galvanometer scanner is used for beam steering, while at stage 2 the sample is moved.

Download Full Size | PDF

Experiments with the UV wavelengths were conducted at a second, customized machining stage. For the generation of the third harmonic (347 nm), laser pulses at 1040 nm are partly frequency-converted to the second harmonic by a lithium triborate (LBO) crystal. The direction of polarization and the temporal delay of the second harmonic are adapted by a waveplate and a calcite plate, respectively. Following behind, the third harmonic is generated by sum frequency generation in a beta barium borate (BBO) crystal. Using customized mirrors and high-contrast spectral filters, the higher wavelength components are removed from the beam in the following path (see also Supplement 1, S2). A vertically mounted standard 5x NUV objective with effective focal length (EFL) of 39.9 mm was utilized to focus the beam onto the sample, which results in a beam diameter of 13 µm in focus. Additional measurements of the ablation threshold at 347 nm were made with an UV-coated plano-convex lens with EFL 100 mm (i.e., 25 µm beam diameter in focus) in order to investigate the impact of a varying spot size. The sample is mounted on a pair of high-precision linear stages, which can be moved with 50 mm/s with respect to the static laser focus. The same stage is employed for the fourth harmonic, using a slightly different beam path. In this case, the internal SHG module of the laser source is utilized. The beam is directed to a BBO crystal, which converts 520 to 260 nm. As in the other beam path, the remaining fraction of the second harmonic is consecutively filtered out. An UV-coated plano-convex lens with an EFL of 50 mm was employed for the 260 nm line, resulting in a beam diameter in focus of approximately 10 µm.

For all utilized wavelengths, the pulse widths were determined with a commercial autocorrelator (PulseCheck, APE), as shown in Supplement 1, S2. Because the external higher harmonic generation setup is designed for pulses around 0.35 ps, both conversion efficiency as well as effective pulse width decrease for longer pulses. Hence, no values of the ablation threshold for pulse widths > 3 ps were obtained for the UV wavelengths.

After laser treatment, the corresponding sample sites were analyzed in a field-emission SEM (Mira3, Tescan). The system is equipped with a customized cathodoluminescence (CL) detection setup, which was employed to measure dark spot densities of GaN samples. Cross-sectional TEM lamellas were prepared in a focused ion beam SEM (FIB-SEM, Helios 5ux, Thermo Fisher), equipped with a detector for scanning transmission electron microscope (STEM) images. Weak-beam dark field (WBDF) analysis of dislocations was conducted in an image-corrected TEM (Titan 80–300, FEI) operated at 300 kV and equipped with a CCD camera (UltraScan 1000XP, Gatan). To meet WBDF conditions for a Bragg vector g, the specimen was tilted to two-beam conditions first. Secondly, the electron beam was tilted by -g and a 10 µm objective aperture was placed to the center of the optical axis.

3. Results and discussion

3.1 Determination of the ablation threshold

The typical morphology of the GaN surface after treatment with high energetic laser pulses is shown in Fig. 2. The pulses impinge onto the surface that features a homogeneous ablation threshold at the scale of the beam diameter. Wherever the threshold is locally exceeded, ablation occurs. Hence, the lateral fluence distribution results in craters with a mostly sharp contour line. The elliptical shape is caused by an ellipticity of the beam itself. In Fig. 2(a) to (c), craters after irradiation with 0.34 ps wide pulses at 260 nm are shown, with an increasing number of pulses from 1 to 100. Upon a single pulse, different regimes appear within the crater. The outer ring exhibits a smooth surface with metallic gallium droplets, which are formed by GaN decomposition [14]. In the center, which is hit by a higher fluence, the morphology appears rougher. With an increasing number of impinging pulses, the lateral crater size stays nearly constant, while more and more material is ablated in the center (see depth profiles in Supplement 1, S3). After 10 pulses, laser-induced periodic surface structures (LIPSS) appear particularly in the outer region of the crater. The spatial periodicity of those LIPSS of 225 nm [see inset in Fig. 2(b)] is slightly below the laser wavelength of 260 nm, as commonly reported in literature [27,28]. Even after 100 pulses, the crater features a sharp contour. The darker contrast in the center of the crater corresponds to the sapphire substrate, which starts being exposed after a higher number of repetitions. In case of the 3.3 ps wide IR pulses depicted in Fig. 2(d) to (f), the crater is confined by a sharp contour after one pulse, but laser damage in the form of additional µm-sized pits is visible in the surrounding area. The pits are formed at damage precursors, as will be discussed in section 3.2 of this paper. After 10 pulses, the crater size noticeably increases: a cohesive ablation site including most of the initially µm-sized pits is formed. Both morphology and contour line appear rough, which implies an increased significance of accumulation effects during the ablation process.

 

Fig. 2. SEM images of laser craters for different parameters. Upper row: 260 nm, 0.34 ps pulse width, pulse energy 0.7 µJ (corresponds to 2.3 J/cm² peak fluence). (a) 1, (b) 10, and (c) 100 pulses. Lower row: 1040 nm, 3.3 ps pulse width, 24 µJ pulse energy (corresponds to 3.8 J/cm² peak fluence). (d) 1, (e) 10, and (f) 100 pulses

Download Full Size | PDF

To analyze the ablation threshold of GaN thin films, the diameter-regression technique was employed [29], which has been described in detail in previous studies [6,15] (see also Supplement 1, S4). Employing this method, the ablation threshold for a specific laser parameter combination can be measured, when a series of pulses with increasing energy is directed to different spots on the surface. In our experiments, the number of pulses per site was controlled by setting the pulse picker to a value of 10,000 or 20,000, corresponding to an effective rate of 10 pulses per second. The laser was then activated for 100 ms to 10 s in order to direct 1 to 100 pulses to the targeted sample site (S-on-1 experiments). We based the determination of the ablation threshold on the diameter of the cohesive crater in the beam center. Laser damage in the surrounding, as for instance visible in Fig. 2(d), was not considered here. This way, we tried to distinguish the actual ablation threshold from the onset of laser damage at lower fluences, caused by locally enhanced absorption at precursors. However, it must be noted that precursors assist in the lateral extension of the central ablation crater, contributing to the observed decrease of the ablation threshold at higher pulse numbers.

Determined ablation thresholds are shown in Fig. 3, all obtained with sample A in order to ensure comparability. The ablation threshold for the GaN layer varies between 0.05 and 3 J/cm² in the analyzed parameter range, depending on the wavelength, pulse width and numbers of pulses per site. In Fig. 3(a), the dependency on the number of pulses is plotted, categorized by the wavelength. As a general trend, one can deduce a decrease of the ablation threshold with the number of pulses per site, which is known as accumulation effect [30]. The impact of wavelength and pulse width on this behavior will be discussed in more detail below.

 

Fig. 3. (a) Ablation threshold of GaN in dependency of the pulse number for different wavelengths and pulse widths. The dotted lines correspond to the empirical model in Eq. (3) (see Supplement 1, S5 for parameter values). (b) Ablation threshold of GaN in dependency of the laser wavelength for 1, 10 and 100 pulses and a pulse width between 0.34 and 0.40 ps. The scaling relation in Eq. (1) based on the skin layer approach is represented by the black line (see text for further explanation). (c) One-pulse ablation threshold as a function of the pulse width for all tested wavelengths. The dotted lines correspond to the rate equation model [Eq. (2)]. Also, different values from literature were added for comparison. (d) Ratio ${F_{\textrm{th}}}({100} )/{F_{\textrm{th}}}(1 )$ in dependency of the pulse width for all tested wavelengths.

Download Full Size | PDF

In Fig. 3(b), the dependency of the ablation threshold on the wavelength is shown, considering the shortest pulse widths of 0.34 to 0.4 ps. For single laser pulses, the datapoints exhibit a steady increase of the threshold fluence, which rises from 0.15 to 0.83 J/cm² between 260 and 1040 nm. A similar behavior is observed for low pulse numbers, while the dependency on the wavelength is less pronounced at 100 pulses. The steady increase of the one-pulse ablation threshold with the wavelength from the UV to the IR is a common phenomenon for most materials [10,14,31]. Gamaly et al. argue that, for very short pulses, the ablation threshold is mainly determined by the penetration depth of the optical skin layer [10]. They reason that the material, independent of the band gap, is brought into a highly excited state already at the beginning of the pulse, and then intraband absorption and impact ionization further dominate the interaction. Hence, the impact of the initial absorption behavior can be neglected. This holds for fluences around the ablation threshold and above. Based on this consideration, they deduce a linear scaling relation for the threshold fluence:

${J_\textrm{i}} \approx $ 7.5 eV is the ionization potential, ${\varepsilon _b} \approx $ 4.4 eV is the binding energy per atom, and ${n_\textrm{a}}$ = 8.9 × 10²² cm^-3 is the density of atoms in case of GaN [32]. Our data roughly match the course of the scaling relation, as indicated in the graph in Fig. 3(b).

Of course, Eq. (1) must be seen as a rough estimate and particular holds in the limit of very short pulses, which may not be reached at 0.4 ps. As demonstrated by several other authors, there is in fact an impact of the initial absorption behavior onto the ablation threshold [31,33,34]. In case of semiconductors, initial absorption is mostly determined by the wavelength and the band gap ${E_\textrm{g}}$. A simple rate equation can be used to calculate carrier generation in the conduction band. In this approach, the surface region of the material is considered, where laser intensity I and hence the carrier concentration n are highest:

$\alpha $ denotes the linear absorption coefficient, ${\beta _m}$ the coefficient of multi-photon absorption of order m, $\sigma $ the avalanche coefficient, and R losses due to Fresnel reflection. Usually, a critical carrier density ${n_{\textrm{crit}}}$ of approximately 10²¹cm^-3 is assumed to be necessary to induce laser damage and ablation [33]. In contrast to the scaling relation in Eq. (1), the rate equation in Eq. (2) does not only account for the impact of the wavelength, but also for the pulse width: The absorption coefficients are dependent on the photon energy, while the pulse width influences the intensity and thus the multi-photon absorption term. In Fig. 3(c), the experimentally determined one-pulse ablation threshold is plotted as a function of the pulse width for each of the four wavelengths. In case of 1040, 520 and 347 nm, the threshold fluence steadily increases with the pulse width, following roughly a ${\tau ^{0.3}}$ dependency. This is described similarly for the damage threshold of different oxide thin films by Mero et al. and was reasoned by an interplay of multi-photon absorption and impact ionization [33]. The applied rate equation model supports this conclusion, as depicted in Fig. 3(c) (see Supplement 1, S5 for exact parameters). It should be remarked that $\alpha $ is as high as 1 × 10⁷m^-1 at 347 nm, since one-photon interband absorption is feasible at this wavelength [35]. This would correspond to an ablation threshold as low as 0.01 J/cm² according to Eq. (2), even without any additional non-linear absorption channel. However, it must be considered that the optical band gap increases with the carrier concentration due to band filling effects: the efficiency of interband absorption for photons at 347 nm decreases significantly for carrier concentrations above 10¹⁹cm^-3 [36] (see also Supplement 1, S5). Hence, we conclude that higher-order absorption processes are still involved in the carrier generation. Consequently, a similar pulse width dependency as in case of 1040 and 520 nm can be observed.

At a wavelength of 260 nm, the situation is different. Here, $\alpha $ remains high up to carrier concentrations close to the critical density, so that one-photon absorption is the dominant process throughout the whole pulse duration. According to the model in Eq. (2), this results in a constant ablation threshold independent of the pulse width. In fact, we even observe a slight decrease between 0.34 and 2.6 ps. The reason for this behavior may be relatively long thermalization times for hot carriers in GaN, which are injected high into the conduction band at 260 nm excitation, resulting in phase space filling. Thermalization processes as slow as 1 ps in highly excited GaN thin films were reported in literature, which was measured by pump-probe spectroscopy [37,38].

To summarize, the simple rate equation model based on Eq. (2) is in accordance with the observed one-pulse ablation threshold dependencies on wavelength and pulse width in the range of 0.3 to several ps. For longer pulses above 3 ps, the threshold fluences for 1040 and 520 nm tend to rise more rapidly than ${\tau ^{0.3}}$. This could be reasoned by the onset of thermal diffusion, which is often considered to cause an increase of the threshold fluence with ${\tau ^{0.5}}$ in the picosecond and nanosecond range [10,39]. On the other hand, it is known that the ablation threshold of GaN stays well below 1 J/cm² for UV pulses with a high one-photon absorption coefficient, even to pulse widths in the nanosecond regime [5]. Hence, thermal diffusion should not be too relevant at pulse widths below 100 ps.

After the analysis of the one-pulse ablation threshold of GaN, the accumulation behavior for an increasing number of laser pulses can be studied by the data presented in Fig. 3(a) and (d). A commonly used empirical model is suited to describe the dependency of the ablation threshold on the pulse number N, which is characterized by a limit $F_{\textrm{th}}^\infty $ and the incubation parameter k [40,41]:

Fit curves according to this model were added in Fig. 3(a). The level of $F_{\textrm{th}}^\infty $ should nearly be reached at 100 pulses, since only a slight decrease of the ablation threshold between 50 and 100 pulses can be observed in most cases [40]. Comparing different wavelengths and pulse widths, it can be noted that the value of $F_{\textrm{th}}^\infty $ is similar for 1040, 520 and 347 nm (0.1–0.2 J/cm²), independent of the huge variations of the one-pulse ablation threshold (see also Supplement 1, S5). In case of 11 ps wide pulses at 1040 nm, this means a reduction of the ablation threshold by a factor of 20 when increasing the number of pulses from 1 to 100, as also depicted in the graph in Fig. 3(d). Such a drastical reduction of the ablation threshold is often linked to damage precursors. As will be discussed in section 3.2, they mainly correspond to threading dislocations in GaN. When directing several pulses below the one-pulse ablation threshold to the same site of the wafer, laser damage in the form of µm-sized pits already occurs after the initial pulse at those precursors. Those sites extend in number and size during the next pulses and eventually result in a cohesive crater at the beam center (see Supplement 1, S6). We assume that these damage precursors give rise to a locally enhanced absorption coefficient ${\alpha _{\textrm{def}}}$, which effectively enables one-photon absorption relatively independent of wavelength and pulse width. In case of 260 nm, the absorption coefficient of such precursors is in the same range as that one of the defect-free crystal, so that pit formation below the ablation threshold cannot be observed. Consequently, the accumulation effect is weak at 260 nm $[{{F_{\textrm{th}}}({100} )/{F_{\textrm{th}}}(1 )\approx 0.5} ]$.

In addition to that, we investigated the impact of the polarization state on the accumulation behavior for a specific parameter combination (520 nm and 0.38 ps pulse width). As shown in the graph in Fig. 3(a), the accumulation effect is less pronounced for circularly polarized light than for the linear polarization state. We assume that pulse-to-pulse damage expansion in lateral direction around the precursors is enhanced by the formation of LIPSS, which only arise under linearly polarized laser pulses [27,28].

3.2 Threading dislocations as damage precursors

When exposing the GaN surface to a fluence slightly below the ablation threshold, laser damage in the form of pits with diameters between 200 nm and several µm appears. This effect can be observed for all tested wavelengths (except for 260 nm) and pulse widths. An example of the resulting surface morphology is shown in Fig. 4(b) and (c). Based on their limited size and erratic distribution, those laser-induced pits can easily be differentiated from ablation craters around the beam center. The pits feature either circular or, in case of the smaller structures, elongated shapes, oriented perpendicular to the polarization direction as it is also known from LIPSS [27,28,42]. They can be observed after single pulses, but increase in size and number after multiple pulses. They also appear in the vicinity of craters, i.e., in the outer low-fluence region of pulses where the ablation threshold is cohesively exceeded in the beam center.

 

Fig. 4. (a) Applied scanning pattern (red dots) in this set of experiments. The pulse parameters remained constant during one scanning sequence. The green circle represents the 1/e² beam width at 520 nm. (b) Laser damage in the form of pits after scanning across the sample at 520 nm, 0.38 ps pulse width and a peak fluence of 0.47 J/cm². (c) As (b), with higher magnification of individual pits, revealing the black core in the center, which likely corresponds to a threading dislocation. (d) Full-area ablation and LIPSS formation at a peak fluence of 0.58 J/cm². (e) Pit density after scanning across the sample as sketched in (a), in dependency of the peak fluence per pulse, for sample A and B, 520 nm wavelength and 0.38 ps pulse width. (f) Surface morphology after scanning across the sample with 520 nm pulses in dependency of the peak fluence and the pulse width (sample A). In the hatched region, the morphology locally varied. 1-on-1 and 10-on-1 ablation thresholds, obtained in section 3.1, are given for comparison.

Download Full Size | PDF

A homogeneous, extensive distribution of those pits can be created when scanning the beam across the sample surface with a dense pattern of impinging pulses, as sketched in Fig. 4(a). During a scanning sequence, the beam is shifted laterally after every impinging pulse, while the laser parameters (wavelength, pulse width and energy / peak fluence) are held constant. Hence, the procedure differs from the S-on-1 experiments presented in section 3.1 A typical example of the resulting surface morphology below the ablation threshold is shown in Fig. 4(b), for a wavelength of 520 nm and a pulse width of 0.38 ps, obtained on sample A. The density of pits is fluence-dependent, as the graph in Fig. 4(e) reveals. For the described parameter combination, laser damage in the form of pits sets in at a peak fluence per pulse slightly below 0.3 J/cm², reaches a plateau and then increases again at around 0.45 J/cm². At higher fluences of 0.5 J/cm², when the ablation threshold is consistently reached across the surface, an abrupt transition to a surface covered with LIPSS occurs, as shown in Fig. 4(d). The pit density in the plateau region between 0.35 and 0.45 J/cm² increases with the density of threading dislocations in the GaN film, as shown by the comparison of data from sample A and B in the graph in Fig. 4(e). This points to the role of threading dislocations as damage precursors, which will be discussed in detail below. In the plot in Fig. 4(f), the morphology of the GaN surface is categorized in dependence of peak fluence and pulse width for the wavelength of 520 nm (sample A), revealing the transition from the laser damage regime to complete ablation.

In order to further elucidate the assumed role of threading dislocations in the formation of laser damage, a set of different experiments was conducted. In all cases, the pits were generated on sample B at a wavelength of 520 nm and a pulse width of 0.38 ps, applying the dense scanning pattern shown in Fig. 4(a). In Fig. 5(a) and (b), an SEM image of the pristine GaN surface with the associated panchromatic CL map is shown. The dark spots in Fig. 5(b) correspond to threading dislocations, which provide an effective path for non-radiative recombination [43]. In general, both screw and edge-type dislocations contribute to the dark spot pattern in CL, which occur with a density of 5.5 × 10⁸cm^-2 for sample B. After exposure to laser pulses below the ablation threshold, the same surface sites were checked by CL a second time, as visible in Fig. 5(c) and (d). Approximately 90% of the laser-induced pits coincide with CL dark spots, i.e., threading dislocations. For this analysis, a surface area of approximately 400 µm² was investigated. Both the circular pits with diameters up to the µm regime as well as the smaller elongated structures form on threading dislocations. The remaining 10% of the pits show a smaller diameter and can partly be attributed to particles on the surface during laser treatment.

 

Fig. 5. (a) SE and (b) panchromatic CL image showing the dark spot distribution at a particular sample site of the GaN film (sample B). (c) SE and (d) CL images at the same site after laser treatment at 520 nm, 0.38 ps and a peak fluence of 0.42 J/cm². In (c), the position of dark spots has been marked with yellow crosses.

Download Full Size | PDF

In order to identify which types of threading dislocations are involved, wet etching tests with concentrated phosphoric acid (H₃PO₄) at 160 °C have been performed. The results of the study are presented in Fig. 6. In Fig. 6(a) and (b), SEM images of the GaN film are shown, which was etched in H₃PO₄ for 15 and 10 min, respectively. Two types of etch pits show up, ‘giant’ and ‘medium’. The medium pits, particularly visible in Fig. 6(b), feature a density of 1.7 × 10⁸cm^-2, which amounts to roughly 30% of the overall threading dislocation density obtained from CL. According to literature, dislocations with a screw component are etched more readily in H₃PO₄, so that we attribute these medium pits to both screw and mixed-type dislocations [44,45]. This assignment is supported by electron channeling contrast imaging, see Supplement 1, S7 [46,47]. The density of the giant pits in Fig. 6(a) is around 2.0 × 10⁶cm^-2, which corresponds to < 0.5% of all dislocations. According to recent studies by Usami et al., as well as earlier investigations, such giant etch pits are likely related to open-core pure screw dislocations [48–51]. In Fig. 6(c), the identical surface area as in Fig. 6(b) is depicted, after exposure to laser pulses at 520 nm, employing the described scanning pattern. Clear laser damage occurs at the giant etch pits, but also additional pits were formed. Their positions are marked with circles in the image in Fig. 6(b). Approximately 2/3 of all marked pits coincide with screw- or mixed-type dislocations (circles marked in red), in particular all the larger circular pits do so. However, there are also sites where no clear etch pit has been formed by the previous treatment in H₃PO₄ (marked in green). Based on the CL data, 90% of the laser-induced pits are correlated to threading dislocations, so we conclude that the sites marked with green circles result from both edge-type dislocations as well as other damage precursors such as surface particles. This hypothesis will be strengthened by TEM results.

 

Fig. 6. (a) / (b) SEM images of the GaN surface after 15 min / 10 min etching in H₃PO₄ at 160 °C. (c) Same surface area as in (b), after additional laser treatment at 520 nm, 0.38 ps and a peak fluence of 0.42 J/cm². The sites of later laser damage have been marked in (b). (d) SEM image of a particular area after laser treatment at 520 nm, 0.38 ps and a peak fluence of 0.42 J/cm², showing the characteristic distribution of laser-induced pits. (e) Same area as in (d), after additional 15 min etching in H₃PO₄. (f) Closer view to a region in between the big etch pits shown in (e), with small dark pits becoming visible. All results were obtained on sample B.

Download Full Size | PDF

In a second set of experiments, the sample was initially exposed to laser pulses, resulting in the distribution of pits shown in Fig. 6(d). Those laser-induced pits act as starting points for locally accelerated etching in H₃PO₄, as demonstrated by the image in Fig. 6(e), which was taken at the same sample site after 15 min of wet-etching. Apparently, the depth of the hexagonally-shaped etch pits differs. This is presumably connected to the penetration depth of the laser damage into the thin film, since the degraded material is readily etched in H₃PO₄. It is likely that sites of smaller damage, i.e., at surface particles or edge-type dislocations, result in rather shallow etch pits. Moreover, giant etch pits as visible in Fig. 6(a) do not appear in Fig. 6(e), even though wet-etching conditions were identical. This implies that the corresponding sites have been transformed into laser-induced pits. Hence, open-core screw dislocations seem to offer an amplified damage potential both in wet and laser-based etching. In Fig. 6(f), a closer view of the surface presented in Fig. 6(e) is shown. In comparison to Fig. 6(b), i.e., the wet-etched surface without initial laser treatment, small dark spots become visible. They feature a high density of 3.7 × 10⁸cm^-2 and most likely correspond to pure edge dislocations. This number also matches the CL dark spot density if added up with the density of screw and mixed-type dislocations, as summarized in Table 1. With respect to the structural sample condition, these findings imply that the impinging laser pulses seem to alter each dislocation. Even though this does not result in pit formation at the surface in case of most of the pure edge dislocations, they are at least attacked more readily when subsequently exposed to H₃PO₄.

Table 1. Comparison of defect densities as measured with different techniques and their associated dislocation types (obtained with sample B).

View Table

In addition to the CL and wet-etching analysis, cross-sectional TEM measurements were conducted. Employing the FIB-SEM, lamellas were prepared at positions of laser-induced pits on sample B. Two exemplary results are presented in Fig. 7. In each set of images, the same cross-sectional area is depicted under different imaging conditions. The FIB-STEM images in Fig. 7(a) and (d) reveal the structural condition of the samples. The first sample in (a) shows lines with clear contrast along the c-axis underneath the two pits. While in case of the left pit two closely spaced narrow lines lead into the pit, there is one broad defect underneath the right pit. Its lateral width varies between 10 nm further towards the substrate and 35 nm close to the surface. According to the weak beam dark field (WBDF) images in Fig. 7(b) and (c), taken under different g vectors, the dislocation line under the pit on the right has a distinct screw component, but vanishes when imaged with a g vector of [03-30]. Based on the g${\cdot} $b-criterion, this hints towards a pure screw-type dislocation [52]. The extended lateral width is known from so called nanotubes, for which diameters between 2 and 40 nm are reported in literature [53,54]. Those defects are often, but not exclusively, connected to open-core screw dislocations [52,55,56], which are likely to cause the giant etch pits shown in Fig. 6(a). The striking contrast change along the defect line, which is apparent in the STEM image in Fig. 7(a), could be related to alternating open and filled segments within the nanotube [53,57]. It is also likely that the intense laser pulses do not only lead to ablation at the surface, creating the pit, but also affect the structure of the underlying dislocation. Hence, the bright contrast could correspond to metallic gallium which was released during laser-induced GaN decomposition along the dislocation line. Contrarily, the two closely spaced lines underneath the left pit do not clearly fade under one of the WBDF imaging conditions, so that they might be related to mixed-type dislocations. In case of the second sample, an isolated laser-induced pit is depicted. In the STEM image in Fig. 7(d), a bunch of dislocations can be seen in the vicinity, even though none of them seems to run directly into the pit. Clear signs of crystal damage are visible underneath, which extend 1-2 µm into the GaN film. Regarding the WBDF images in Fig. 7(e) and (f), the dislocations around the pit can be attributed to edge-type dislocations, since they show a clear contrast under a g vector of [02-20], shown in Fig. 7(f). In this image, one can also see a dislocation which ends in the center of the pit. This finding substantiates the conjecture that also edge-type dislocations show precursor activity for laser pulses.

 

Fig. 7. Cross-sectional analysis of two different sites of laser damage. (a), (d) STEM images after lamella preparation in the FIB-SEM. (b), (e) TEM WBDF images obtained with a g vector along the c axis to identify screw components. (c), (f) TEM WBDF images obtained with a g vector along the m axis to identify edge components. The site of laser damage in the upper image sequence seems to relate to an open-core pure screw dislocation (right crater), and two dislocations running in parallel (left crater), possibly mixed-type. In the lower row, the crater is located on a pure edge-type dislocation.

Download Full Size | PDF

The presented data demonstrate that threading dislocations efficiently take effect as precursors for laser damage in GaN, especially in case of screw-type dislocations. It is generally assumed that threading dislocations in GaN act as accumulation centers for point defects, which shape the local electronic structure [58]. Depending on the type of dislocation, those point defects give rise to a range of states within the band gap and may noticeably enhance absorption [59]. In n-GaN, the defect states within the gap are filled by electrons up to the Fermi energy, which leads to a negatively charged region screened by ionized donors [60,61]. This creates local electric fields and band deformations as illustrated in Fig. 8(b). Based on transmission electron holography measurements, negative charge densities as high as 7 × 10¹⁹cm^-3 have been reported, filling a cylinder of several nm diameter around the dislocation core [60,62]. Regarding the location of the occupied defect states deep within the band gap, a seed population in the conduction band can readily be created by defect-to-band transitions also in case of below-band gap radiation [63]. For IR wavelengths, where the photon energy might not be enough to raise electrons from the uppermost occupied defect states into the conduction band, lower-order absorption processes or transitions via higher unoccupied states may still increase the overall absorption cross-section [64]. As indicated by the strong accumulation effect present in GaN at visible and IR wavelengths, the activity of dislocations as precursors is additionally enhanced by previous laser pulses. This is likely caused by ongoing laser-induced chemical modifications around the dislocation line, e.g. by the capture of further conduction band electrons after a pulse or generation of additional trap states [65]. As in the case of defect-free regions at higher fluences, intraband absorption and impact ionization act on the created seed carrier population and cause ablation once a critical density is exceeded. During this process, diffusion of highly excited free carriers can lead to the formation of a radially expanding absorption front. This way, pits with larger diameters than the width of the actual dislocation may arise [66,67]. Consistently, Yoo et al. observed similar laser-induced pits, but with bigger diameters up to tens of µm, when directing nanosecond pulses at 1064 nm and fluences > 5 J/cm² to an n-doped GaN surface [21].

 

Fig. 8. (a) Schematic illustration of the ablation process at dislocations when the ablation threshold of defect-free material is not yet exceeded. The vertical lines represent different dislocations. (b) Band diagram for an electronically active dislocation with dispersed states within the band gap in n-doped GaN. The states are filled up to the Fermi level, resulting in band bending. Typical values of R and ${V_\textrm{B}}$ are 20 nm and 2.2 V, respectively [58]. Possible photon-driven transitions of electrons to higher states are illustrated.

Download Full Size | PDF

Our results imply that screw and mixed-type dislocations provide more effective excitation channels than edge-type. As reasoned before, this is likely caused by different densities of defect states within the band gap. A dense distribution of states within the band gap, delocalized along the dislocation line, also leads to enhanced electrical conductivity along those one-dimensional defects and thus vertical leakage channels in electronic devices like pn-junctions or vertical FETs [48,49,68]. Even though this has controversially been discussed throughout the last decades, recent results imply that screw dislocations, especially those with a closed core, exhibit the strongest electronic activity, whereas edge dislocations are likely to be rather electronically inactive [51,59,69].

Besides a modified electronic structure at dislocations, geometric aspects have to be taken into account. According to the etching tests in H₃PO₄, especially open-core screw dislocations act as damage precursors for ultrashort laser pulses. Even if they do not provide the same density of dispersed gap states as their closed-core counterparts, the activity may additionally be enhanced by their geometric shape. According to Bloembergen, surface imperfections exceeding a critical size of 10 nm can lead to a strong local field enhancement [70]. In literature, diameters of up to 40 nm are reported for open-core screw dislocations, which consequently may further amplify laser damage formation [53]. As argued by Bloembergen, chemical and geometric features can simultaneously contribute to the precursor activity of defects.

4. Conclusion

In this paper, we describe the influence of various parameters on the ablation threshold of unintentionally doped GaN films with laser wavelengths from the IR to the deep UV. A clear impact of the wavelength can be observed, which leads to a steady increase of the ablation threshold for single sub-ps pulses from 0.15 to 0.83 J/cm². Additionally, an increase of the one-pulse ablation threshold with the pulse width is apparent in the range of 0.3 to 10 ps for the wavelengths of 347, 520 and 1040 nm. We attribute this behavior to a wavelength and pulse width dependent generation of seed carriers in the conduction band, based on multiphoton absorption, which is consistent with a simple rate equation model. In case of a wavelength of 260 nm, where one-photon interband absorption is dominant throughout the whole pulse duration, we contrarily found a slight decrease of the ablation threshold with increasing pulse width. This behavior may be caused by phase space filling, due to carrier thermalization times in the low picosecond range.

When testing the multi-pulse ablation threshold, we observe a strong accumulation effect in particular for below-band gap radiation. The reduction of the ablation threshold after multi-pulse treatment is closely linked to the activity of damage precursors. Laser damage in the form of µm-sized pits appears at fluences below the one-pulse ablation threshold. These sites grow in number and size during subsequent pulses and eventually merge to a coherent crater, effectively reducing the multi-pulse ablation threshold. We could demonstrate that those precursors reach densities in the range of 10⁷cm^-2 and are linked to threading dislocations in GaN. A thorough investigation including CL, wet-etching tests in H₃PO₄ and TEM measurements revealed that in particular threading dislocations with a screw component exhibit precursor activity, which is attributed to a distinct distribution of dispersed states in the band gap for this type of dislocation. Further, geometric aspects could play a role for open-core screw dislocations. We conclude from these findings that dislocations do not only act as centers of non-radiative recombination and promote yellow luminescence, but also act as local hot-spots during optical excitation. This phenomenon must be kept in mind when processing GaN-based devices with lasers, since those precursors may give rise to laser damage in the vicinity of laser-defined structures. In particular, this effect may enhance leakage current when using ultrashort pulses for laser lift-off, where a fraction of the laser light reaches the active area on the front side of the wafer [6].