Quantum Cascade Lasers Grown by Metalorganic Chemical Vapor Deposition on Foreign Substrates with Large Surface Roughness

Abstract

The surface morphology of a buffer template is an important factor in the heteroepitaxial integration of optoelectronic devices with a significant lattice mismatch. In this work, InP-based long-wave infrared (~8 µm) emitting quantum cascade lasers with active region designs lattice-matched to InP were grown on GaAs and Si substrates employing InAlGaAs step-graded metamorphic buffer layers, as a means to assess the impact of surface roughness on device performance. A room-temperature pulsed-operation lasing with a relatively good device performance was obtained on a Si template, even with a large RMS roughness of 17.1 nm over 100 µm². Such results demonstrate that intersubband-operating devices are highly tolerant to large RMS surface roughness, even in the presence of a high residual dislocation density.

1. Introduction

The monolithic integration of high-performance mid-infrared quantum cascade laser (QCL) devices onto foreign substrates enables cost-efficient and scalable manufacturing for a variety of applications including gas sensing and free-space communication links, making the technology more accessible to customers. Recently, there has been significant progress reported for QCLs hetero-epitaxially grown on GaAs and Si substrates.

The first experimentally demonstrated QCLs grown on foreign substrates were InAs-based using a GaSb buffer on miscut silicon for long wavelength emission (λ~11 µm), which demonstrated a performance comparable to that for devices grown on native substrates [1]. By contrast, the first reported ternary AlInAs/InGaAs-active InP-based QCLs on GaAs substrates grown by molecular beam epitaxy (MBE) suffered from a high threshold current density (4.1 kA/cm²), and only milliwatt output powers were measured [2]. Many efforts since then have been undertaken to improve the epitaxial material quality and mitigate the device performance degradation. Recently, a room-temperature continuous wave operation for MBE-grown QCLs on miscut Si substrates with an output power greater than 0.7 W and a low threshold current (1.3 kA/cm²) has been achieved [3]. Meanwhile, our recent work shows that the InP-based QCLs on foreign substrates grown by metalorganic chemical vapor deposition (MOCVD) can also have a comparable device performance to its counterpart on native substrates, even in the presence of a high dislocation density [4,5,6]. However, these prior studies have not elucidated the role of the root-mean-square (RMS) surface roughness of buffer layers on subsequent device performances.

In general, one of the key pathways for achieving the improved performance of QCLs on foreign substrates is the development of metamorphic buffer layers (MBLs), which are typically used as the intermediate template to integrate the actual device onto a foreign substrate, as well as to accommodate the large lattice mismatch. For most of the above-mentioned work of InP-based QCLs on either GaAs or Si, regardless of the buffer layer strategy, the threading dislocation density (TDD) value is in the mid-10⁸ cm⁻² range, and the buffer surface is generally maintained at a relatively smooth level with RMS roughness below 2 nm over 100 µm² area. The QCL devices are unipolar intersubband operation devices, solely dependent on the transition and tunnelling behavior of electrons in conduction bands. Therefore, the non-radiative electron and hole recombination process, which is an important recombination pathway in interband transition quantum well-based lasers, is avoided. This is so that the presence of threading dislocations cannot affect the operation for QCLs. By contrast, the surface morphology may affect the structural quality of the active region layers and then degrade the QCL device performance, because the rough interfaces can influence the electron transport and phonon scattering, although the actual mechanism needs to be further investigated.

In this paper, InP-based long-wave infrared (λ~8 µm) lattice-matched quantum cascade laser structures grown on miscut GaAs and Si are investigated. The use of a thick InAlGaAs step-graded MBL leads to a high surface roughness and cross-hatching surface morphology. Yet, we observe that devices grown on both GaAs and Si exhibit lasing characteristics with comparable performance. The results in this work show that QCL devices are generally tolerant to high surface roughness.

2. Materials and Methods

A Si-doped graded metamorphic InAlGaAs quaternary buffer was grown on a 1 mm thick GaAs substrate and a 650 µm thick Ge/Si-engineered template by MOCVD with orientation in 6° off towards <111> direction, as shown in Figure 1. The growth temperature and pressure were nominally set at 615 °C and 50 mbar, respectively, with the Si-doping level targeted at around 1 × 10¹⁸ cm⁻³. The composition of the In_xAl_(1−x−y)GayAs was step graded from In_0.05Al_0.19Ga_0.76As to In_0.54Al_0.39Ga_0.07As using twelve 0.16 µm thick intermediate layers at fixed intervals with an average growth rate of 12 µm/h. A thick cap layer of 0.9 µm In_0.56Al_0.365Ga_0.075As was grown after the step grade buffer at a growth rate of 8 µm/h. Finally, a 0.3 µm thick Si-doped (1 × 10¹⁸ cm⁻³) InP layer was grown as a cap layer for the subsequent QCL growth at a growth rate of 4 µm/h.

The properties of the InAlGaAs MBL on GaAs and Si substrates were analyzed by atomic force microscopy (AFM) and the electron channeling contrast imaging method (ECCI). ECCI is a non-destructive characterization technique based on the back scattered electron (BSE) detection in a scanning electron microscope (SEM). The contrast changes in the channeling image can detect small changes in the orientation and strain surrounding the defects. Therefore, it has been widely used as a rapid technique to measure the surface defect density from III–V materials [7].

The band structure for the lattice-matched QCL active region utilized in this study is based on a two-phonon resonance design [8]. This is also the same QCL structure previously adopted for realizing QCLs on on-axis GaAs and Si by MOCVD, and the growth details and full laser structure can be found here [4,5]. As shown in Figure 2, the QCL on the Si wafer was fabricated into a deep-etched ridge waveguide with an average core region width of ~25 µm. First, a 1 µm silicon nitride layer was deposited onto the wafers, serving as the hard mask for subsequent etching processes. The ridge was formed using a combination of inductively coupled plasma (ICP) etching for the active region layers and wet chemical etching for the InP claddings, which stops at an n-doped InGaAs etch-stop layer. Then, a 0.3 µm thick silicon nitride insulation layer was deposited everywhere. The silicon nitride layer on top of the ridge was etched away and a Ti/Pt/Au multilayer was used as top contact metals. A similar process was applied to make side contact with a stack of AuGe/Ni/Au metals, which was ~50 µm away from the laser core. Before cleaving, the backside of the wafer was thinned down to ~80 µm for the QCL on Si. However, due to the miscut angle in the Si substrate, an oblique angle (~3.4°) was found for the front facet from a top view. Nevertheless, planar interfaces between the Ge/Si substrate, the step-graded InAlGaAs, and the InP cladding layer were observed from a cross-sectional view. The same fabrication process was also applied to the GaAs substrate wafer, and the cleaved laser facet was observed in this case to be nearly flat and parallel. Using a finite-difference time-domain (FDTD) method and configuring a 3D structure of the uncoated ridge-guided laser with a facet angle of 3.4°, the reflectance of each facet was calculated to be 22.9%, which was about 3% less than the normal flat facet. Thus, the estimated mirror loss was increased by about 8%.

The devices on GaAs and Si substrates were then cleaved into 3 mm long bars and epi-side-up mounted onto copper heat sinks without any front and back facet coating. The laser device tests were performed at 293 K and a pulsed operating condition with a 200 ns pulse width and a 20 kHz frequency rate. The lasing spectrum was measured by FTIR at a resolution of 0.125 cm⁻¹.

3. Results

3.1. Material Characterization

After growing the InAlAs step-graded buffer on GaAs and Si, both wafer surfaces displayed a slightly hazy appearance to the eye. Deeper trenches and a high peak-to-valley ratio, but a less dense cross-hatching surface morphology, were observed for the growth on the Si wafer via optical microscopy as well as seen in a top view SEM image from Figure 3a,c from the observed dark-to-light contrast. In addition, the threading dislocation density was estimated to be ~2.2 × 10⁷ cm⁻² for the buffer on GaAs and ~1.9 × 10⁹ cm⁻² for the buffer on Si, as shown in Figure 3a,c via the ECCI method.

With sample surface morphology measured over 100 µm² area per scan by AFM, the 3D surface roughness was evaluated from selected amplitude parameters [9], such as the root-mean-square deviation in axial directions and full images, as well as the maximum peak height. The averaged RMS roughness over the full AFM image was estimated to be 6.5 nm on GaAs and 17.1 nm on Si, as shown in Figure 3b,d. Also, it was noticeable that, due to the presence of a cross-hatching pattern, the surface of both buffers was smoother along the longitudinal direction and had an RMS roughness value of 2.8 nm on GaAs and 6.0 nm on Si, respectively. Furthermore, we observed a higher maximum peak height of 103.1 nm for the buffer on Si, and this value was only 49.4 nm for the buffer on GaAs. These 3D roughness values indicate that the quality of the buffer on either GaAs or Si was quite poor, and as such, these RMS roughness values are generally thought to be not suitable for any laser growth, as they were significantly larger compared with prior reports of graded InAlAs buffer on Si with 3.3 nm over 400 µm² area [10] and the InP buffer on GaAs with an RMS roughness of 0.4 nm over 100 µm² area [4].

After growing the full QC lasers on both GaAs and Si templates, the Si wafer surface became significant hazier to the eye. X-ray diffraction (XRD) analysis was used to examine the material’s structural quality, as shown in Figure 4. The full width at half maximum (FWHM) of (004) InP peak for the QCL on Si and on GaAs 329 and 240 arcsec, respectively, are significantly larger compared to the value obtained from same laser structure grown on the on-axis Si (144 arcsec) with a low roughness InP buffer (1.5 nm) and moderately high TDD (7.9 × 10⁹ cm⁻²) [5]. The rough buffer layer surface, together with a high density of dislocations, is a possible factor leading to the observed peak broadening. In addition, similar broad satellite superlattice (SL) peaks were also observed, circled in Figure 4, which have also been reported previously for MBE-grown QCLs on foreign substrates with miscut angles [3,8]. In contrast, well-defined and distinguishable active region SL peaks were reported for MOCVD-grown QCLs on foreign substrates without miscut [4,5,6]. Thus, the use of miscut substrates and a non-optimized core region growth condition are believed to play a role in the large fringe broadening in Figure 4. The wide SL fringe peaks indicate possible interface roughening and compositional mixing for layers in the core region and would be expected to result in an inferior laser performance.

3.2. Laser Test Reuslts

The 3 mm long uncoated facet lasers for the QCL on GaAs and Si were tested at room temperature under pulsed operating conditions, and the results are compared in Figure 5a. Both laser devices exhibited a similar maximum output power, ~0.8 W per facet, while the threshold current density for the QCL on GaAs was much lower than the value on Si, 1.52 vs. 2.36 kA/cm⁻², respectively. Also to be noted, the operating voltage applied to QCLs on both substrates was similar, within a 0.8 V difference, and the central laser emission for both devices, as shown in Figure 5b, were around 8.12 µm, measured just above the threshold current, indicating the band structure of the core region was well maintained on both substrates. However, the series resistance for devices on GaAs was higher than for devices on Si, 3.09 vs. 2.23 W. Taking the shorter dynamic range observed for the QCL on GaAs and the above device observations into consideration, it is reasonable to conclude that the doping level for the QCL on GaAs was actually lower than on the Si substrate. This could be caused by a temperature reduction, due to the use of a thicker GaAs substrate (~1 mm thick) compared to a Si substrate (~650 µm) or due to the pocket temperature variation in the reactor chamber. Even though the QCL on Si displays a higher dynamic range, the maximum output power and the wall-plug efficiency were still lower than for devices on GaAs. A likely reason may be the angled cleaved facets, which could result in a higher mirror loss.

4. Discussion

As the growth condition was not yet optimized for the laser structure and the buffer template on miscut substrates, rough and hazy surfaces with a high estimated TDD was observed. Surprisingly, the QCL devices grown on those unoptimized templates still show lasing characteristics with reasonably low threshold current densities and pulsed power output. To further investigate the impact of the buffer’s surface roughness on QCL’s device performance, the above device results were compared with that from our prior research on the work of QCLs on on-axis GaAs [4] and Si [5] with the same active region design, as shown in

Table 1. Device performance comparison for QCLs with same active region design on different InP-on-Si and InP-on-GaAs templates.

Lattice-Matched QCL on	RMS Surface Roughness (nm)	Dislocation Density (cm−2)	Threshold Current (kA/cm−2)	Dynamic Range (kA/cm−2)	Resistance (ohms)	Max Power (W)	Slope Efficiency (W/A)	Central Peak λ @ ~Ith (µm)
Miscut Si	17.1	1.9 × 109	2.36	2.76	2.23	0.78	0.49	8.11
Miscut GaAs	6.5	2.2 × 107	1.52	2.31	3.09	0.81	0.61	8.13
On-axis Si [5]	1.53	7.9 × 108	1.50	4.55	2.69	1.64	0.72	8.10
On-axis GaAs [4]	0.40	4.8 × 108	1.86	6.50	1.58	1.95	0.66	8.42

. By contrast with the devices on the miscut Si and GaAs, the RMS surface roughness for the devices on on-axis Si and GaAs of the same active region design were measured and found to be as low as 1.53 nm and 0.4 nm, respectively. These are the typical acceptable roughness values for a using a QCL on foreign substrates [2,3,4,5,6,10,11].

Comparing the characteristics of the QCL grown on miscut Si and on-axis Si, the threading dislocation density as well as the series resistance were not much different between both set of devices, indicating that the injector doping levels are likely at a similar level. In general, higher doping in the injector of the active region would lead to a larger threshold current and larger dynamic range. However, even given the similar doping level, a higher threshold current density (2.36 vs. 1.50 kA/cm⁻²), a shorter dynamic range (2.76 vs. 4.55 kA/cm⁻²), and a lower slope efficiency (0.49 vs. 0.72 W/A) were observed for the QCL on miscut Si compared to that on on-axis Si. This may stem from two factors: (1) the facet is slightly angled on the miscut Si, leading to higher mirror losses, and (2) the RMS surface roughness could play an important role in electron transport and interface roughness scattering in the active region, resulting in elevated self-heating and increased carrier leakage [12,13].

The measured series resistance for the QCL on miscut GaAs was double that of the value for on-axis GaAs, 3.09 vs. 1.58 ohm, possibly reflecting a lower Si doping incorporation in the active region, which would be expected to lead to a large change in the threshold current density. However, only a slightly lower threshold current density was actually observed, 1.52 vs. 1.86 kA/cm⁻². Note that the significantly lower dislocation density of the MBL on the miscut substrate, compared with that on-axis, may also play a role in the slightly lower threshold current density observed. Furthermore, the dynamic range was significantly reduced, 2.31 vs. 6.50 kA/cm⁻², which may be attributed to the combined effect of large surface roughness and the lower doping.

Despite the fact that there are many factors which may be influencing the observed device results, such as oblique facet and growth run variations, the laser results on the miscut substrate clearly show that the impact of surface roughness on device performance is not as significant a factor as might be expected. A higher threshold current, a lower output power and a smaller dynamic range could be anticipated from the large roughness, due to an increased electron scattering in the active region as well as optical scattering in the waveguide. However, quantitative analysis of the template roughness on the active region structural properties still needs further investigation. Scanning transmission electron microcopy (STEM) or atom probe tomography (APT) could be useful to assess the nanoscale interlayer material properties [14].

This work presents a demonstration of mid-IR InP-based QCLs directly grown on a pretreated Si template and fully by MOCVD. The results presented help elucidate the unique and robust property of QCLs: (1) their insensitivity to a large density of dislocations, as previously found [15], and (2) their high tolerance to substrate roughness newly observed from this work, pointing out that a simple buffer layer is adequate to attain devices with a reasonable performance. Nevertheless, suitable QCL active region growth conditions still need to be established for producing planar superlattice interfaces, which are expected to lead to an improved device performance for QCLs on foreign substrates, similar to the state-of-art high-efficiency QCLs on native InP substrates.

The above insights provide a practical solution for realizing compact and low-cost MIR devices integrated with silicon electronics. However, other laser device characteristics like in ref. [9], such as beam quality, reliability, and yield analysis, etc., should be further studied for completeness.

5. Conclusions

Previously, many efforts were undertaken to develop a high-quality buffer with a smooth surface and low defect density, which were traditionally thought to be a prerequisite for the successful integration of QCLs onto foreign substrates. In this work, we demonstrate InP-based QCLs on GaAs and Si substrates employing InAlGaAs step-graded metamorphic buffers fully grown by MOCVD with high surface roughness and high TDD. After developing the proper growth conditions, for the InGaAs/AlInAs/InP ternary material active region, room-temperature pulsed lasing with a relatively good performance is observed.

The above results indicate QCL devices to be relatively insensitive to both threading dislocations and surface morphology. This observed high tolerance could be attributed to the unique intersubband transition in QCL operations. However, a more in-depth understanding of the underlying mechanism responsible for such QCL behavior requires further investigation.

In addition, since MOCVD is the crystal growth method widely used for high-throughput III-V optoelectronic manufacturing, the above achievement will provide the impetus towards realizing QCL commercial products as well as integrating them with other opto- and electronic components onto complex Si photonic platforms.