Variation in thermal stability of Ge_1−xSn_x films for infrared device applications

Development of group-IV alloys targeting a range of infrared (IR) photonic applications has been ongoing for decades. A ternary Si_xGe_1−x−ySn_y film can theoretically be tuned through a wide range of bandgaps and lattice constants,¹ and with sufficiently high Sn concentration, an indirect-to-direct bandgap transition is realized.² Photodetectors responsive at 2 μm and longer wavelengths are desirable for fiber optic communications and thermal imaging. IR detectors responsive out to 3.3 μm have been achieved.³ Because of its compatibility with existing Si device processing methods, SiGeSn would be competitive with II-VI and III-V semiconductor materials in the same bandgap regime.⁴ Direct integration of photonic materials into Si-based CMOS technologies would dramatically reduce power demands and size for intracomputer communication and enable the development of new smart-pixel, multimodal focal plane arrays, and remote hyperspectral imaging technologies.⁵ The alloy system also has potential as a thermophotovoltaic material with lower costs and material toxicity than the dominant III-V technologies.^6–9 

Film quality degradation in response to elevated temperatures is a significant challenge for SiGeSn. It has to be grown epitaxially at very low temperatures relative to ideal conditions for Si or Ge due to the low solubility of Sn in each – 0.1% and 1%, respectively.⁵ Once grown, SiGeSn layers may still respond to the application of heat by strain relaxation and Sn segregation.^10–18 Consequently, the temperature has to be carefully considered for every successive step of device manufacturing, including subsequent layer growths, dielectric deposition, and annealing for material quality and contact formation. These considerations cumulatively are called the thermal budget. The operating temperature of devices also needs to be matched to the tolerance of the material.

Previous studies have considered various aspects of annealing and material degradation in Ge_1−xSn_x. Both rapid and long anneals have been found to cause relaxation in some films.^12,17,18 The intensity of photoluminescence can either increase or decrease in response, depending on the initial state of strain in the film and the conditions of the anneal;^11,16 generally, a partially relaxed film is found to have increased PL intensity, while a fully strained film is likely to have decreased PL intensity. Sn segregates in many annealed samples, with the temperature at which segregation initiates being correlated to Sn concentration, strain state, and layer thickness.^16,17 Samples grown by molecular beam epitaxy (MBE) versus by chemical vapor deposition (CVD) sometimes respond differently,¹⁴ presumably because their defect profiles are different¹⁹ as a consequence of the very different temperature profiles in growth.

In this work, we use short anneals with fast ramping rates to determine the maximum temperature withstood by Ge_1−xSn_x films before their degradation, as denoted by Sn surface segregation. Fast anneals were chosen to help indicate thermal processing conditions which rapidly exceed a film's thermal budget. A range of alloy compositions were investigated to help establish a relevant trend with composition and to provide context for a wide variety of potential Ge_1−xSn_x applications.

Samples in this study were grown on a custom-designed VG-90 MBE system, with Ge evaporated by an electron beam and Sn evaporated by a solid source effusion cell. Before use, undoped (001) Ge wafers were cleaned for 1 min in 1% aqueous HF, then quickly loaded into the vacuum environment. As the first step of growth, wafers were heated to 775 °C to remove any reformed oxide, then cooled to 375 °C, and 250 nm buffers of Ge were grown to smooth the surface. Wafers were then cooled to low temperatures for the growth of 100 nm of Ge_1−xSn_x. Substrate temperatures were monitored by a thermocouple located behind the substrate and at temperatures above approximately 350 °C by an optical pyrometer. Growth conditions are listed in Table I.

Fast anneals were carried out under a constant flow of 4.5 l/min N₂ or He gas on a Linkam HFS600 heating stage with a purged volume of approximately 0.1 l. Separate pieces, approximately 1 cm on a side were cleaved from near the center of the as-grown wafer for each annealing temperature. Sample temperatures were ramped at 100 °C/min and then held for 1 min at the target temperatures, then cooled at 100 °C/min. Reported temperatures were corrected for the gradient from the heater to the sample’s front side based on measurements taken with a thermocouple bonded to the front side of a Ge reference wafer during heating. The front side temperature was 90.6% of the back side temperature. Assuming a linear gradient through the sample, the highest annealing temperature represents a 0.01 °C drop across the thickness of the GeSn film. Temperatures exceeded the setpoint by no more than 0.25 °C.

After removing samples from the heating stage, their surfaces were examined by differential interference contrast (DIC) optical microscopy before and after annealing, to look for Sn droplet formation and surface degradation. The film Sn contents and crystallinity were measured by high-resolution x-ray diffraction (HRXRD) on a Bruker D8 HR diffractometer, with a Cu K_α1 source. The beam limiters were 1 mm by 0.6 mm. Samples were measured by 004 and 224 + 2θ-ω line scans, with reciprocal space maps (RSMs) performed where substantial relaxation was indicated. HRXRD measurements were repeated on annealed samples to examine the change in crystallinity and strain states. Sn content was determined by modeling 004 2θ-ω line scans in Bruker LEPTOS software, assuming Vegard's law relationship between the out-of-plane lattice parameter and Sn composition for samples shown to be pseudomorphic by RSM. Crystallinity and optical properties before and after annealing were also probed by Raman spectroscopy on a Horiba LabRam Evolution equipped with a 532 nm laser, 1800 gr/mm grating, 50× objective, and a spectral resolution ≤ 0.6 cm⁻¹. The resulting spot size was 1.2 μm, and the expected penetration depth of 532 nm light in Ge was approximately 17 nm.

As expected, higher Sn compositions and the resulting higher film strains led to Sn segregation occurring at lower annealing temperatures, as shown in Fig. 1. The temperature threshold of segregation was determined by the first appearance of Sn droplets in DIC microscopy, such as the bright dots visible at the end of the segregation trails in Fig. 2(d). The segregation threshold temperatures determined in this work were dependent in part on the film thickness, which was uniform between samples rather than proportional to the critical thickness for strain relaxation. Previous works^14,17 have found that occurrence of strain relaxation before Sn segregation forms is dependent on both the initial strain condition and the closeness of pseudomorphic films to their critical thickness. According to the People and Bean model, the expected critical thickness of P6 (6.5% Sn) is approximately 235 nm.^20,21 P6, due to its lower Sn content, was further from its critical thickness than P8 (8.4%, h_c ≈ 150 nm), for example, and may, therefore, be expected to persist unaffected to higher annealing temperatures than a 6.5% film of 200 nm thickness.

In the pseudomorphic films, Sn segregated to the surface in clusters of moderate to large droplets, which then traveled preferentially along $⟨110⟩$ lattice directions, as shown in Fig. 2 and as seen in previous studies.²² In P6, P8, and P10, there were slight increases in surface roughness without segregation after some annealing, but Sn droplets that emerged immediately began to migrate. This can be seen in Figs. 2(a)–2(d), which shows the surface of P6 sample pieces as-grown, annealed but unsegregated, and two cases following annealing with Sn surface segregation. In the case of sample R14, which began to strain and relax during growth, clusters of much finer Sn droplets initially formed as rough patches on the surface, as seen in Figs. 3(a)–3(c). Raising the annealing temperature by 25 °C resulted in larger clusters of comparably fine droplets. These droplets did not begin to migrate until the 321 °C anneal, at which time they moved outward in arbitrary radial directions from assorted regions that also contained small stationary Sn droplets. This is consistent with previous works, which have proposed that Sn segregates to the surface along threading dislocations when they are present.^14,15 R14, which has partly relaxed, will have a higher density of threading defects along which Sn can rise to the surface. The preferential surface migration along $⟨110⟩$ lattice directions seen for samples P6 (6.5%) and P8 (8.4%) was missing from sample R14 and less clear in P10 (9.7%) Sn, shown in Fig. 3(d). We attribute this change in behavior to the increased density of threading defects induced by the higher strain and onset of relaxation in higher Sn-content films.

Defect formation and changes to strain were observed by HRXRD in some annealed samples. 2θ-ω (004) symmetric scans of each sample, seen in Fig. 4, show two or more peaks: the Ge substrate peak at 66.06°, and a peak (or two peaks for the partially relaxed R14 sample) at a lower angle, resulting from the increased out-of-plane lattice constant of the strained GeSn layer. In highly uniform crystalline films, Pendellösung interference fringes form around the GeSn peak. The sampling area of HRXRD was much larger than the feature sizes seen in optical microscopy, so data may be assumed to sample a representative portion of the bulk material.

The lowest Sn content sample in this study, P6, had interference fringes as grown. After annealing at 367 °C, despite there being no evidence of surface segregation in optical microscopy, the fringes were substantially less well-defined and the overall sample signal intensity was reduced, with an asymmetric shift toward the substrate peak. Two plausible causes are the generation of dislocations, or interdiffusion at the GeSn/Ge boundary resulting in a layer of lower-Sn alloy.^14,18 In light of previous findings in the literature¹⁴ and the slightly increased surface roughness seen in Fig. 2(b) compared to Fig. 2(a), dislocation formation is more likely. Surface segregation may have been suppressed relative to the onset of dislocations in P6 by its initial low density of dislocations along which Sn could diffuse to the surface. This would explain why Sn surface segregation trails seen in Fig. 2 radiate from fewer points of origin after annealing at 412 °C than at 457 °C; increasing annealing temperatures enabled a larger number of dislocations to form.

The 004 line scan of P8 after a segregated anneal, 321 °C, showed no difference relative to its as-grown scan. This, in contrast to P6, is consistent with previous works that have found that fully strained films that have not been grown past their theoretical critical thickness decompose before strain relaxing.^11,17 It also shows that P8, which as indicated by HRXRD, had relatively few defects as-grown, also had little room for defects to be repaired by annealing. The unresponsiveness of the bulk to the surface segregation of Sn in the annealed sample of P8 indicates that Sn rejection was strongly localized to the sites from which droplets emerged and spread, rather than being uniform.

The 2θ-ω line scan of P10 as grown had clear but slightly asymmetric Pendellösung fringes, which became more symmetric after both unsegregated and segregated anneals. P10 had a small amount of strain or composition nonuniformity, which annealing improved. Unlike many previous works, the gradual onset of Sn surface segregation in R14 was not shown by HRXRD to be associated with increased relaxation of the film.^13–16 This was determined from RSMs taken of R14 before and after a 253 °C anneal, shown in Fig. 5. Each map has two peaks, a narrow Ge substrate peak at (q_x 5.01, q_z 7.07) and a GeSn peak with its greatest intensity at the fully strained position (q_x 5.01, q_z 6.83) with a tail representing strain relaxation. The relative distribution of strained material shown in RSMs was unchanged by annealing. We speculate that this is due to the intersection of our short anneal time and low temperatures relative to previous studies that have found progressive relaxation using temperatures of more than 500 °C or annealing periods on the scale of hours.^11,16

Raman spectroscopy supports the hypothesis that annealing improved the uniformity of interatomic spacing, which results from consistent alloy composition and strain state, only in samples that already showed early signs of relaxation or deteriorated crystallinity. Plots of each sample as-grown are shown normalized to a Ge reference in Fig. 6(a). The Ge reference results in a single symmetric Raman shift peak at 299.2 cm⁻¹, while GeSn films have a Ge-Ge peak at a slightly reduced shift overlapped by a Ge-Sn peak that appears as a tail toward smaller shifts. The intensity and full width at half-maximum (FWHM) of the Ge-Ge peaks only marginally changed in samples P6 and P8, following anneals that did not cause segregation and which showed little to no relaxation in HRXRD. The spot size for Raman spectroscopy was 1.2 μm, which is comparable to the segregated Sn feature sizes and much smaller than the spacing between features in the first segregated P6 sample. The minimal evidence of the effect on P6, which appeared to have deteriorated in HRXRD, may indicate that the data collected came from regions that did not happen to include newly formed defects.

While the fully pseudomorphic samples produce comparable as-grown and annealed Raman data, the intensity increased notably for annealed samples of P10 and R14, including for annealing which begin to induce Sn segregation. For P10 (9.7% Sn), the 276 °C anneal, which caused no surface segregation, the Raman peak intensity doubled while reducing the FWHM by 9%. For R14 (14.4% Sn), the 253 °C anneal, which caused very slight segregation, increased the intensity by 90.6%, although the FWHM increased by 3.9%. A stronger Raman response indicates a greater regularity in strain or atomic distribution after annealing.

Raman data were taken for P10 for two anneals that did not induce segregation and one that did, to examine the efficacy of anneals on the sample nearest its critical thickness for relaxation in growth. The 276 °C anneal, which was the highest temperature P10 withstood without Sn segregation, created the greatest increase in Raman intensity relative to the sample as grown, while a lower temperature anneal at 231 °C produced a 64% increase in intensity. Intensity also increased slightly for the 321 °C anneal, which did induce Sn segregation.

The Ge-Ge optical longitudinal peak is separated from a Ge-Sn peak by fitting two Gaussian functions to the measured data.¹⁸ A representative example, P10 after a 276 °C anneal, is pictured in Fig. 6(d).²³ The Ge-Ge peak shifted to smaller wavenumbers between samples as the increasing Sn content increases the average interatomic distances. The most dramatic shift was observed for R14, which was partially relaxed and, therefore, had the most dramatic change of Ge-Ge positioning. The Ge-Sn peak position shifted to smaller wavenumbers between P6 and P8, consistent with the increased Sn content, but increased again at P10, where the first traces of relaxation were seen. The peak positions responded weakly to annealing until segregation was visible, for 321 °C in P10 and 253 °C in R14. In these samples, both peak centers shifted to smaller wavenumbers. This, as with the original Ge-Ge peak position of R14, is attributed to the restructuring of dislocations forming the original relaxation. A summary of peak features is listed in Table II.

In this study, we investigated the effects of annealing on 100 nm GeSn films of compositions between 6.5% and 14.4% Sn grown by MBE on Ge wafers. Material quality was measured after rapid thermal anneals, with increasing Sn content leading to reduced maximum temperatures before Sn droplets segregate to the sample surface. Samples that were fully strained to the Ge substrate after growth were found to begin Sn segregation and other forms of visible surface degradation without substantial improvement in bulk film properties. Sn droplets formed on the surface of pseudomorphic samples migrated along $⟨110⟩$ directions. Samples that, during or after growth, passed the critical thickness for relaxation were found to improve in uniformity in response to anneals leading up to and even surpassing the Sn segregation threshold. The migration of Sn droplets on the surface of partially relaxed samples was less strictly aligned to $⟨110⟩$ directions. For these short anneals, little or no relaxation was observed in any sample. We attribute this to the high energy threshold of beginning relaxation in pseudomorphic films and the low temperatures used on the relaxed films, which for a short anneal is a small cumulative energy transfer.

Funding for this research was provided by the U.S. Air Force Office of Scientific Research (AFOSR) grant (No. FA9550-20-1-0188) and DURIP grants (Nos. FA9550-15-1-0352 and N00014-17-1-2591) from the AFOSR and Office of Naval Research, respectively. We gratefully acknowledge Thachachanok Menasuta for assistance with Raman data collection. Raman spectroscopy was performed at the Center for Nanoscale Systems (CNS) at Harvard University, a member of the National Nanotechnology Coordinated Infrastructure Network (NNCI), which is supported by the National Science Foundation (NSF) under NSF Award No. ECCS-2025158. This work made use of the Materials Research Laboratory Shared Experimental Facilities at MIT for x-ray diffraction.

The authors have no conflicts to disclose.

Amanda N. Lemire: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Kevin A. Grossklaus: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (supporting); Methodology (equal); Supervision (lead); Visualization (equal); Writing – review & editing (equal). Thomas E. Vandervelde: Conceptualization (equal); Funding acquisition (equal); Methodology (supporting); Resources (equal); Supervision (supporting); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.