Improved Thin Film Quality and Photoluminescence of N-Doped Epitaxial Germanium-on-Silicon using MOCVD

Ge-on-Si structures in-situ doped with phosphorus or arsenic via metal organic chemical vapor deposition (MOCVD) were investigated. Surface roughness, strain, threading dislocation desnity, Si-Ge interdiffusion, dopant diffusion, and photoluminescence were characterized to study the impacts of defect annealing and Si substrate offcut effects on the Ge film quality and most importantly, the light emission properties. All samples have a smooth surface (roughness<1.5 nm), and the Ge films have a small tensile strain of 0.2%. As-grown P and As-doped Ge films have threading dislocaiton densities from 2.8e8 to 1.1e9 cm^(-2) without defect annealing. With thermal cycling, these values reduced to 1-1.5e8 cm^(-2). The six degree offcut of the Si substrate was shown to have little impact. In contrast to delta doping, the out-diffusion of dopants has been successfully suppressed to retain the doping concentration upon defect annealing. However, the photoluminescence intensity decreases mostly due to Si-Ge interdiffusion, which also causes a blue-shift in the emission wavelength. Compared to a benckmarking sample from the first Ge laser work doped by delta doping method in 2012, the as-grown P or As-doped Ge films have similar photoluminescence intensity at a 25% doping concentration and smoother surface, which are promising for Ge lasers with better light emission efficiencies.

For Ge-on-Si lasers, important material requirements commonly include 1) very high ntype doping above 1×10 19 cm -3 , 2) tensile strain > 0.2% , 3) low threading dislocation density (TDD), 4) surface roughness in nm scale or lower, and 5) minimized Si-Ge interdiffusion at Ge/Si interfaces to prevent detrimental indirect gap behavior. These requirements are also closely related and some can be traded-off such as the doping and strain requirements. Adding high concentrations of n-type dopants in Ge is crucial to occupy the electron energy states in the indirect conduction valleys (L valleys) [15]. Many efforts have been made on that. Delta-doped layer and gas immersion laser doping were used to achieve up to 5×10 19 cm −3 activation of P doping [18][19][20].
The spin-on dopant process and multiple implantation were also successful in doping Ge up to 1×10 20 cm −3 [22][23]. Laser thermal annealing has been successfully employed to achieve a doping activation above 10 20 cm -3 in implanted bulk Ge [24] and in-situ doped Ge-on-Si epilayers [25].
For doping from external sources such as the dopant diffusion from delta-doping method, a high temperature annealing step is commonly used to drive dopants in and activate them. Pellegrini et al. achieved an activated carrier concentration ranging from 2.5×10 19 cm −3 to 2.1×10 20 cm −3 by incorporating P dopants during the Ge growth followed by annealing [26]. For in-situ doped Ge, defect annealing such as thermal cycling of high and low temperature (HT/LT) is commonly used to reduce threading dislocation density (TDD) right after the Ge film epitaxy step. The high thermal budget associated with these steps is not desired, as it drives Si-Ge interdiffusion as well as dopant out-diffusion [27], and thus counteracts the efforts in bandgap engineering. Especially, interdiffusion is enhanced with n-type doping by a factor of 2 to 6 with P doping and 1.5 to 3 times with As doping in low to mid-10 19 cm -3 range [28][29][30].
Lee et al. studied the impact of high concentration arsenic (As) on Ge epitaxial film grown on Si (001) with 6° off-cut. The observation was that the TDD was reduced by at least one order of magnitude to < 5 × 10 6 /cm 2 , which was attributed to the enhancement in the velocity of the dislocation motion in an As-doped Ge film [3]. However, Ge films without HT/LT annealing were not studied. Recently, Zhou et al. systematically studied the doping and defect annealing impact on Ge film quality for P and As doped Ge on Si (001) with 6 o off-cut. The finding was that P and As doping can reduce etch pit density (EPD) without the HT/LT defect annealing step. This suggests that n-type doped Ge films without defect annealing have lower defect density and minimized Si-Ge interdiffusion, which are very promising for Ge light emission. These n-doped Ge films also have smooth morphology and tensile strain of about 0.16 to 0.2%. However, photoluminescence (PL) measurements of the n-doped Ge were not tested, and Ge grown on onaxis Si (100) wafers were not studied either.
This work is a follow-up and more in-depth study of Zhou et al.'s recent work [28]. 1) This work provides photoluminescence (PL) measurements and analysis of n-Ge films on (100) Si substrates without HT/LT annealing, which are more relevant to light emission and to the mainstream Si wafer type, and 2) Zhou et al.'s work used EPD to characterize TDD, which was effective for undoped Ge, but was not suitable for n-Ge [32]. In this work, electron channeling contrast imaging (ECCI) was used to measure TDD of n-Ge [33][34] to provide much more accurate TDD measurements.

EXPERIMENT DESIGN, RESULTS AND DISCUSSIONS
Structure Design, Growth and Defect Annealing. Figure 1 Schematic diagrams of the 7 samples grown by MOCVD. X stands for As or P. Table 1. Samples without defect annealing were noted as "NA" in the names. "5TC" means with five thermal cycling for defect annealing. "MHT" means the defect annealing was performed only with the merged high temperature steps. "off" means on 6° off-cut (100) Si, and "on" means on on-axis (100) Si. Seven n-doped Ge-on-Si samples with 3 different annealing conditions (no annealing, 5 HT/LT thermal cycles, and merged HT annealing) were investigated ( Table 1). The schematic structure is shown in Fig. 1. The sample matrix was designed to study the impact of the defect annealing and the 6 degree offcut. There were more samples without annealing, as they were shown to have higher PL than the annealed ones. The merged HT annealing was shown to be equivalent to the 5 HT/LT annealing in our previous study [28].The "MIT Ge control" sample was from the first electrically pump Ge laser work published in 2012 [16] to serve as a benchmark.
All 7 samples were grown on 8-inches Czochrolski (CZ) Si wafers in a metal-organic chemical vapor deposition (MOCVD) tool and the model is CRIUS CCS from Aixtron. The Si substrates are either on-axis (100) Si or (100) Si with 6° off-cut towards the [110] direction. P and As doping levels were chosen as the highest concentration achievable in the epitaxial growth tool.
The "MIT Ge control" sample was grown by an ultra-high vacuum CVD tool (UHVCVD) at MIT.
The CVD tool model is Sirius 300 from Unaxis. The doping of the "MIT control sample" was conducted with a delta doping technology, where multiple atomically thin P layer were deposited on top of Ge and was later driven in with a high temperature annealing step [16].
For the 7 samples grown by MOCVD, before a Si layer was deposited, the Si substrate was treated at 1050 ± 10 °C for 10 minutes under H2 ambient at 400 mbar. Then, a 600 nm Si layer was grown at 950 ± 10 °C under H2 ambient at 100 mbar. To improve the Ge film quality and reduce the threading dislocations caused by the Ge-Si lattice difference, a 100 nm Ge seeding layer was grown at 400 ± 10 o C under H2 ambient at 100 mbar (low-temperature Ge growth) on top of the Si layer. Finally, a Ge film about 600 nm was grown at 650 ± 10 °C under H2 ambient at 100 mbar (high-temperature Ge growth). We denote these layers as "the top Ge layers" in the discussion below to differentiate from the Ge seeding layers. Immediately after the growth procedure, some of the samples were annealed inside the growth tool while another half were left unannealed (NA) for comparison. Post-deposition thermal cycling was performed by repeating a H2 annealing cycle between low temperature (LT) and high temperature (HT) ranging from 600 °C to 850 °C for 5 times (5TC). Each annealing step was 10 minutes at HT and 5 minutes at LT to improve the quality of the Ge epitaxial film. The ramping rates for heating and cooling were around    Threading Dislocation Density Characterization. EPD has been widely used to characterize TDD of undoped Ge [32,39,40]. However, for doped Ge, the etch behaviors are quite different. The etch recipes that work for undoped Ge may not work for doped Ge, and the etch rate is also highly dependent on the doping level [32]. ECCI has been proven to be a reliable, fast and non-destructive method for TDD characterizations [33][34]. The stress fields associated with threading dislocations cause a local deformation of the crystal lattice, which can be imaged using ECCI. TDD measured by ECCI for undoped Ge were in good agreement with EPD and transimisson electron microscopy (TEM) results [34]. Therefore, we believe that ECCI can overcome the weakness of EPD methods for doped Ge, and provide more accurate TDD measurements.
It can be seen that the n-Ge films have TDD values above 1 × 10 8 cm -2 , and are commonly three orders of magnitude larger than the TDD by EPD. Defect annealing helped to reduce TDD Pdoped Ge.  As seen in Fig. 4 (a), the interdiffusion is minimal for the unannealed samples. The 6 degree offcut has little impact on the Ge and dopant profiles. Some differences can be seen in the concentration of the P and As segregation peaks concentrations at Ge/Si interfaces, which may be due to SIMS uncertainty at the interfaces. The SIMS data of annealed n-Ge/offcut-Si samples are in Fig. 4 (b), where significant interdiffusion can be seen. According to the diffusion theory, diffusion and thus interdiffusion is isotropic for cubic crystals. Therefore, we expect that the Ge profiles of n-Ge/on-axis-Si to be the same as those of Ge/offcut-Si. Due to the interdiffusion, the sharp Ge/Si interfaces changed to thick alloy regions. In the full Ge x range, the effective interdiffusivity of Sample P (̃) is 1.5 to 3 times higher than that of sample A (̃), and ̃ is 1.5 to 2 times higher than that of Sample U (̃) [28].

Ge and Dopant Concentration
PL Characterization. The samples were characterized for PL using a Horiba LabRAM HR Evolution instrument with a Symphony II InGaAs detector. A 1064 nm laser was used on all the samples. The excitation power is ~20 mW at the surface of the sample, and a 50X near infrared objective lens was used. The acquisition time was 6 seconds per spot. The grating used was 300 gr/mm. In order to reduce the noise, first a measurement using "signal" mode was taken, followed by a measurement with the same conditions and under dark condition. The data was then obtained using "signal minus dark" mode. This process was repeated for the measurement of each sample. The PL characterization results are summarized on Fig. 5 above. Note that the As and P doped samples are blue shifted compared to the MIT Ge control. This is due the lower dopant concentration of the samples (1×10 19 and 7×10 18 for the As and P doped samples respectively and 4×10 19 for the MIT Ge control sample as shown on Table 1). Higher doping leads to band narrowing [41], therefore the redshift in PL peak positions. On the other hand, the PL intensities of the P and As doped samples are similar to that of the MIT samples despite of the lower n-doping concentration. This result suggests better material quality from the MOCVD grown samples compared to the delta-doped sample. The slight difference in PL intensity between the As and P doped sample could be attributed to the higher EPD value of the As-doped unannealed Ge. Fig. 6 below shows a comparison for the PL of the annealed and un-annealed samples. The blue shift observed on the annealed samples is mostly due to the Si-Ge interdiffusion, given that the dopant concentration did not decrease after annealing as shown on Fig. 4. This is also evidenced by the fact that the indirect transition is enhanced relative to the direct after the annealing, which is observed by a decrease on the relative intensity of direct transitions on the annealed sample. The reduced overall intensity of the annealed sample can be explained by the fact that Si diffused to Ge makes the material more indirect. Regarding the intensity of the on-axis and off-axis samples, this is related to the differences in the dislocation density between the samples. For the P doped sample, the dislocation density seems to be lower for the on axis sample and for the As doped sample the dislocation density for the on-axis sample is slightly higher on average, as shown in Table 2. However this has little impact in comparison to the annealing process and Si-Ge interdifussion. Fig. 7 below shows the impact of the annealing process and TDD as measured by ECCI on the PL intensity.   Fig. 7 below. The energy difference between the direct and indirect gaps as well as the (red) shift on emission (band gap narrowing) shows the differences on dopant concentration between the different samples. Given that there is a linear dependence between the Γ-point bandgap shrinkage and heavy n-doping [29], the band narrowing can be predicted using the dopant concentration given on Table 1 for each of the samples. Using a first order phenomenological model for band gap narrowing and using the values of = 0.013 and Δ = 10 −21 / −3 given for Ge [41], the narrowing was calculated and is summarized on Table 4 below. Furthermore, it is possible to calculate the dopant concentration from the observed energy difference between the direct and indirect bandgaps on Table 3. The results are also shown on Table 4 as follows. Note that the calculated band gap narrowing shows agreement with the measurements above and that the calculated dopant concentration agrees with the measured concentrations within an order of magnitude.

Summary.
To summarize, this work studied thin film quality and photoluminescence of n-doped epitaxial Ge-on-Si. Surface roughness, strain, etch pit desnity, Si-Ge interdiffusion, dopant diffusion, and photoluminescence were characterized to study the impacts of defect annealing and Si substrate offcut on the Ge film quality and most importantly, the light emission properties. All samples have a smooth surface (roughness < 1.5 nm), and the Ge films have a small tensile strain. Six degree offcut was shown to have little impact. Defect annelaing was shown to decrease the photoluminescence intensity greatly due to Si-Ge interdiffusion, which also causes a blue-shift in the emission wavelength. Compared to a benckmarking sample from the first Ge laser work in 2012, the P or As-doped Ge films have similar photoluminescence intensity, smoother surface and less doping concentrations, which are promising for Ge lasers with better light emission efficiencies.