Design and growth of multi-functional InAsP metamorphic buffers for mid-infrared quantum well lasers on InP Free

Strain-relaxed metamorphic buffers have been extensively studied as a virtual substrate to grow wide-ranging photonic and optoelectronic devices that are highly lattice-mismatched to binary substrates.^1–3 The main functionality of the metamorphic buffer layers is to serve as a mechanical bridge that tunes the in-plane lattice parameter to the desired value for coherently strained active region growth, while effectively reducing the density of threading dislocations (TDs) that arise at the buffer/substrate interface.^4,5 High performance metamorphic solar cells, detectors, and laser diodes have been demonstrated on Si, GaAs, and InP substrates with relatively thick buffer layers.^6–11 

While the metamorphic buffer is indispensable to confine TDs away from device active regions where injected or generated electrons and holes recombine or are collected, buffer layer growth can significantly increase material growth cost and time. The buffer layer thickness typically ranges from 2 to 5 μm, depending on the desired TD reduction efficacy.¹² Therefore, it would be of significant benefit to be able to engineer metamorphic buffers with multiple functionalities in addition to their primary role as a mechanical bridge. Doing so offers significant savings in growth cost and time, as well as the potential for improving device performance.

We recently reported the demonstration of room-temperature (RT), mid-infrared quantum well laser diodes grown on InP substrates emitting at 2.7 μm, in which the laser operation relied on a multi-functional metamorphic buffer (MFMB) layer.¹³ In the laser design, compositionally graded InAs_xP_1-x buffers grown on the InP substrate served not only as a virtual substrate to grow coherently strained multi-quantum wells (MQWs), but also as a graded-index optical bottom cladding layer for mode confinement, enabling extension of the Type-I QW emission wavelength up to 2.74 μm. Note that the 2.5–3.0 μm wavelength spectrum is not accessible by either conventional QW lasers or quantum cascade laser (QCL) technology on InP due to the high strain required in the laser active region.^14,15 Instead, GaSb-based Type-I QW lasers have been well developed for the aforementioned spectral regime. However, the use of InP wafers over GaSb could provide several advantages such as well-established commercial processing infrastructure, reduced cost, and higher thermal conductivity.

In this paper, we report on the design and growth of InAsP MFMBs with varying compositions and compare the performance of Type-I QW lasers grown on them. High-resolution x-ray diffraction (HRXRD) indicates high crystalline quality InAs QWs grown on the MFMBs with intense satellite peaks, while cross-sectional transmission electron microscopy (XTEM) confirms coherent strain (i.e., no misfit dislocations) in the MQW active regions for the three laser samples. Strong RT photoluminescence (PL) intensities across a wavelength range from 2.70 to 3.05 μm are observed, demonstrating the tunability available in this material system. Fabry-Perot lasers were processed and show characteristic temperatures and differential quantum efficiencies that are strongly affected by the energy band offset between the InAs QWs and InAsP MFMB waveguide.

Figure 1(a) illustrates the structure of an InAsP MFMB laser diode grown on an InP substrate. Three samples were grown on n-InP (001) substrates in a Veeco Mod Gen-II solid source molecular beam epitaxy (MBE) chamber. After oxide desorption and an n-InP buffer layer growth, step-graded n-InAsP metamorphic buffers were grown with final InAsP target compositions of x = 0.5, 0.6, and 0.7. The MFMBs were designed to enable growth of InAs QWs thicker than the ∼5 nm critical thickness (h_c) on InP substrates while maintaining threading dislocation densities (TDDs) <5 × 10⁶ cm⁻². The designs of the three laser diode active regions are summarized in Table I, and more details about the InAsP buffer MBE growth conditions can be found in our previous publication.¹⁶ The bottom InAs_xP_1-x waveguide layer is 450 nm thick to ensure near-complete strain relaxation. A five-QW strain-balanced InAs/In_yGa_1-yAs MQW structure was grown in the active region to improve the optical gain. The compositions and thicknesses of the In_yGa_1-yAs barriers were determined by the zero-stress model and were adjusted in each case according to the InAs QW thickness and the composition of the InAsP MFMB layer.¹⁷ As shown in Table I, the barrier In content and thickness were increased with increasing QW thickness in order to maintain near-zero strain in the active region. Figure 2 shows in situ reflection high-energy electron diffraction (RHEED) images during the MQW growth for Sample A. Streaky patterns were observed throughout the five QW growths, indicating that a smooth surface morphology was maintained. After the growth of the MQW active region, an upper InAs_xP_1-x waveguide layer was grown. Due to the asymmetric waveguide/cladding structure, the thickness of the upper waveguide layer was determined from mode confinement simulation.

Figure 1(b) presents the COMSOL-simulated results showing refractive index (red) and TE mode intensity (blue) over the growth direction for Sample A; the refractive index of InAsP graded buffers (bottom cladding) is linearly interpolated from the values for InP (n = 3.1) and InAs (n = 3.51).¹⁸ The confined light intensity peak is located at the center of the active region. Both sides of InAs_0.5P_0.5 were doped gradually from 5 × 10¹⁷ to 1 × 10¹⁷ cm⁻³ toward the active region to reduce free carrier absorption. For the upper cladding material, we chose a highly lattice-mismatched Al_0.5Ga_0.5As layer because of its low refractive index that is suitable for transverse optical confinement. Also, a high density of edge misfit dislocations at the Al_0.5Ga_0.5As/InAs_0.5P_0.5 interface due to the large lattice mismatch helps to prevent threading dislocations from penetrating back to the QW active region.¹ To promote formation of edge dislocations, a 50 nm thick GaAs layer was first grown at a low growth rate of 0.1 μm/h. Finally, the epitaxial laser structures were capped by a 200 nm thick 2 × 10¹⁹ p-doped GaAs contact layer.

The optical micrograph in Fig. 3(a) shows a smooth surface morphology with clear cross-hatches along the $[11¯0]$ and $[110]$ directions; a clear, biaxial cross-hatch pattern is typically indicative of long misfit dislocation line-lengths, efficient strain relaxation, and low TDD.¹ No other extended defects such as faceted trenches or cracks were observed from the micrographs.¹⁹ Atomic force microscopy (AFM) images reveal a root-mean-square roughness of ∼5 nm as shown in Fig. 3(b). The highly lattice-mismatched AlGaAs cladding growth did not worsen the surface morphology even with the high degree of tensile strain relaxation.

We conducted (004) and (115) XRD reflections to study the composition, strain, and tilt of the samples. The (004) 2Theta-Omega linescans in Fig. 4(a) reveal the entire structure of the MFMB lasers including strong satellite peaks from the InAs MQW, which indicates a high crystalline quality in the active region. The tilt of the MFMB layers are minimal at ∼0.05° from the (004) reciprocal space maps (RSMs) (not shown here). The (115) RSMs of Figs. 4(b)–4(d) reveal coherent InAs MQWs with no strain relaxation when grown on MFMB layers with appropriate strain balancing. Combining the (115) RSMs with the film tilts [from the (004) RSMs], the cap compositions of the final MFMB layers were determined to be 51%, 61%, and 69% for Samples A, B, and C, respectively; all were found to be ∼100% relaxed.

The XTEM images of Fig. 5 show crisp interfaces between InAs MQWs and InGaAs tensile strained barriers. No misfit dislocations are observed from the XTEM investigations, confirming that the MQWs are coherently grown on their MFMB layers as shown in the HRXRD measurements. It should be noted that coherent growth of strained QWs on a metamorphic layer is challenging, because misfit dislocations often form even under the calculated h_c.^14,20 This is because the existing TDs are a heterogeneous source for misfit dislocations at the interface.¹ Furthermore, in the case of MQWs, strain energy keeps accumulating as more QWs are grown and may eventually surpass the maximum allowable strain energy, even if the single QW layer does not. To resolve these challenges, a combination of low QW growth temperature and strain-balanced QW structure was employed.¹³ We found that low growth temperatures (420 °C) for strained QWs can suppress formation of misfit dislocations by kinetically limiting the dislocation glide velocity.

Figures 6(a) and 6(b) show photoluminescence (PL) spectra from the three samples at 77 K and 300 K. The PL full-width at half-maximum (FWHM) values at 77 K are 22.9 meV, 18.8 meV, and 15.5 meV for Samples A, B, and C, respectively. It is interesting to note that Sample C, which has the highest As content (69%) in the InAsP MFMB, shows the smallest FWHM among the samples. The larger FWHM of Sample A may be due to the thinner QWs (8 nm), where slight roughness at the QW/barrier interface can result in higher variation in the quantum confined energies. Also, the larger lattice misfit between the QW and barrier in Sample A may increase the interface roughness, as can be observed in the X-TEM images of the three samples (Fig. 5). At 300 K, the PL peaks red-shifted to 2.70, 2.86, and 3.05 μm for Samples A, B, and C, respectively, closely matching the predicted emission wavelengths for the three samples (Table I).

The three laser samples were fabricated into 10 μm-wide Fabry-Perot lasers. Details on the processing steps and device measurement setup were described elsewhere.¹³ Figure 7(a) shows lasing spectra from the three lasers in the pulsed mode. At 77 K, the lasing wavelengths from the three lasers match the PL peak wavelengths, confirming that the lasing action is achieved from the QW active region, not from either the waveguide or barrier layers.⁸ The observation of laser oscillation over a range of designed wavelengths and compositions—beyond our initial demonstration on InAs_0.5P_0.5 in Ref. 10—illustrates the versatility of the MFMB as a platform for metamorphic lasers. As the testing temperature increases, the emission from each laser red-shifts due to the narrowing bandgap of the InAs MQWs. The maximum operating temperatures were 250 K, 200 K, and 125 K for Samples A, B, and C, respectively. More detailed discussion about the maximum lasing temperature is presented in the Discussion section along with the active region band alignments.

Figure 7(b) displays light-current-voltage (LIV) curves for Sample A. The threshold current density is as low as 180 A/cm² at 77 K and gradually increases to 4 kA/cm² at 250 K. The turn-on voltage for the laser is ∼2.5 V at 77 K, which indicates high series resistance across the laser diode; such a low-bandgap diode would ordinarily be expected to be well beyond turn-on at <1 V.²¹ We speculate that the band offset at the interface between the InAsP waveguide and GaAs upper cladding layer contributes to the high turn-on voltage. Samples A and B operate in continuous-wave (CW) mode up to 175 K and 150 K, while Sample C did not lase CW at 77 K. This heating issue could be improved by employing grading layers at the interfaces such as p+GaAs and p-AlGaAs.^22,23 Figure 7(c) shows the pulsed mode threshold current densities as a function of temperature. Samples A and B show changes in characteristic temperatures around 150–175 K. We attribute this to thermal carrier escape from the MQWs to the barriers, especially electrons in the conduction bands.

To understand the thermal performance of the three lasers grown on InAsP MFMBs, we investigate the energy band alignments in the MQW active regions as shown in Figs. 8(a)–8(c). The energy levels and band offsets were determined based on the model-solid theory,²⁴ and the materials’ parameters were taken from Ref. 25. Figure 8 shows that the valence band offsets between the InAsP waveguide (MFMB final layer) and the ground state of the QWs are relatively large for all three samples. However, the conduction band offsets are small at 129 meV, 103 meV, and 80 meV for Samples A, B, and C, respectively. We attribute the relatively poor thermal performance of the QW lasers to these low conduction band offsets and the device heating issue.

While the use of high-quality InAsP MFMBs enables emission at wavelengths of 2.5–3.0 μm from InP-based Type-I QW lasers, their thermal performance needs significant improvement to be considered for practical applications. Adding an electron blocking layer such as InAlAs and n-modulation doping in the barriers may raise laser operation temperatures.¹³ Also, the aforementioned heating issues could be diminished by optimizing the doping levels in the InAsP MFMBs and by modifying the compositions of the Al_0.5Ga_0.5As upper cladding layers, both of which may allow operation at considerably lower voltages.

In summary, we have presented InAsP MFMBs with a range of compositions designed to enable mid-infrared Type-I MQW lasers on InP substrates. The MFMB allows for coherent growth of InAs MQWs on InP by engineering the strain level between InAs and InP. Strong RT PL was observed in the 2.70–3.05 μm wavelength range, where neither conventional Type-I QW nor quantum cascade lasers operate efficiently. Lasing action was confirmed from electrically-pumped Fabry-Perot type lasers in the aforementioned wavelengths. Temperature-dependent measurements revealed performance degradation at longer wavelengths, largely due to carrier escape from the QWs. Future work will look to focus on improving electron confinement through blocking layers, as well as reducing turn-on voltage and parasitic series resistance to minimize sample heating. We show that carefully designed MFMBs can serve not only as a virtual substrate, but also as a laser cladding layer, which reduces device growth cost and time. The MFMB concept could be extended to other material systems such as GaAsP on Si for laser diodes or solar cells.

M.L.L. gratefully acknowledges support from the National Science Foundation (NSF) under Grant No. 1713068. D.W. gratefully acknowledges support from the National Science Foundation (NSF) under No. DMR-1629570.