Introduction

The GaP/AlGaP system has gained interest for use as crystalline mirror coatings in precision measurements that demand extremely low thermal noise in the optics: long baseline interferometer-based gravitational wave detectors and stable cavities in optical atomic clocks [1,2]. Crystalline coatings comprising of III-V semiconductors, including this work, have been shown to exhibit much lower mechanical losses than conventionally used amorphous oxides [3,4]. As mechanical loss is directly related to thermal noise, crystalline coatings have low thermal noise as well [5,6]. To date, research-scale crystalline coatings of AlGaAs/GaAs that were grown on GaAs substrates, lifted-off, transferred, and then bonded to silica, yielded a ten-fold reduction in thermal noise compared to state-of-the-art amorphous coatings [3].

One key advantage of the GaP/AlGaP system is the ability to deposit mirrors directly onto Si, which avoids the lift-off, transfer, and bonding process required for non-lattice-matched systems such as GaAs/AlGaAs [3]. The GaP/AlGaP system leverages the use of Si as the optical mirror substrate, which has low thermal noise and absorption around 1550 nm and enables the possibility of scaling the mirror with the size of the Si substrate. Silicon boule sizes are currently available up to 450 mm in diameter whereas there is little commercial driving force to grow GaAs boules beyond 150 mm in diameter. The size of the substrate determines the largest possible crystalline mirror that can be produced and because GaP/AlGaP is closely lattice-matched to Si (0.37% mismatch at room temperature), it can be grown directly on the final substrate and is not limited in size by an intermediary growth substrate such as GaAs in the case of GaAs/AlGaAs (4% mismatch to Si). This is a distinct advantage for applications, such as the proposed future gravitational wave detectors [4], which demand large area optics.

Epitaxial integration of GaP on Si is possible due to the close lattice-match between GaP and Si but the presence of antiphase domains inherent to polar-on-nonpolar growth can degrade material quality. There has been a large effort in refining Si surface preparation techniques to induce defect-free III-V growth, however, surface preparation alone is insufficient for achieving high quality GaP on Si [7,8]. After proper surface preparation, two-step growth is commonly used to decouple nucleation and growth, where growth rate, substrate temperature, and other parameters can be optimized for each step. In a two-step growth process for growing GaP on Si, the nucleation step offers finer control of adatom migration and incorporation on the Si surface. In the second step, finding a growth regime where antiphase domain annihilation is favored is the key to achieving high quality films [9–12].

We present a study of GaP and Al_0.5Ga_0.5P films nucleated on Si to elucidate how growth temperature, growth initiation, and nucleation method affect the resulting morphology of the nucleation layer. By influencing adatom diffusion length through macroscopic knobs such as substrate temperature and variations in growth initiation, it is possible to achieve a desired two-dimensional nucleation layer that is suitable for subsequent growth, enabling high-quality GaP films on Si. We then utilize the optimized AlGaP nucleation layers in a 10-pair GaP/AlGaP mirror structure grown on Si and present initial results of the mirror structure. The mechanical loss measurements are described in a separate publication [4]. This work focuses on demonstrating an AlGaP/GaP mirror on Si using molecular beam epitaxy (MBE) to provide high-quality films while avoiding potential complications of impurity incorporation from metal organic chemical vapor deposition (MOCVD) precursors where impurity concentrations are dependent on growth conditions such as growth temperature and Al content in layers [13,14]. Fortunately, MOCVD of GaP on Si is well established [15–18] and may provide a pathway for scaling these mirrors to larger optics.

Experimental procedure

Thin GaP and Al_0.5Ga_0.5P films were grown on Si (100) with 4° offcut towards [110]. The moderately-doped n-type Si substrates were prepared by a modified RCA cleaning method followed by deposition of a 200-nm Si buffer in an Applied Materials reduced pressure chemical vapor deposition (CVD) system. Post-growth annealing of the Si wafer was performed at 1000°C under 40 slm H₂ flow for 20 min to promote double atomic step formation on the Si surface and provide hydrogen passivation [8,19]. After these steps, the Si surface typically had a root mean square (rms) roughness less than 2 Å. The wafers were dipped in dilute hydrofluoric acid for oxide removal and hydrogen termination on the surface before immediately loading into a load chamber connected to two Varian Modified Gen II MBE chambers.

Once in the dual-chamber MBE system, a 750°C anneal for 30 min was performed in a dedicated group IV chamber, in absence of any group V background. The wafers were then transferred to a III-V MBE chamber where they were exposed to P₂ flux for 5 minutes, then growth was initiated with 4, 6, and 12 monolayers (ML) of either GaP or Al_0.5Ga_0.5P at a substrate temperature of 525°C, nominal growth rate of 0.2 μm/hr, and V/III beam equivalent pressure (BEP) ratio of 3.5. This exposure and growth initiation are part of the nucleation step, the first step in our two-step GaP growth process. For another experiment, the Si wafer was exposed to either As₂ or P₂ flux, followed by 10 ML Al_0.5Ga_0.5P and 200 nm GaP buffer layer to investigate antiphase domain annihilation. The parameters investigated in these runs were group-V pre-exposure species (As vs. P), nucleation layer composition (GaP vs. AlGaP), nucleation method (simultaneous deposition of group III and V species vs. alternative deposition of group III and V species, also known as migration enhanced epitaxy (MEE)), and nucleation layer thickness.

Mirror coatings comprising of 10 GaP and Al_0.9Ga_0.1P pairs were grown with quarter-wave thicknesses optimized for highest reflectivity at 1550 nm: 127 nm and 140 nm, respectively, based on dispersion relations from Pikhtin [20]. The mirror structures were grown on both Si and GaP substrates for comparison, with the one grown on Si utilizing an optimized 10 ML Al_0.5Ga_0.5P nucleation layer followed by a 200 nm GaP buffer layer to induce antiphase domain annihilation prior to the mirror layers. The mirror layers were grown at a substrate temperature of 630°C, growth rate of 0.4 μm/hr, and V/III BEP ratio of 6.

The thin films and mirror structures were characterized by a variety of techniques to determine film orientation, surface and interface morphology, reflectivity, and absorption. A Park Systems XE-70 atomic force microscope (AFM) was used in non-contact mode with tips of a nominal radius of curvature of 6 nm. A 200 kV Tecnai F20 transmission electron microscope was used to examine the (110) cross-section of Si and III-V layers. In addition to qualitative evaluation of layer interfaces, TEM provides a way to distinguish antiphase domains in GaP where certain diffraction conditions are used to enhance the difference in beam amplitudes, arising from the lack of two-fold symmetry along <110> in GaP [21,22]. These diffraction conditions effectively enhance image contrast between domains of different orientations. X-ray diffraction reciprocal space mapping along the symmetric 004 and asymmetric 224 directions were performed for the mirror structure. Mirror reflectivity at and around 1550 nm was measured using a white light source and spectrophotometer. Room temperature optical absorption at 1550 nm was measured using two techniques: transmission and reflection from power loss and photothermal common-path interferometry (PCI) [23]. PCI, a well-established technique for measuring absorption in optical coatings, was used to benchmark this first result so that amorphous coatings and future GaP/AlGaP coatings will have a direct comparison.

Results and analysis

Changes in surface morphology and interfacial quality of III-V on Si are seen when different growth variations and composition in the nucleation layer are used. Table 1 summarizes different nucleation layers grown on Si and the corresponding AFM images are in Fig. 1. Based on the morphology and rms roughness, the effect of adding Al into GaP layers at a nucleation temperature of 525°C is similar to nucleating GaP at 325°C. Both conditions yield films with more complete surface coverage and a three times reduction in rms roughness compared to GaP films nucleated at 525°C. Minimizing surface roughness during the first monolayers of nucleation will result in subsequent layers being smoother.

Table 1. List of samples with 1x1 μm² rms roughness values taken from the corresponding AFM data in Fig. 1. AlGaP nucleation layers consistently yield surfaces with lower rms roughness.

View Table

 

Fig. 1 1x1 μm² AFM images of GaP and Al_0.5Ga_0.5P nucleation layers with growth variations: (a) 4 ML GaP by MEE, (b) 2 ML GaP by MBE followed by 4 ML GaP by MEE, (c) 2 ML GaP by MBE followed by 10 ML GaP by MEE, (d) 2 ML GaP by MBE followed by 10 ML GaP by MEE (at 325°C), (e) 4 ML AlGaP by MEE, (f) 2 ML AlGaP by MBE followed by 4 ML AlGaP by MEE, and (g) 2 ML AlGaP by MBE followed by 10 ML AlGaP by MEE. All samples were nucleated at a high temperature (525°C) with the exception of sample (d). TEM images show the interface between the Si substrate and (h) GaP nucleation layer in sample b and (i) AlGaP nucleation layer in sample f. The AlGaP nucleation layer yields an abrupt and uniform interface.

Download Full Size | PDF

The addition of a 2 ML GaP or AlGaP layer deposited by typical MBE co-deposition, preceding the MEE layers provides a barrier and helps to prevent Ga droplets from forming and etching into the Si substrate [24]. Lattice fringes are observed in high-resolution TEM images of sample b in Fig. 1(h) and sample f in Fig. 1(i) indicating excellent crystallinity and no sign of dislocations in these 10-ML-thick nucleation layers. The AlGaP film provides an atomically abrupt interface between it and Si whereas the GaP film interface with Si is rougher, indicating interdiffusion and an island growth mode that is also consistent with the surface morphology. These results indicate that adatom diffusion suppression from the presence of Al adatoms and from reduced substrate temperature are desired to yield smooth nucleation layers.

During the nucleation step, one predominant orientation is nucleated. Domains of the opposite orientation also exist and because the base width of the antiphase domains is determined by the first monolayers, investigating the effect of the first monolayer is extremely relevant. Due to the preference of group V atoms to dimerize and self-terminate to one monolayer, we investigated using different group V species (As and P) as the first monolayer on Si. Group V adatoms are self-terminating at one monolayer under the typical growth temperatures used and provide a more robust process compared to nucleation with group III atoms.

Figure 2 shows cross-section TEM images of As- and P-initiated growths of 10 ML AlGaP followed by 200 nm GaP, under the same growth conditions with the only difference being the group V species to which the Si substrate was exposed for 5 minutes before growing the nucleation layer. The As- and P-initiated samples had many similar features: small antiphase domains less than 20 nm in height, which annihilate near the III-V/Si interface (vertical arrows) and twins/stacking faults propagating from the interface to the film surface (horizontal arrows). The magnified inset of Fig. 2(a) shows antiphase domains which annihilate at 10 nm from the Si interface. One key difference between the two conditions is the P-initiated growth also had large antiphase domains propagating to the film surface, resulting in a bimodal distribution of antiphase domain sizes, which was not present with As-initiation. An explanation for this result is the diffusive behavior of P by diffusing into the Si substrate to leave a non-uniform first monolayer and diffusing along the antiphase boundaries. If there was excess P at the antiphase boundary, the boundary would consist of both Ga-Ga and P-P bonds, forcing the boundary to propagate along the 110 plane rather than along the 111 plane that has potential for annihilating with another 111-type boundary. The driving force for boundary propagation along certain crystallographic planes is explained by antiphase boundary spacing and the ratio of Ga-Ga to P-P bonds at the boundary [25]. Arsenic is also known to replace Si dimers on a (100) surface, providing more uniform coverage than in the case of phosphorus [26,27].

 

Fig. 2 (110) cross-section TEM of GaP on Si with an (a) As-initiated AlGaP nucleation layer and a magnified image of the interface and (b) P-initiated AlGaP nucleation layer. Blue, vertical arrows point to antiphase domains and black, horizontal arrows point to stacking faults in the GaP buffer layer. P-initiation yields antiphase domains which propagate to the film surface.

Download Full Size | PDF

Applying the AlGaP nucleation layer, a 10-pair GaP/AlGaP mirror structure for high-reflectivity at 1550 nm was grown on n-type Si. Similar to a homoepitaxial GaP/AlGaP mirror on a GaP substrate (not shown), all of the layer interfaces are smooth and abrupt in Fig. 3(a). Despite the implicit generation of dislocations from strain relaxation of the GaP and AlGaP layers with respect to Si in Fig. 4, the overall high-quality growth and absence of anti-phase domains disrupting the layer morphology yielded the expected 83% reflectivity. This measured reflectivity matches the reflectivity curve calculated by transfer matrix method and AlGaP dispersion relations from Pikhtin [20], as shown in Fig. 3(b) with the red point and blue curve, respectively. Achieving the target reflectivity that was predicted for the 10-pair layer structure at 1550 nm shows promise for reaching higher reflectivity with more mirror pairs. Most applications will require greater than 99.9% reflectivity, or at least 40 pairs of GaP/AlGaP, but adding more mirror pairs is relatively straightforward once an initial, antiphase domain-free GaP buffer layer on Si is established.

 

Fig. 3 (a) TEM cross-section of the AlGaP/GaP mirror structure on Si indicating abrupt and smooth interfaces and (b) 83% reflectivity achieved for a 10-pair mirror as expected from calculations. Red point is the experimental measurement; blue curve is calculated by transfer matrix method and AlGaP dispersion relations in [20].

Download Full Size | PDF

 

Fig. 4 XRD RSM data shows evidence of strain relaxation along the (a) 004 symmetric and (b) 224 asymmetric scans. Based on the ~0.4% mismatch between GaP/AlGaP and Si, the peak positions in reciprocal space are expected to be separated, as seen in this data, and implies that dislocations were generated to allow film relaxation.

Download Full Size | PDF

Optical absorption in the current mirror could arise from several sources: diffusion of Si into GaP buffer layer, the presence of antiphase domains where Ga and P are substitutional impurities, and impurity incorporation from the growth process. Data from secondary ion mass spectroscopy (SIMS) reveals carbon and oxygen concentrations of 10¹⁶ and 10¹⁷ cm⁻³, respectively, throughout the mirror structure and both peaking at the Si/GaP interface, indicative of a monolayer or less at the interface. All of these impurities could be a contribution to the 2.3% absorption at room temperature, measured using PCI and power loss measurements. Oxygen tends to get incorporated at the beginning of every AlGaP layer due to the gettering effect of Al (higher solid solubility of O in Al-containing layers) while the concentration of C is relatively uniform throughout the GaP and AlGaP layers. These contaminants from MBE growth could be reduced by using a dedicated chamber for these mirrors. At the time, issues with the MBE tool precluded the growth of additional material to isolate different absorption mechanisms in the mirror. Future work should be aimed at reducing the impurities as well as performing low temperature PCI measurements to gain insight into any temperature-dependent mechanisms and correlate these mechanisms with materials metrology data.

Mechanical loss, a standard proxy for thermal noise, of the 10-pair crystalline GaP/AlGaP mirror on Si was 1.2*10⁻⁵ at 12K. This value is a factor of 14.3 lower than ion beam sputtered amorphous SiO₂/Ta₂O₅ mirrors at room temperature and a factor of 64 lower than the same SiO₂/Ta₂O₅ mirrors at 20K [4, 28]. This low level of mechanical loss shows great promise for these epitaxially-integrated crystalline coatings to Si and is encouraging for further development of these GaP/AlGaP coatings. In these studies, AlGaP nucleation layers provided significantly smoother interfaces and surfaces. As-initiated monolayers on Si yielded smaller antiphase domains which annihilate within the buffer layer, rather than propagating to the film surface.

Conclusions

These findings have demonstrated the high-quality growth of GaP/AlGaP on Si, which exhibit extremely low levels of mechanical loss. Therefore, GaP/AlGaP crystalline coatings provide a pathway towards reducing the effect of thermal noise and hence increasing the sensitivity of future generations of gravitational wave detectors.