Open Access
Add to BibSonomy
Abstract
High quality 940 nm AlxGa1-xAs n-type distributed Bragg reflectors (DBRs) were successfully monolithically grown on off-cut Ge (100) substrates. The Ge-DBRs have reflectivity spectra comparable to those grown on conventional bulk GaAs substrates and have smooth morphology and reasonable periodicity. These results strongly support VCSEL growth and fabrication on more scalable bulk Ge substrates for larger scale production of AlGaAs-based VCSELs.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Vertical cavity surface emitting lasers (VCSELs), compared to edge emitting lasers, have the advantages of high power efficiency, low beam divergence, narrow spectrum, higher spectral temperature stability, and low-cost testing and packaging [1]. The demands for VCSELs as near infrared illumination sources in three-dimension (3D) sensing and imaging have increased dramatically in recent years as VCSELs power some of the most popular new features in smartphones and augmented reality (AR) applications, such as Face ID, Animoji and proximity-sensing [2–3]. They are also widely used in auto-navigation, smart retail, industrial automation and robotics. The global market for VCSELs was predicted to increase to $2.4 billion by 2026 from the $1.2 billion market size in 2021 at a 13.6% compound annual growth rate [4–7]. So far, VCSEL production is still largely based on 3-, 4- and 6-inch bulk GaAs wafers [8]. Scaling up the substrate wafer size is a very effective way to boost the production volume, lower the manufacturing cost and make larger VCSEL arrays more ubiquitous.
The most common emission wavelengths of VCSELs are in the range of 750–1100 nm. A typical VCSEL epitaxial structure is about 5 to 15 microns thick, with the majority of the thickness taken by the bottom and the top distributed Bragg reflectors (DBRs) made of AlxGa1-xAs/AlyGa1-yAs superlattices. The lattice constant mismatch is about 0.14% between AlAs and bulk GaAs at room temperature, which is acceptable for VCSEL production based on 3-inch and 4-inch GaAs wafers [9]. When the growth is transited to larger GaAs wafers, the inherent strain formed during VCSEL growth results in bowed and warped wafers, which lead to low chip yield and reliability problems. One solution is to replace the bulk GaAs substrates with bulk Ge substrates, as the lattice constant of Ge locates between those of GaAs and AlAs. The less lattice mismatch between AlAs and Ge allows VCSELs epitaxy with lower residual strain, which reduces bow and warp after growth on Ge wafers [10]. With neglectable bow and warp, thicker DBR epilayers for longer wavelength VCSELs and larger diameter substrates can be used. Ge wafers also have some other attractive properties, including much lower dislocation densities, which are associated with yield loss and reliability failures in VCSELs on bulk GaAs substrates [11]. Commercially, most of the bulk Ge wafers produced are used as multijunction solar cell substrates. The price of mass-produced 6” Ge wafers is not higher than 6” GaAs wafers. Most importantly, Ge wafers can be made to 8- and 12-inch diameter, which is very hard to achieve for more brittle GaAs wafers. All these arguments make Ge wafers promising to replace GaAs wafers as the substrates for the mass production of VCSELs. In 2020, IQE first demonstrated 6” bulk Ge wafer based 940 nm VCSELs [12]. With unoptimized processes for Ge-VCSELs, they produced comparable DBR stopbands, active region photoluminescence (PL), lasing performance as GaAs-VCSELs and have much less wafer bow and slip [12]. It is the only report on bulk Ge-based VCSELs so far. However, no Ge substrate specifications, Ge to GaAs transition layers, DBR growth and testing conditions were revealed in their presentation.
In this paper, we report the successful monolithic integration of AlxGa1-xAs n-type distributed Bragg reflectors (n-DBRs) on bulk Ge substrates, which is the first and crucial step of bulk Ge-based VCSELs. Important material, processing and testing conditions are discussed in detail.
2. Experiment design and epitaxy growth
The schematic structure of the DBRs grown on Ge substrates is shown in Fig. 1. The Ge wafers used in this study were n-type 375 µm thick 4-inch (100) Ge wafers with 6-degree miscut towards (111) provided by Umicore. This thickness was chosen to match the 3” Si adaptor thickness so that the surfaces of the Ge samples are flush with the Si adaptor surface. Due to the more mechanical robustness, Ge wafers can be thinner than GaAs wafers of the same sizes. The miscut is to reduce antiphase-domains of the subsequent GaAs layers on Ge and to promote step-flow growth. 100 nm SiN thin films were deposited on the Ge wafer backsides to prevent Ge evaporation.
The AlGaAs n-DBR superlattices are not directly on top of the Ge surface. Instead, an InGaP nucleation layer, a 950 nm n-Ga0.985In0.015As layer lattice matched to Ge and a 50 nm n-GaAs layer with 5×1018 cm−3 doping were grown on Ge in order to have a clean and anti-phase free surface. The epitaxy growth of these transition layers was originally developed for solar cell applications. The growth was performed in an Aixtron 2800G4R metal organic chemical vapor deposition (MOCVD) reactor and the details are in [13]. After the backside SiN deposition and frontside GaAs/In0.015Ga0.985As/InGaP layer growth, the 4” Ge wafers were 376 µm in thickness. They were cut into 1 cm × 1 cm pieces. 3-inch 380 µm thick Si dummy wafers with 1 cm × 1 cm square holes cut by laser machining were used to position the Ge pieces in the 3” wafer pockets of a MOCVD reactor at LandMark Optoelectronics Corp for the n-DBR growth. This work was about our first epitaxy growth on Ge bulk substrates. An existing, mature DBR design with well-calibrated epitaxy growth recipes for GaAs substrates at LandMark was chosen to show the impact of Ge substrates on the DBR performance, layer thickness, uniformity, strain status and defect density.
The epitaxy growth at LandMark began with a 500 nm thick GaAs layer growth to have a fresh surface for the subsequent n-DBRs. A 3” 2-degree offcut bulk 629 µm thick GaAs wafer was used as the substrate of the control sample. Each DBR consists of 40 periods of n-type 3×1018 cm−3 doped four layers: 20 nm AlxGa1-xAs (x: 0.12→0.9) / 56 nm Al0.9Ga0.1As / 20 nm AlxGa1-xAs (x: 0.9→0.12) / 49 nm Al0.12Ga0.88As (from the bottom to the top in one period), which was designed with 940 nm as the center wavelength of the reflectance stopband. On top of the DBR, one 20 nm AlxGa1-xAs (x: 0.12→0.9), a 56 nm Al0.9Ga0.1As and a 5 nm undoped GaAs layer were grown sequentially as the capping layers. The growth temperature was set to 760 °C and the vapor pressure was 50 mbar. For the DBR growth process, the sources used for GaAs growth was H2, SiH4, AsH3 and Ga(CH3)3, which is the preferred metalorganic source of gallium for MOCVD. During the AlxGa1-xAs growth, Al2(CH3)6 was added to provide the Al source.
3. Characterization results and discussions
Good optical and material properties of the DBRs, i.e. reflectance spectra (with on-target stopband center wavelength, width and shape), smooth morphology and low threading dislocation density, are essential for the operation of VCSELs. To investigate the impact of the Ge substrates on the DBR quality, atomic force microscope (AFM) imaging, optical reflectance measurements, high-resolution X-ray diffraction (HRXRD), scanning electron microscopy (SEM), etch pit density (EPD), and electron channeling contrast imaging (ECCI) analysis were used.
3.1 Surface quality
Under the optical microscope and AFM, both GaAs-DBRs and Ge-DBRs show good surface quality, which are clean and free of cracks and antiphase domains. Some minor cross-hatch pattern can be seen in the AFM image of the Ge-DBR surface, as shown in Fig. 2, which is a result of some residual strain in the epitaxial layers due to the small lattice mismatch between AlGaAs, GaAs and Ge. The root mean square (RMS) roughness of the GaAs/InGaAs/InGaP/bulk-Ge/SiN substrates measured by AFM is 0.43 nm before the DBR growth. The RMS roughness of GaAs-DBRs and Ge-DBRs are 0.28 and 0.83 nm respectively measured on 50 µm x 50 µm areas.
3.2 Reflectance
Figure 3 shows the normal-incidence reflectance spectra of the Ge-DBR and GaAs-DBR, which were obtained with a Filmmetrix F20 thin-film analyzer. Reflectance measurements are sensitive to the position and tilt of the surfaces. A rotation stage was used underneath the samples to check the reflectance variation brought by the small surface tilt. Each DBR was measured with a series of rotation angles from 0 to 180 degrees. The highest reflectance was recorded, when the sample surface, the incident light and the reflection light detector satisfy the law of reflection. Ge-DBR has a stronger rotation angle dependence with 5.28% variation due to the rougher surface and larger non-uniformity, compared with 4.17% variation for bulk GaAs-DBR.
Fig. 3. (a) Normal-incidence reflectance spectra of three pieces of bulk Ge-based DBRs and two pieces of GaAs-based DBRs grown at the center of 3-inch Si dummy adaptor. All spectra were taken at the center of each sample normalized by the maximum GaAs DBR reflectance value. (b) Normal-incidence reflectance spectra of 6 pieces of Ge-based DBRs (a to f) placed at different positions of 3-inch Si dummy adaptor showing the cross-wafer growth non-uniformity, which are normalized by the maximum Ge-DBR reflectance of the spectrum from the center spot growth “a” (the blue line). (c) Normal-incidence reflectance spectra of bulk GaAs-based DBRs measured at 6 spots (a to f) across a 3-inch bulk GaAs wafer showing the cross-wafer reflectance uniformity, which are normalized by the maximum GaAs DBR reflectance of the spectrum from the center spot “a” (the blue line).
Figure 3(a) shows that Ge-DBRs have comparable stopband shapes, widths and maximum peak heights as those of the control GaAs-DBRs. The as-collected peak reflectance values of the Ge-DBRs and the GaAs-DBRs are in the range of 113.31 to 115.45 and 111.03 to 114.94 respectively. These reflectance values are normalized to a Si calibration sample of Filmmetrix. The peak reflectance values of the Ge-DBRs are from 98.58% to 100.44% of the peak GaAs-DBR reflectance. The errors mainly come from the non-uniformity of the DBR growth and the measurement method.
The stopband widths of the Ge-DBRs and the bulk GaAs-DBRs are 74.4 nm and 77.3 nm respectively. This difference is within the cross-wafer deviations of the Ge-DBRs shown in Fig. 3(b). There is a noticeable spectral shift between the Ge-based and GaAs-based samples. The stopband centers of GaAs-DBRs locate at 940 nm, while those of the Ge-DBRs range from 920 to 940 nm. SEM microscope images taken at different locations reveal that the average Ge-DBRs and GaAs-DBRs are 5848 nm and 5924 nm respectively (see section 3.4), which means that the Ge-DBR layers are about 1.45% thinner than those of the GaAs-DBRs. This is consistent with the smaller average center wavelength of the Ge-DBR stopband. Although the 629 µm thick GaAs wafer with a two-degree offcut and the 376 µm thick Ge pieces with a six-degree offcut were processed in the same DBR epitaxy growth, the Ge pieces were fixed by Si dummy adaptors, while the GaAs substrate stayed in the wafer pocket as a full 3” wafer. The difference in substrate offcut angle, heat conduction and gas flow could result in different growth rate and uniformity of the Ge-DBRs compared with the GaAs-DBRs.
3.3 X-ray diffraction
In order to check the quality of the epitaxial growth of Ge-DBRs, HRXRD ω/2θ rocking curves were measured for both Ge-DBRs and control GaAs-DBRs. Ge (220) 4-bounce monochromator for incident beams and a 1 mm slit for the detector were used to characterize both structures.
In Fig. 4, the (004) ω/2θ diffraction curves of the control GaAs-DBR and Ge-DBR are plotted in comparison with the simulated GaAs-DBRs curve. The simulation was conducted by X’Pert Epitaxy 4.1 software. The best-fitting simulated GaAs-DBR rocking curve in Fig. 4 was based on a uniform DBRs structure with a 146.5 nm periodicity and a 5.94 µm total thickness with capping layers, which is consistent with the SEM observations.
Fig. 4. (004) omega-2theta scan of Ge-based DBRs, bulk GaAs-DBRs and best-fitting simulated GaAs-DBRs curve for comparison
In Fig. 4, a very good agreement on the peak positions and intensity between the simulated and measured GaAs-DBR can be observed, indicating very good DBRs thickness uniformity and crystal quality. The slight shift of GaAs peak from 33.0239 degree in simulation to 33.0287 degree in measurement is from the slight tensile strain in the grown buffer GaAs layer beneath the DBRs, which is not included in the simulation. In comparison, the measured peaks from the Ge-DBR are slightly broader leading to minima that are less defined. The satellite-peaks distribute still at the same positions, which means that the material compositions are as designed and a good epitaxy growth quality is achieved. The DBR periodicity of the Ge-based DBRs is 144.3 nm and the total thickness is 5.85 µm with capping layers calculated from the satellite peaks fitting. The Ge-based DBRs’ thickness and the difference between Ge-based DBRs and GaAs-based DBRs are consistent with the SEM observations. The broader peaks of Ge-based DBRs, however, indicate that there is more variation in the layer thickness, surface roughness due to the strain relaxation (as seen in the AFM images in Fig. 2), and a higher density of crystal defects like threading dislocations, which were confirmed by SEM imaging and dislocation density measurement discussed below.
3.4 Cross-sectional SEM images
SEM imaging of the cross-sections of the Ge-DBRs and GaAs-DBRs was conducted to observe the DBR layers. The SEM samples were manually cleaved with a diamond scriber. No surface polishing or coating was conducted. The images in Fig. 5 and 6 were collected using a FEI Nova NanoSEM system. The measurement mode was the immersion mode with a gaseous analytical detector (GAD). The operation voltages used were 10 to 15 kV and the working distance was 5–6 mm. SEM images of the Ge- and GaAs-DBR surfaces (not shown) reveal smooth surfaces without any cracks or surface defects within the imaging area of about 7.5 × 105 µm2. The SEM images of the cross-sections of bulk GaAs- and Ge-DBRs in Fig. 5(a) and Fig. 5(b) show good periodicity and uniformity for both types of DBRs. The shadowy lines in Fig. 5(b) are due to the roughness from the sample cleaving. A noticeable difference between the two images is on the thicknesses of the DBRs. The thickness of the GaAs-DBR is 5.93 µm, while that of the Ge-DBRs is 5.84 µm. Both are thickness readings from the SEM images. These are consistent with the 5.94 and 5.85 µm thickness values from the best-fitting rocking curves in Fig. 4. The Ge-DBR is 1.5% thinner, consistent with the shorter center wavelengths of the Ge-DBRs. The thicknesses of every two repeated pair of a GaAs-DBR and a Ge-DBR are shown in Fig. 6(a) and Fig. 6(b), and that of the Ge-DBR is 1 to 5% thinner than that of the GaAs-DBR. This thickness difference was not unexpected, as the DBR growth conditions used were optimized for GaAs-DBRs to obtain optimal yield, uniformity and VCSEL performance at LandMark. Ge substrates are 376 µm thick in comparison with 629 µm thickness of the GaAs substrates, which may have led to small variations in thermal coupling and finally slightly different surface temperatures. Also, Ge samples are much smaller (1 cm by 1 cm), and were placed inside the Si adaptors. Due to the edge effects, the non-uniformity of the smaller Ge-DBRs is much larger than that of the full GaAs-DBRs. To avoid these problems, in the next epitaxy growth, 3” full Ge wafers will be used. Moreover, the MOCVD growing conditions will be fine-tuned for Ge-DBRs to further improve the thickness, uniformity and stopband characteristics.
Fig. 5. Cross-section SEM images at 30 K × magnifications of a (a) bulk GaAs-based DBR, (b) Ge-based DBR.
Fig. 6. Epitaxial DBR layer thickness distribution of (a) bulk GaAs-based DBR, (b) Ge-based DBR. The yellow numbers are the thicknesses of two DBR periods measured from the SEM images.
3.5 Dislocation density measurements
Threading dislocations (TDs) are key defects that degrade laser efficiency and lifetime. EPD and ECCI measurements were conducted on the Ge substrates before and after the DBR growth respectively to measure the threading dislocation density (TDD). The two methods were chosen according to their suitable TDD ranges and the non-destructive nature of ECCI. For the EPD method, the 500 nm GaAs layer on the top of Ge substrates was stripped by nanostrip and H2O2 for 30 mins. The EPD etchant solution consisted of CH3COOH (100 ml), 70% by weight HNO3 (40 ml), 49% by weight HF (10 ml), and I2 (30 mg). Figure 7(a) is an optical image of the Ge substrates after EPD etching. The area of this image is 4.443 mm2 and the number of threading dislocation is 40. The GaAs/InGaAs/InGaP/bulk-Ge/SiN substrate (post the transition layer growth and before the DBR growth) has a TDD of about 9.6 × 102/cm2 obtained over a total area of 13.3 mm2 by EPD.
Fig. 7. (a) An optical image of the etched Ge substrates before growth and (b) an ECCI image taken on Ge-based DBRs after growth showing one dislocation (the white spot) in the image.
The ECCI measurement was performed with a SEM (Apero 2 model by ThermoFisher Scientific) using the PivotBeam mode. The crystal orientation was optimized using the sample stage tilt and rotation, and crystal defects imaging was conducted in the Immersion mode. All ECCI imaging was performed in high vacuum and no conductive coating was applied. Figure 7(b) is one of the ECCI images of the Ge-based DBRs after growth to calculate the TDD. There is only one dislocation spotted within the area of 48.2 µm2. The Ge-DBR has a TDD of about 8.3 × 105/cm2 obtained over a total area of 482 µm2 measured by ECCI. As the dislocation density is low for ECCI imaging, in most of the surveyed areas, there are no dislocations observed. Therefore, this 8.3 × 105/cm2 value is believed to be an overestimation of the TDD of the Ge-DBR.
3.6 Future work
As discussed, our work was based on GaAs/InGaAs/InGaP/bulk-Ge/SiN substrates that were readily available. This type of substrates was originally developed for solar cells, and thus solar pumped VCSEL lasers can be envisioned with these substrates. In principle, other epitaxial GaAs growth technologies on bulk Ge substrates may also be effective to serve as the virtual substrates of the DBRs. Due to the constraints on the resources, other types of Ge to GaAs transition layers were not investigated, which is a good topic for future studies. Our next growths of partial or full VCSEL epitaxy on Ge wafers have been planned. Calibration runs will be performed to tune the Ge-DBRs growth recipe to have the stopbands centered at 940 nm. Full Ge wafers are going to be used to avoid growth condition variation and improve the growth uniformity and quality. More material analysis and device performance measurements will be conducted.
4. Conclusion
High quality AlGaAs DBRs were successfully monolithically grown on bulk Ge substrates. The Ge-DBRs have reflectivity spectra comparable to those grown on conventional bulk GaAs substrates and have smooth morphology and reasonable periodicity. No APDs are formed in DBR epitaxy on Ge substrates, and the threading dislocation density of Ge-DBR is about 8.3 × 105/cm2 or less. These results are relevant to the R&D of VCSEL growth and fabrication on large-area Ge substrates to address the dramatically increased demand of VCSELs.
Funding
Huawei Technologies Canada.
Acknowledgments
Umicore N. V., Belgium, is acknowledged for providing the bulk Ge wafers with GaAs/GaPAs epitaxy layers and SiN back coating. ThermoFisher Scientific corporation was acknowledged for the ECCI measurements.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. K. Iga, “Surface-emitting laser-its birth and generation of new optoelectronics field,” IEEE J. Sel. Top. Quantum Electron. 6(6), 1201–1215 (2000). [CrossRef]
2. A. Extance, “Faces light up over VCSEL prospects,” SPIE Professional magazine, 09 April (2018).
3. M. Dummer, K. Johnson, S. Rothwell, K. Tatah, and M. Hibbs-Brenner, “The role of VCSELs in 3D sensing and LiDAR,” Proc. SPIE 11692, 116920C (2021). [CrossRef]
4. N. Anyadike, “3D sensing technology booming on new applications,” Electropages, 30 January (2019).
5. J. F. Seruin, D. Zhou, G. Xu, A. Miglo, D. Li, T. Chen, B. Guo, and C. Ghosh, “High-efficiency VCSEL arrays for illumination and sensing in consumer applications,” Proc. SPIE 9766, 97660D (2016). [CrossRef]
6. H. Moench, M. Carpaij, P. Gerlach, S. Gronenborn, R. Gudde, J. Hellmig, J. Kolb, and A. van der Lee, “VCSEL-based sensors for distance and velocity,” Proc. SPIE 9766, 97660A (2016). [CrossRef]
7. Yole Dévelopment, “VCSEL – Technology and Market Trends 2021 report,” Technical report, 2021.
8. D. Wiedenmann, M. Grabherr, R. Jäger, and R. King, “High volume production of single-mode VCSELs,” Proc. SPIE 6132, 613202 (2006). [CrossRef]
9. C. Bocchi, C. Ferrari, P. Franzosi, A. Bosacchi, and S. Franchi, “Accurate determination of lattice mismatch in the epitaxial AlAs/GaAs system by high-resolution X-ray diffraction,” J. Cryst. Growth 132(3-4), 427–434 (1993). [CrossRef]
10. M. Bosi and G. Attolini, “Germanium: Epitaxy and its applications,” Prog. Cryst. Growth Charact. Mater. 56(3-4), 146–174 (2010). [CrossRef]
11. B. Depuydt, A. Theuwis, and I. Romandic, “Germanium: from the first application of Czochralski crystal growth to large diameter dislocation-free wafers,” Mater. Sci. Semicond. Process. 9(4-5), 437–443 (2006). [CrossRef]
12. A. Johnson, A. Joel, A. Clark, D. Pearce, M. Geen, W. Wang, R. Pelzel, and S. W. Lim, “High performance 940 nm VCSELs on large area germanium substrates: the ideal substrate for volume manufacture,” Proc. SPIE 11704, 1170404 (2021). [CrossRef]
13. M. Feifel, D. Lackner, J. Ohlmann, J. Benick, M. Hermle, and F. Dimroth, “Direct growth of a GaInP/GaAs/Si triple-junction solar cell with 22.3% AM1.5 g efficiency,” Sol. RRL 3(12), 1900313 (2019). [CrossRef]
