1. Introduction

Quantum cascade lasers (QCLs) are important infrared light sources with numerous applications in defense and civilian fields. Low optical absorption in the atmospheric window spanning from 4.4µm to 4.8µm has given rise to a number of applications. While QCL performance at longer wavelengths has been improving significantly [1], culminating in continuous wave (CW) room temperature wallplug efficiency (WPE) exceeding 20% and optical power of 5W at ~5µm [2], the 13% WPE and optical power of 3W results reported in Ref. 3, are still among the best results in the technologically important 4.5µm to 4.8µm spectral region.

Injection efficiency for the upper laser level is usually considered to be the main factor limiting laser performance of the short wavelength QCLs. Thermal carrier leakage from the upper laser level 4 to continuum states located above the barriers reduces laser population inversion at a given pumping current density and therefore increases laser threshold current density. Another carrier leakage path is due to carrier excitation to level 5 and subsequent scattering to states other than level 4 [4]. Carrier escape to continuum states C and level 5 can be suppressed by increasing energy spacings E_C4 and E₅₄ (Fig. 1 ). A straightforward approach to increase E_C4 is to increase potential barriers height by employing higher strain In_xGa_1-xAs/ Al_yIn_1-yAs composition. In addition to larger E_C4, employment of higher barriers also leads to larger E₅₄ for the same laser transition energy. Therefore, use of higher strain composition is a very attractive tool for suppressing carrier leakage from the upper laser level.

 

Fig. 1 Active region band diagram of the new design based on Al_0.78In_0.22As/In_0.72Ga_0.28As.

Download Full Size | PDF

Another advantage in using a high strain composition is that position of indirect X and L-valleys relative to direct Γ-valley minimum increases as strain increases [5]. Carrier leakage from the upper laser level through indirect states progressively increases as laser emission wavelength reduces below 5μm. Therefore, short wavelength QCLs with high strain composition should benefit from suppressed carrier leakage through indirect states. Increase in barriers height, however, also increases linewidth γ₄₃ of the laser transition. Linewidth is proportional to [6]:

Typical thicknesses for the first and the second active region barriers are in the range of two to five monolayers (six to fifteen Angstroms). Depending on epi-growth quality, a barrier thickness of two to five monolayers is comparable to, or even less than total roughness at the both interfaces of a barrier 2Δ. In other words, the whole barrier thickness d may be irregularly graded. In this case, Eq. (1) may not be directly applicable to calculate laser transition linewidth since it was derived assuming d > 2Δ. Using electroluminescence data we show in this work that very good epi-quality can be achieved for laser structures with ~2.0% lattice mismatched barriers and ~1.3% lattice mismatched quantum wells.

2. Laser design

In the new 4.7µm active region design with highly strained barriers and quantum wells our goal was to reproduce, as closely as possible, the Non Resonant Extraction, or NRE, design [3, 7], as shown in Fig. 1. The laser transition was kept vertical and the overlap between the upper laser level and the lower laser level wavefunctions was also the same as in the design from Ref. 3. With the exception of the energy spacing between the upper laser level 4 and the top of the barriers, E_C4, and the energy spacing between the upper laser level and the active region level located above it, E₅₄, all design parameters, such as laser transition matrix element, upper and lower laser lifetimes and voltage defect, for the new structure were very similar to the parameters of our earlier design [3]. E_C4 and E₅₄ were increased from 230meV and 63meV for the old design up to 420meV and 72meV, respectively, for the new design as a result of a larger band offset of the higher strain composition.

The 40-stage quantum cascade laser active region, along with the waveguide and contact layer sequence discussed in Ref. 3, was grown by molecular beam epitaxy. First, a small part of the new material was processed into round mesas for electroluminescence (EL) testing. Measured pulsed EL was centered at 4.65μm and had full width at half maximum (FWHM) of only 22meV, which is 10% lower than EL FWHM for design reported in Ref. 3 that was based on Al_0.64In_0.36As/In_0.67Ga_0.33As resulting in ~1% strain both in barriers and quantum wells. The reduction in FWHM cannot be explained by improvement in epi-quality since both structures were grown under the same conditions. Also, as mentioned above, the laser transition was vertical in both cases and therefore it is unlikely to be the reason for the observed EL narrowing. Better understanding of the observed effect may require study of the exact composition of the ultra thin active region layers, its dependence on strain and numerical analysis of electron levels/wavefunctions in such structures that does not rely on bulk band offset approximation. This encouraging result shows however that an excellent epi-quality can be achieved for structures based on a very high strain composition.

3. Tapered waveguide geometry

The remaining wafer was then processed into tapered buried heterostructure (BH) geometry. We have observed for 4.7μm QCLs that the optical damage threshold for the front facet coated with Al₂O₃ lies at ~10MW/cm², which results in maximum single-ended optical power of approximately 3W. The output facet is more susceptible to optical damage than any other part of the device because it is placed where the optical density is the highest. In addition, the dielectric anti-reflective coating has a relatively low thermal conductivity, unlike the back facet whose high reflection coating contains a metallic layer that enhances heat removal. While the utilization of wider straight devices would reduce the optical power density on the facet, this approach leads to lower beam quality and increases active region self-heating. Instead, we fabricated devices comprising a long and narrow straight section, which strongly favors the TM₀₀ order mode and a short tapered section at the output facet to increase the optical damage threshold.

The tapered waveguide is illustrated in Fig. 2 . Devices were designed to have a total cavity length of 10mm and ridge width W_R of 7.5μm in the straight section. The output facet width W_T was chosen to be 20μm, in order to reduce the power density and thus increase the damage threshold by a factor of two compared to typical straight devices, while keeping the self-heating at the facet location to a minimum. The taper length L_T was chosen to be 0.5mm, so that the taper angle α_T was 0.7° and the mode expanded adiabatically in the tapered section. Even though it was not necessary to reduce the optical power density at the back facet, we used the same taper geometry there as well, so that our devices were symmetrical. The resulting length of the straight section was 9mm.

 

Fig. 2 Schematic of a tapered waveguide QCL.

Download Full Size | PDF

Tapered waveguides have previously been used in the design of QCLs, to realize master oscillator power amplifier devices [8, 9] and high-peak-power, high-beam-quality pulsed devices [10, 11]. However, the motivations to use tapered waveguides in these applications were different from ours, as were the resulting waveguide geometries. In Refs. 8 and 9 the taper was used as a monolithically integrated single-pass external optical amplifier, i.e. a device through which the light emitted by the distributed feedback QCL oscillator, travels only once before leaving the device. In the latter application, on the other hand, the taper was part of the laser cavity as in our application. However, the motivation for tapering the waveguide in this application was also very different from ours. In Refs. 10 and 11, the main purpose of the taper was to increase the device area in order to increase its peak power in low-duty-cycle pulsed operation while maintaining a good beam quality. In such devices, the length of the tapered section was typically more than half of the entire length of the device and its area typically constituted most of the area of the entire device. Since self-heating is not significant in low-duty-cycle pulsed operation, the output facet width was typically 50-200 μm. While adequate for these operating conditions, such large waveguide widths are not compatible with high-duty-cycle and/or CW operation because of the high active region self-heating resulting from the significant waste heat generated in QCLs. In our work the purpose of the taper was to reduce optical intensity at the output facet, while keeping self-heating to a minimum for high-performance CW operation. Therefore the taper width and length are kept to a minimum. As a result, the taper length is equal to only a small fraction of the length of the entire device and the taper area equals a small fraction of the entire device area.

4. Silicon carbide submounts

A tapered 10 mm laser was packaged into a hermetically sealed butterfly package for testing. Typical semiconductor laser package assembly involves bonding of the laser to a submount (substrate material) that, in turn, is mounted on a heat spreader or a heat sink. Two critical parameters need to be considered when choosing the substrate on to which the semiconductor laser chip is mounted. The first is the thermal conductivity of the submount, which needs to be as high as possible in order to remove the heat efficiently from the laser. The second is the coefficient of thermal expansion (CTE) of the submount with respect to the coefficient of thermal expansion of the semiconductor laser material. The CTE of the submount must be close to that of the laser in order to minimize mechanical stress in the laser chip during large temperature cycles, both during the operation of the laser but also during the mounting process which, for reliability purposes, typically uses AuSn eutectic or other hard solder.

One of the most technologically important submount materials for InP-based semiconductor lasers is aluminum nitride (AlN) because of its relatively high thermal conductivity of 200W/mK and because its CTE is perfectly matched to that of InP [12]. However, even though AlN has a relatively high thermal conductivity, higher laser performance can be achieved employing submount materials with even higher thermal conductivity.

Diamond has the highest known thermal conductivity (~2000 W/mK) of all natural materials, exceeding that for AlN by an order of magnitude. Therefore, employment of diamond submounts is expected to significantly reduce laser active region temperature under the same laser driving conditions. However, there is a large CTE mismatch between InP and diamond. Therefore, to avoid the undesirable mechanical stress, it is necessary to use soft solders, such as indium, for laser bonding. Soft solders suffer from electromigration from the bonding area to the laser facets, which eventually leads to facet damage and laser destruction. This significantly reduces laser reliability. Therefore, diamond is rarely used as a submount material for commercial semiconductor laser applications.

The second best material for high heat waste applications is silicon carbide (SiC) with thermal conductivity of up to 500 W/m*K, two and a half times that for AlN. As with AlN it can be made semi-insulating, which facilitates all the electrical interconnects. In addition, its CTE of 4.0·10⁻⁶ K⁻¹ is close to that of InP equal to 4.5·10⁻⁶ K⁻¹. However, even this relatively small CTE mismatch, still, can lead to reduced laser reliability, especially when the final assembly undergoes large thermal cycles during packaging and subsequent laser operation.

In this work we used a thin layer of AlN as a buffer layer between SiC wafer and InP laser for the purpose of absorbing mechanical stress caused by the CTE mismatch between the two materials [13]. Since AlN has a relatively high thermal conductivity and its thickness is only 10-20 μm, presence of this buffer layer does not significantly increase overall thermal resistance. Thermal simulation showed that the 4.7 µm QCLs with the active region and waveguide designs discussed in this work mounted epi-down on an AlN/SiC submount has average active region temperature 10 degrees lower under roll over conditions compared to a similar laser mounted on an AlN submount.

We processed an AlN(20μm)/SiC(250μm) wafer into laser submounts. As a hard solder we used AuSn eutectic solder system that does not exhibit electromigration. After numerous large temperature cycles (20°C to 300°C) we did not observe any visible mechanical damage to the laser, which indicated that there was no strain buildup at the laser/submount interface.

5. Experimental data

CW light vs. current and voltage vs. current (LIVs) characteristics of a hermetically packaged 10 mm long laser with 7.5µm-wide central section tapered up to 20 µm at laser facets mounted on AlN/SiC submounts are shown in Fig. 3 . The laser temperature was set to 283K. Maximum optical power of over 4.5W and efficiency of 16.3% have been demonstrated for the laser, the best result at this wavelength. Maximum efficiency is reached at 4W, close to the maximum power, which is important for practical high power applications. Also, the laser had a very low threshold current density of 0.8kA/cm². This is the first demonstration of tapered CW BH QCLs.

 

Fig. 3 CW LIV of a hermetically packaged 10 mm long laser with 7.5µm-wide central section tapered up to 20 µm at laser facets mounted on AlN/SiC submounts. The laser temperature was set to 283K.

Download Full Size | PDF

For wide-ridge devices higher order modes are expected to dominate even at threshold, as shown in Ref. 14. Beam image of the tapered laser at pumping current slightly above threshold is shown in Fig. 4 . The beam is single-lobed, i.e. consists of a zero order mode, along the two axes, with a ratio of the two diameters equal to about three. When this ellipticity is not desirable for an application, it can be corrected by the use of circularizing optics. The fact that the beam is purely TM₀₀ shows that the 7.5 μm-wide, 9 mm-long straight section of the waveguide is sufficient to prevent higher-order transverse modes from lasing, even though facet reflectivity is higher for these modes in TM polarization. The beam shape stayed unchanged up to current density three times that at threshold, with some beam steering observed at maximum current.

 

Fig. 4 Beam of the hermetically packaged 10 mm long laser with 7.5µm-wide central section tapered up to 20 µm at laser facets mounted on AlN/SiC submounts. The laser current was set to 0.75A (see Fig. 3 for LIV). The laser temperature was set to 283K.

Download Full Size | PDF

Figure 5 demonstrates reliability data for the same laser packaged into an air-cooled QCL system delivering 3W in a collimated beam. Negligible laser performance degradation has been observed for over 2,000 h. This result demonstrates that employment of a very high strain composition and new AlN/SiC submounts does not compromise laser reliability. It also shows that tapered waveguide geometry is an effective tool for increasing optical damage threshold. High laser efficiency, leads to low heat waste and as a consequence low TEC power consumption. As a consequence, measured wallplug efficiency of the entire system was in excess of 10%, including thermoelectric cooler power consumption. High overall efficiency for the laser with thermoelectrical cooler assembly is crucial for practical applications with limited cooling capability.

 

Fig. 5 Reliability data with periodic hourly laser turn ons/offs to subject the QCL to a large thermal stress.

Download Full Size | PDF

6. Conclusion

In conclusion, we have presented experimental data on tapered 4.7 µm QCLs based on the nonresonant extraction approach. Electroluminescence FWHM of only 22meV showed that excellent epi-quality was achieved despite using highly strained active region quantum wells and barriers. WPE of 16.3% and optical power of over 4.5W have been demonstrated for a 10 mm by 7.5 µm HR-coated laser mounted on an AlN/SiC submount. Reliable CW operation was demonstrated for a laser system for over 2,000 hours at 3W level with overall system efficiency (including power input to the TEC) exceeding 10%.