Layer transfer of bulk gallium nitride by controlled spalling

Interestingly, part of the commercial success of gallium nitride-based solid state lighting is attributed to the defect-tolerant nature of these materials with respect to optical generation. This defect-tolerance permits functional and efficient blue multi-quantum well (MQW) diode structures to be grown on highly lattice mismatched substrates such as sapphire or silicon carbide. For applications that demand higher current density, such as laser diodes or power electronics, crystalline defects in GaN have been shown to degrade both the breakdown voltage and lifetime of the devices.^1,2 For those applications, heteroepitaxial GaN with high defectivity (∼10⁸ cm⁻²) is inadequate and bulk or “free standing” GaN wafers with a lower defect density (∼10⁶ cm⁻²) become necessary. There are two important challenges to overcome for enabling widespread adoption of bulk GaN substrates; the first is the high cost associated with producing these wafers, and the other is that the limited thermal conductivity of GaN would require substrate thinning for high power applications. Because GaN is very hard and chemically inert, mechanical thinning is difficult in practice.

A layer transfer technique applicable to bulk GaN wafers, which does not perturb the crystal quality, could potentially address both of these challenges; however, such a process has not yet been demonstrated. Laser lift-off has been successfully applied to GaN on sapphire; however, attempts to use a modified laser lift-off process³ on bulk GaN required the use of a buried epitaxial layer (InGaN) that compromised the GaN quality. Attempts to use ion implantation based layer transfer have been made,^4,5 and successful wafer scale bonding was demonstrated; however, the persistent ion damage had to be removed, and subsequent epitaxial GaN regrowth was performed to recover low-defectivity material.⁵ Likewise, other lift-off techniques generally require specialized buried epitaxial layers (ZnO, graphene, etc.)^6,7 that tend to compromise subsequent GaN epitaxy quality.

In previous work, we described the Controlled Spalling process^8,9 for material-agnostic wafer scale layer transfer. We have demonstrated the separation of epitaxial GaN light-emitting diode (LED) layers from sapphire substrates using Controlled Spalling and have shown functional devices and equivalent material quality.¹⁰ Controlled spalling has also been successfully applied to completed Si CMOS circuits¹¹ as well as Ge and GaAs-based solar cells.¹² 

In this paper, we extend the Controlled Spalling process to include wafer scale layer transfer of bulk GaN substrates that does not alter the defectivity within the error of XTEM analysis. This fracture-based method obviates the need for specialized epitaxial layers or ion-implantation steps for GaN layer transfer, both of which introduce damage to the crystalline structure. Additionally, we use spalling as a new tool to measure important material properties of high-quality GaN.

As described previously, the Controlled Spalling process relies upon the deposition of a tensile stressor layer on the surface of the substrate to induce fracture and propagate a subsurface crack across the wafer. The stressor material itself must be able to possess high values of tensile stress without yielding or cracking, and Ni has been found to be suitable for this purpose. Ni has the additional advantage that deposition with a controlled amount of tensile stress is relatively straightforward for both sputtered and electrodeposited films.

The substrates used in this work were 300 μm thick, HPVE grown n-type 2″ bulk GaN wafers from a variety of suppliers with no observed difference in the spalling process window for the different wafers. DC magnetron sputtering was first used to deposit a 90 nm thick Ti adhesion layer followed by a 1 μm thick Ni seed layer on both sides of the wafer. Next, the remainder of the Ni stressor layer was deposited by electroplating using a NiCl₂ and H₃BO₃ based solution at room temperature. Vacuum transferred through an electrically conductive elastomer was used to keep the substrate flat throughout the spalling process and permit electrical connection to the wafer during Ni plating while allowing deposition up to the very edge of the wafer. The tensile stress in the Ni was measured by wafer bowing and was consistently between 450 and 500 MPa independent of the thickness (up to 30 μm). By adding 10 g/l NH₃Cl, the stress could be increased to approximately 800 MPa. The Ni deposition rate was fixed at 1 μm/min using a constant current density of 50 mA/cm². After Ni deposition, a sheet of 25 μm thick polyimide tape was roll applied to the surface of the Ni plated wafer to serve as the handling layer. By lifting one edge of the tape, the fracture was initiated and propagated as a single fracture front using the handling layer to mechanically guide the cleaving. This manual method of guiding fracture does not require any additional “pulling” force on the handle layer; once the crack is initiated at the wafer edge, the mechanical energy to propagate fracture is contained in the Ni stressor layer. This mechanically guided singular fracture front enables wafer scale crack-free layer transfer and is referred to as Controlled Spalling. An illustration of the process is shown in Fig. 1(a), and an image of a bulk GaN wafer undergoing the process (using 500 MPa Ni) is shown in Fig. 1(b). A plot of the measured Ni plating rate as a function of the distance along a 2 in. wafer is shown in Fig. 2 and includes data points to within 200 μm of the wafer edge.

There is a combination of stressor thickness and stress values above which spalling mode fracture is energetically possible and is referred to as the critical loading condition. If the fracture toughness of the material is known, then the critical loading conditions can be calculated using existing fracture mechanical models. In this work, we inverted the problem by using the experimentally determined critical loading conditions to calculate the effective fracture toughness of the bulk GaN substrates by fitting those results to the model¹³ by Suo and Hutchinson (S&H),

In Eqs. (1) and (2), K_I and K_II are the type I and type II stress intensity factors, respectively, P is the edge load, M is the moment induced by the stressor layer, U, V, and γ are dimensionless constants originating from the calculation of elastic energy stored in the beam structure behind the crack tip, h is the stressor layer thickness, and ω is a dimensionless number that depends on the elastic dissimilarity of the stressor and the substrate and also depends on the crack depth. The procedure for calculating these values is given in the Appendix of S&H, and the specific calculation for the case of Ge substrate spalling is given in the supplementary section of Ref. 8. The earlier analysis assumes a pre-existing substrate crack and therefore represents critical loading for the case where crack initiation has already occurred. Generally, the fracture is crack initiation-limited, and therefore, the above analysis does not represent the mechanical conditions for spontaneous fracture.

In previous work, we showed a dense data set of Ni stress and thickness combinations that defined a continuous Controlled Spalling window for 4 in. Ge wafers.⁹ However, a number of limitations in the present study limit both the Ni stress range and number of data points permissible when studying the GaN spalling window. First, whole wafers are used to avoid edge-related effects observed on small samples and the limited availability and very high cost of bulk GaN substrates restricted the number of full-wafer spalling experiments compared to equivalent Ge or Si experiments. Also, the Ni thickness required to spall GaN is roughly four times that of Si and therefore electroplating becomes the only practical method of stressor deposition. The Ni plating process must provide a constant stress value throughout the thickness of the film to avoid bending moments not included in the S&H analysis and only a limited number of bath configurations were studied.

In order to determine the critical loading condition using a single substrate, a bulk GaN wafer was secured to a 400 μm thick sapphire wafer using the double-sided polyimide tape to keep it flat, and Ni was electroplated on the surface of the GaN to an initial thickness. The wafer was then removed from the bath, and a small piece of tape was mounted on the surface of the Ni to exclude that region from subsequent Ni deposition. This process was repeated until a range of Ni thickness regions existed and spontaneous spalling occurred from the edge. Once spalling spontaneously initiates, it propagates until the fracture front reaches a region where the Ni is below the critical thickness. The stress for this method of depositing Ni was measured using a sapphire monitor wafer and found to be 570 MPa. Figure 3 shows an image of the test sample used to determine the critical Ni thickness for spalling and the fracture can be seen to stop at the third Ni region. The Ni thickness was then measured accurately by masking off areas within the different regions and etching down to the GaN surface using FeCl₃-based Ni etchant and measuring the step height using profilometry. Also, the fracture remained arrested after removing the small tape mask, suggesting that the effect of the tape on the critical loading condition is small in this case.

These data are summarized in Fig. 4 and also included are four data points representing electroplated Ni thickness and stress values that resulted in full 2 in. (Ga-face) bulk GaN wafer spalling including the 25 μm thick handle layer. For the 800 MPa Ni stressor layer, spalling became possible at a stressor thickness of 8.0 ± 0.5 μm whereas for a 500 MPa Ni layer 16.0 ± 0.5 μm was required. Since these values for Ni represent the critical loading conditions for bulk GaN spalling-mode fracture (K_II = 0) parallel to the [0001] planes, Eqs. (1) and (2) can be used to determine the fracture toughness (K_IC) by calculating the value of K_I at a depth in the GaN corresponding to K_II = 0. Using this analysis, a value for K_IC = 1.7 MPa $m$ was obtained, and the critical load range corresponding to this value is represented by a dashed curve in Fig. 4. For combinations of Ni stress and thickness below this line, the spalling-mode fracture is not possible and defines the subcritical region. If deposition of the stressor layer continues long enough, fracture will self-initiate, usually at imperfections at the edge of the substrate. The remaining metastable region defines the operable range for Controlled Spalling.

Figure 5 shows a picture of the thin GaN surface layer after spalling (using 22 μm of 500 MPa Ni) and mounting the handle tape to a metal frame to prevent film curling. Superimposed on the image are the spalled GaN film thickness measurements (in units of μm) made using a calibrated Oxford Instruments X-ray fluorescence analyzer. The spalled GaN film thickness is primarily dependent on the Ni film uniformity although local fracture perturbations can cause spalling depth variation. Figure 6 shows the surface morphology after fracture for two distinct regions on the residual (post-spalled) bulk GaN wafer. For surface regions where fracture occurred at a relatively constant velocity, the resulting morphology is smooth and uniform with an amplitude of approximately ±500 nm over a length scale of several millimeters. The [0001] planes are not cleavage planes for the wurtzite GaN system; recent work¹⁴ suggests that the m- and a-planes permit proper cleavage, and therefore, spalling of semipolar and non-polar GaN wafers may have much lower as-spalled roughness. When fracture propagation is arrested and restarted, a line feature is created on the surface, orthogonal to the direction of propagation. The profilometry data in Fig. 6 show that these “start/stop” artifacts represent local depth perturbations with an amplitude between 1 and 2 μm.

Figure 7 shows cross-sectional transmission electron microscopy (X-TEM) images from the residual (post-spalled) substrate and the corresponding spalled film (using the 500 MPa Ni process). These images represent the as-fractured surfaces from both sides of the crack path: the bulk substrate side and the spalled film side. No dislocations were observed in either of the prepared samples (low magnification inset) indicating that the material quality has not been altered within the detection limit of these XTEM images (mid 10⁷ cm⁻²). The high magnification images show the absence of any near-surface structural damage confirming that the fracture was strictly brittle.¹⁵ 

Once spalling is complete and a thin layer of GaN is removed from the substrate, the handle, stressor, and spalled layer constitutes a three layer strain-sharing stack. Figure 8 shows an illustration of the strain change in the film stack after spalling (in the absence of bending). By assuming the film stack is held flat (in practice, by using a vacuum chuck) and the films are in equilibrium (⁠$∑Fx=0$⁠), the strain change (ΔL/L) is given by

where M_x is the biaxial modulus, h_x is the thickness, and ε_x is the initial (before spalling) strain of layer x (x = 1 is the GaN, 2 is the Ni, and 3 is the polyimide handle layer). Because the length of these films is on the order of centimeters, very small changes in film length can easily be measured on a precision x-y sample stage. By comparing the distance between known features on the bulk wafer and the same features on the spalled film, the GaN strain (Δε) can be measured directly. This can be repeated for a variety of spalled GaN films with varying Ni thickness and stress to provide a range of GaN strain values. These compressively strained GaN films can be measured using Raman spectroscopy to establish the E₂(high) peak shift as a function of strain without any reference to material constants. Figure 9 shows the relationship between the Raman E₂(high) peak shift and the direct strain measurement for a range of spalled 2 in. bulk GaN films. Care must be taken to measure film strain over regions with uniform GaN and Ni thickness because this method yields average strain, whereas the Raman data are localized. The best-fit line to the data gives Δω(cm⁻¹) = 1411ε, with the error bars giving an upper bound of the strain coefficient of 1420 (cm⁻¹) and a lower bound of 1320 (cm⁻¹). There are at present three values^16–18 often cited in literature for the Raman E₂ peak shift coefficients and they are given in terms of stress. If we use the biaxial modulus for a hexagonal crystal structure: $MBiax=C11+C12−2C132C33$⁠, a value of 479 GPa is calculated. Converting the above strain Raman coefficient to stress gives Δω(cm⁻¹) = 2.95σ(GPa), in good agreement with the Davydov et al.¹⁶ value of 2.7 ± 0.3 cm⁻¹/GPa, but significantly different from the coefficient proposed by Kisielowski et al.¹⁷ (4.2 cm⁻¹/GPa) and that of Kozawa et al.¹⁸ (6.2 cm⁻¹/GPa).

Although we have demonstrated that Controlled Spalling can be used to remove thin surface films from bulk GaN substrates without generating defects or cracks in either the film or the remaining wafer, the surface artifacts presently make wafer reclaim challenging. Improvements in the quality of as-fractured surfaces by spalling with constant velocity (no start/stop artifacts) or new higher speed grinding and lapping techniques are required for successful wafer reclamation. Because some of the bulk GaN wafers used in this study were double-side polished, after spalling the Ga-face side, the wafer was flipped over, and the N-face was subsequently spalled. No obvious differences were noted in the critical loading conditions or the fracture characteristics between Ga and N-face spalling.

The calculated fracture toughness value K_IC of 1.7 MPa $m$ differs significantly from the literature value¹⁹ of 0.79 MPa $m$ measured by indentation. This is likely due to the fact that fracture is occurring predominantly on the [0001] planes in this work, whereas indentation is likely inducing fracture on the natural cleavage planes. Analysis using the S&H model also provides a calculation of spalling depth. Applying the conditions used for the layer transfer shown in Fig. 5 (500 MPa, 22 μm Ni) yields a calculated spalling depth of 22.1 μm, in reasonable agreement with the thickness values measured on the sample. However, for the higher stress Ni condition (800 MPa, 8.9 μm Ni), the calculated spalling depth is 13.2 μm, but the measurement indicated depth variation from 9 to 18 μm. The strong correlation between Ni thickness and spalling depth observed in Ge and Si⁹ is not observed in the GaN system and requires additional study.

In this work, we have experimentally determined the critical loading conditions for Controlled Spalling of bulk GaN substrates using Ni stressor layers. By comparing these loading conditions with the S&H analytical model, an effective fracture toughness value of 1.7 MPa $m$ was calculated and represents fracture parallel to the [0001] planes. XTEM analysis showed that the crystalline quality is the same before and after spalling within the error of the technique unlike other bulk GaN layer transfer approaches. Upon release from the bulk GaN substrate, the multilayer film stack (polyimide/Ni/GaN) undergoes in-plane contraction due to the force exerted by the tensile Ni layer. By holding these films flat and measuring the amount of this contraction using a precision x-y sample stage, a direct measurement of film strain could be determined without reference to any material parameters. By comparing the measured film strain with the E₂(high) Raman peak shift, a measured relationship of Δω(cm⁻¹) = 1411ε was determined. These constitute the first such measurements of strained bulk GaN (low defect density) and when converted to a stress-relationship Δω(cm⁻¹) = 2.95σ(GPa), it is found to agree well with the coefficient of Davydov et al. (2.7 cm⁻¹/GPa) but in disagreement with those of Kisielowski et al. (4.2 cm⁻¹/GPa) and Kozawa et al. (6.2 cm⁻¹/GPa).

This work was supported, in part, by the ARPA-E SWITCHES program. The authors would like to thank the IBM management team for their support.