Sapphire substrates slash the cost of deep UV lasers
Optimised growth enables the first optically pumped, low-threshold deep UV lasers on sapphire
BY XIAOHANG LI FROM KING ABDULLAH UNIVERSITY OF SCIENCE AND TECHNOLOGY (PREVIOUSLY WITH GEORGIA TECH), AND THEERADETCH DETCHPROHM AND RUSSELL DUPUIS FROM GEORGIA TECH
Among all the electromagnetic waves in the universe, the most relevant to us are those in the visible spectrum. It is the radiation at these wavelengths that enables us to see our surroundings and live, by breathing in oxygen generated by photosynthesis.
Just a short reduction in wavelength, however, and radiation may be far less welcome. In the deep UV, which corresponds to wavelengths of 280 nm or less, photons have higher energies and can interact with living cells, exciting electrons and disrupting critical chemical processes. This has led to mutation and cancer. Fortunately, however, UV light also has a good side. Highly energetic photons can be used to excite, engineer or break materials and chemical bonds in numerous applications.
One example of the positive use of UV light is as a source for enabling the identification of unknown biochemical substances. Radiation in this spectral range is ideal for this task, because all the functional groups are absorptive in the deep UV, compared to just a few in the visible. This makes for stronger signals that speed detection.
Using UV light to identify substances is not easy, however. Many applications cannot be taken out of the lab, because today's gas and solid-state deep UV lasers are bulky, complex, and costly.
To tackle this issue, DARPA has funded and led two consecutive programmes: Compact Mid-Ultraviolet Technology (CMUVT) and Laser UV Sources for Tactical, Efficient Raman (LUSTER). These programmes had a primary purpose of developing deep UV semiconductor diode lasers for efficient, low-cost, tactical Raman spectroscopy.
Benefits of a compact, deep UV laser are by no means limited to providing a source for Raman spectroscopy. This device could also aid non-line-of-sight (NLOS) communication, which is enabled by special characteristics of our atmosphere. In its upper region, ozone absorbs nearly all the sun's deep UV light. Consequently, light in this solar band is nearly non-existent on the earth's surface. The result is a low noise environment for the deep UV radiation, where photodetectors can reach a quantum-limited level of photon-counting detection.
Another appealing feature of the NLOS communication technology is that close to ground level, molecules and aerosols produce strong angle-independent scattering of deep UV light. This creates numerous communication paths from the source to the receiver, enabling the transfer of information even when line-of-sight obstacles are in the way − such as buildings in an urban city.
What's more, deep UV light from the source is absorbed moderately by the atmosphere. This limits the distance that information may be transmitted, making this communication technology ideal for tactical applications.
The most promising class of materials for making a compact, reliable, low-cost, and efficient deep UV semiconductor laser is the III-nitrides. Its attributes include high chemical and mechanical toughness and a very suitable range of bandgaps − the wavelength of the band edge can be as short as 200 nm.
Providing further motivation for the development of a deep UV III-nitride laser is the success of this material system in blue LEDs and laser diodes. But replicating performance at shorter wavelengths is far from easy. As of today, the wall-plug efficiency of most commercial deep UV LEDs is still in the low single-digit range, and there is yet to be a demonstration of an accompanying laser diode.
One of the biggest challenges associated with developing a deep UV laser is that the highly mature, blue-emitting InGaN system is not up to the task, due to its smaller bandgap. Stretching emission from the blue to UV demands the addition of aluminium to GaN, and this increases the in-plane lattice mismatch between the III-nitride and the most common substrate, sapphire. Material quality degrades, with imperfections dragging down the quantum efficiency.
On top of this, it is difficult to design a deep UV laser diode. One challenge is to develop a structure that provides sufficient optical confinement. Judged in these terms AlGaN is not ideal, with changes in the aluminium content producing small variations in refractive index.
Another concern is that there is yet to be an experimental report of a switch in polarisation from the transverse-electric to the transverse-magnetic mode during the onset of stimulated emission. If the polarisation mode switches, this has a big impact on device design, because the transverse-magnetic mode leaks light deeper into the absorptive p-region, due to its broader beam profile.
Finally, p-doping in UV devices is far more tricky than its visible counterparts, because the activation energy of the magnesium acceptor is considerably higher in aluminium-rich AlGaN than it is in GaN. The upshot is an insufficient hole density in the active region that hampers stimulated emission.
Solving all these issues simultaneously is very challenging. Consequently, many research groups start by putting aside issues related to p-doping and focus on the rest. By taking this approach, their first goal is to produce optically pumped deep UV lasers, preferably with short wavelengths and low thresholds.
Pioneering this approach is a team from Kogakuin University. In 2004, this group reported the first deep UV III-nitride laser. Formed on a SiC substrate, this optically pumped, 241.5 nm laser has a very large threshold of 1200 kW/cm2.
The next milestone in deep UV laser development came in 2011, when a partnership between Palo Alto Research Centre and the US Army Research Laboratory announced the demonstration of an optically pumped 267 nm laser with a threshold power density of just 126 kW/cm2. One of the big differences between this laser and that produced at Kohgakuin University is the choice of substrate "“ it had been switched from SiC to bulk AlN. Reasons for preferring the latter include a low dislocation density "“ it is around 104 /cm2 "“ and a similar lattice constant and thermal expansion coefficient to that of aluminium-rich AlGaN. Thanks to these merits, there is a low dislocation density in the AlGaN heterostructures.
During the last few years, more groups have had success with AlN substrates. Low thresholds have been obtained by optical pumping while pushing the lasing wavelength ever shorter, to reach to 237 nm. These accomplishments have highlighted the AlN substrate as a technically promising platform for demonstrating and developing the deep UV laser diode.
However, there are issues associated with the AlN substrate. Supply is limited, costs are high, and it is only available in small sizes. What's more, the current manufacturing process for making it introduces carbon impurities, which is a hindrance for UV devices, because they absorb light in this spectral range.
What about the alternatives? SiC is not ideal, due to relatively high costs that hamper manufacture. Far cheaper is sapphire, but it is plagued by large lattice and thermal expansion mismatches with AlN. These differences can lead to dislocation densities of over 1010 /cm2, unless special growth processes are adopted. And such a high level of imperfections is a major concern, because the internal quantum efficiency of a deep UV emitter is generally inversely proportional to the density of the dislocation-related, non-radiative recombination centres.
The crux of the matter, however, is this: Is a dislocation density as low as that of the AlN substrate essential for obtaining a high internal quantum efficiency and low-threshold lasing? Perhaps not. A detailed photoluminescence study by a partnership between Meijo University and Nagoya University has shown that despite a dislocation density of low 109 /cm2, the internal quantum efficiency for aluminium-rich AlGaN multiple quantum wells can exceed 50 percent. Note that these samples were grown on an AlN-on-sapphire template.
This finding is encouraging, given that similar or lower dislocation densities for AlN-on-sapphire templates have been realized by a mainstream growth technique, MOCVD. A common approach with this deposition technology is epitaxial lateral overgrowth (ELO), where AlN layers are re-grown on patterned seeding AlN templates. However, this technique is costly, takes considerable time and leads to uneven surfaces. These issues stem from a process that involves lithography, etching and a re-growth process that attempts to form over the patterned templates a coalesced, flat layer that is several microns thick.
One common MOCVD modification is pulsed atomic layer epitaxy (PALE). Nitrogen and aluminium precursors are supplied in a pulsed mode, in a manner that allows aluminium atoms sufficient time to mobilize on the epitaxial surface. This approach is simple, but increases the growth time and thus the cost.
Figure 3. (a) Laser emission spectra and (b) spectral integrated intensity and linewidth of the 249 nm laser versus pumping power, using 193 nm excitation.
Some groups have united the two, combining ELO and PALE to accelerate coalescence over patterned templates. Additional attempts to improve material quality have involved growth temperatures exceeding 1200 °C, because this increases aluminium atom mobility. This has been employed independently, and jointly with the ELO and PALE, to produce heterostructures with a low-dislocation density and a smooth surface. High-temperatures have their downsides, however. Reaching high temperatures requires a special reactor configuration, and these growth conditions can cause considerable thermal stress and cracks, due to the large thermal expansion mismatch.
Our goal at Georgia Tech has been to develop an MOCVD growth process for an AlN-on-sapphire template that is simple, efficient and yields a dislocation density of low 109 /cm2. To trim costs and save time, we avoid the ELO and PALE technologies.
We were also keen to see if lower temperatures could be used for AlN growth. Experiments had previously been carried out at 1100-1500 °C by teams from Meijo University and Linköping University, using an incremental step of 100 °C. This step is too large, however, because it prevents an insight into the variation in material quality with small changes in temperature "“ and it ultimately fails to uncover the optimum temperature.
With our study, growth temperatures for the AlN template were varied from 1050 °C to 1250 °C using an incremental step of just 18 °C. This led to a very interesting discovery. At temperatures exceeding 1212 °C material quality deteriorates − the surface gets rougher, surface bunching begins, and the linewidth in the X-ray diffraction scan increases (see Figure 1). We also found that high-quality AlN templates can probably be produced with growth temperatures as low as 1100 °C.
Figure 4. (a) Laser emission spectra and (b) spectral integrated intensity and linewidth versus pumping power densities of the 256 nm laser. Measurements taken at room-temperature, using 193 nm excitation.
Using this temperature, we optimised our growth conditions. We discovered that the best results came from a relatively simple template, formed from continuous growth of three layers: a 15 nm AlN buffer layer, a 50 nm AlN intermediate layer, and a 3400 nm AlN template layer. These were grown at 930 °C, 1130 °C, and 1100 °C, respectively, on a c-plane planar sapphire substrate. Encouragingly, deposition involved a relatively high growth rate of 2.30 µm/h, realised with a growth efficiency of 2200 µm/mol that indicates a low degree of parasitic reactions.
Scrutinising the template with conventional inspection techniques provided proof of good material quality. An atomic force microscopy scan over a 1 µm2 area uncovered well-defined terraces and determined a root-mean-square roughness of just 0.07 nm (see Figure 2), while transmission electron microscopy revealed a total threading dislocation density of 2à—109/cm2.
Equipped with these templates that have a dislocation density that we targeted, we grew pseudomorphic AlGaN heterostructures for an optical pumping experiment. They contained an AlGaN grading waveguide layer, five periods of AlGaN multi-quantum wells designed for laser emission at 250 nm, and an AlGaN cap for surface passivation. The composition and thickness of the AlGaN layers were optimized to enhance optical confinement and thus reduce threshold.
Following growth of the heterostructures, wafers were scribed, either by laser or hand, and cleaved into Fabry-Pérot laser bars. Laser scribing led to smoother facets. These chips were not given a high-reflectivity coating, and had a reflectance in the UV region of around 20 percent.
We formed a 249 nm laser using laser scribing that had a threshold of 95 kW/cm2 (see Figure 3). We also produced a 256 nm laser, using hand scribing, that had a threshold 61 kW/cm2 (see Figure 4). Both these thresholds are low values for this spectral range. The smoother facets resulting from laser scribing produced a higher slope efficiency and smaller linewidth. For both spectra, stimulated transverse-electric polarised emission dominated optical output.
To design high-performance, deep UV lasers operating at shorter wavelengths, it is imperative to know the spectral range where the dominant lasing mode switches from transverse magnetic to transverse electric. This transition is governed by the valence band structure of AlGaN. The transverse electric mode dominates when lasing results from a transition from the conduction band to a heavy hole band that sits at the top of the valence band. To reach shorter emission wavelengths, the aluminium content in AlGaN must increase, and this shifts the split-off hole band closer to the heavy hole band. The switch in lasing from the transverse-electric to the transverse-magnetic mode occurs when the topmost band changes from the heavy hole band to the split-off hole band. Note that we are the first group to report this switching of stimulated emission from the transverse-electric mode to the transverse-magnetic mode for lasers grown on the same type of substrate.
Our study shows that to ensure that the transverse-magnetic mode dominates lasing, lasers must emit at shorter wavelengths than those of our initial structures. One way to do this is to increase the aluminium composition in the active region, but this reduces optical confinement. So we prefer to decrease the composition contrast between the grading layer and active region. This diminishes the net interface polarization charge, increases the transition energy in the multi-quantum well, and drives emission to shorter wavelengths.
Efforts in this direction have led to a portfolio of lasers emitting at 243 nm and below. We have produced structures emitting at 239 nm, 242 nm, and 243 nm with laser thresholds of 280 kW/cm2, 250 kW/cm2 and 290 kW/cm2, respectively (see Figure 5). All these lasers were scribed by hand, so they didn't have the smooth facets that ensure the narrow linewidth of our 249 nm laser. Nevertheless, thresholds are lower than those of lasers at similar wavelengths that were reported by other groups and are grown on bulk AlN substrates.
Our lasers operate as we intended, with spectra dominated by transverse-magnetic, polarized stimulated emission (see Figure 6 for details). When operating above threshold, the difference in peak wavelength between the two modes is negligible. This reveals a minimal energy separation between the heavy hole and split-off hole bands and those have crossed over in our lasers.
Although we have focused on the edge-emitting laser in our discussion in this feature, this is not the only UV laser that could serve many important applications. There is also the deep UV VCSEL, which promises high-speed modulation, good beam quality, and great control of the production process. Progress with this vertical emitter is lagging behind its edge-emitting cousin, but we have recently made some important progress, using optical pumping to demonstrate the onset of deep UV stimulated emission from the surface of AlGaN heterostructures grown on AlN-on-sapphire templates (see Figure 7). Key milestones ahead of us and our peers include the development of high-reflectivity, high-conductivity distributed Bragg reflectors that could lead to the fabrication of the first electrically injected deep UV VCSEL.
While that may be some way off, we are undeniably making good progress, by optimizing growth processes and improving material and structural quality. This has enabled us to demonstrate low-threshold deep UV lasers, the switching of laser emission from the transverse-electric to the transverse-magnetic mode, and report the first observation of the onset of surface stimulated emission from III-nitride heterostructures on a sapphire substrate (see Figure 7).
What is needed now is to address a number of technological barriers that are preventing the demonstration of a deep UV laser that operates via electrical injection, rather than optical pumping. The good news is that many teams are working on this goal, both in the US and in many institutions overseas. Hopefully these efforts will deliver a major breakthrough in p-doping of aluminium-rich AlGaN, as today this is a major barrier to electrically driven, deep UV lasers.
