AlInN Mirrors Spur VCSEL Progress
Our community has produced several types of nitride device that offer better performance characteristics than their GaAs equivalents. GaN HEMTs can produce far higher power densities, while nitride LEDs can emit at shorter wavelengths that allow single chip, high-brightness white-light sources.
However, there is a gaping hole in the nitride portfolio – a GaN VCSEL. Such a device does not exist because it is very difficult to produce highly reflecting mirrors that operate in the blue-violet spectral range with good electrical conductivity. But a GaN VCSEL could offer several desirable attributes over a GaN edge-emitting laser including a lower threshold current, a circular emission profile that simplifies coupling into optical fibers and a reduced price, thanks to higher yields and on-wafer testing.
Despite these advantages, it is highly unlikely that a GaN VCSEL would replace an edge-emitter in today s Blu-ray and HD DVD players. That s because it would need to be in the form of an array to deliver sufficient power, and major changes would be needed in the technology employed for reading/writing functions. However, this device could enjoy success in a variety of other applications. Xerox has built some very high-quality 2400 dpi printers with infrared VCSEL arrays, and even higher resolutions may be possible by switching to GaN-based equivalents. These shorter-wavelength arrays could also serve in small laser-based projectors, and single GaN VCSELs could offer an alternative to their red-emitting cousins employed in optical mice.
All VCSELs feature a pair of high-quality mirrors with a reflectivity of at least 99%, which compensate for the small active volume that is a consequence of this emitter s vertical design. The mirrors for conventional arsenide-based VCSELs are relatively easy to manufacture because they can be produced by making distributed Bragg reflectors (DBRs) from alternating layers of GaAs and AlAs. These two materials have good conductivity, very similar lattice constants and a significant refractive index contrast, which means that high-reflectivity defect-free mirrors that are capable of good carrier injection can be produced from an acceptable number of mirror pairs. Nitride-based VCSELs, however, can t call on an equivalent pair of nitride materials that are so well matched for making mirrors.
Up until now, research into nitride-based DBRs has focused on AlGaN and GaN, but it has proved impossible to produce mirrors of sufficient quality with this pair. The combination has a good refractive index contrast, but this comes at the expense of a lattice mismatch of up to 2.4%, which causes the epiwafer to crack and prevent laser manufacture. Sophisticated strain-engineering solutions can overcome this cracking, but even free-standing GaN substrates can t prevent the high dislocation densities that result from plastic relaxation.
The mirrors employed in an infrared emitting VCSEL are not just there to reflect light back into the device – they also transport carriers from the metal contacts to the central part of the active region. Injecting these carriers is relatively easy for arsenide-based VCSELs, but it is a major stumbling block when designing a nitride equivalent. The problems can be bypassed with intracavity contacts, but these can hinder lateral current spreading, which is needed to transport the carriers to the device s center.
A high current must be confined within an area of a few microns to produce the current density required for lasing. In an arsenide VCSEL, this confinement is produced by partial chemical oxidation of an AlAs layer inserted in the active region. However, this process is incompatible with GaN-based VCSELs, because AlN increases the structure s strain and defect density, and it can t be oxidized. Current confinement from selective oxidation or etching of a nitride alloy could be a solution, but this requires sophisticated processes, such as photochemical etching and electrochemical oxidation.
To overcome some of the difficulties associated with AlGaN-based mirrors, our team headed by Nicolas Grandjean from the Ecole Polytechnique Fédérale de Lausanne (EPFL), Switzerland, has developed microcavities based on the Al1-xInxN alloy (figure 1). This material has a perfect lattice match with GaN at an indium content, x, of 17%, and a far higher refractive index contrast than that of AlGaN-based mirrors (figure 2).
AlInN is difficult to work with because it is thermodynamically unstable. The Al–N and In–N covalent bonds have a large mismatch, which drives phase separation and compositional inhomogeneity. Incorporating indium into the AlN matrix is very challenging because InN prefers growth temperatures below 600 °C and AlN is grown at 1100 °C.
Despite these difficulties, we have developed an MOCVD process that can routinely produce high-quality Al1-xInxN layers with an indium content of less than 30%. This is done using growth temperatures between 800 and 850 °C, a pressure of less than 100 mbar and nitrogen as a carrier gas that prevents indium desorption. This growth temperature window prevents the degradation of AlInN s crystalline quality, which would occur at lower temperatures, and allows sufficient indium to be incorporated, which is not possible at higher temperatures. Reducing the growth rates to 0.2 µm/h prevents surface roughening and the formation of indium clusters, and there is no variation in indium content, according to X-ray diffraction measurements.
Thanks to this lattice-matched growth, we can produce AlInN/GaN DBRs on sapphire substrates that are free from any additional dislocations. These structures typically have 40 mirror pairs and 99.4% reflectivity, and microcavities built from them have a quality factor (the cavity mode s wavelength divided by its linewidth) of 2800. This key figure of merit for any VCSEL – which is inversely proportional to its optical losses – is a big improvement over the previous best value for AlInN microcavities of 800, but it is still considerably less than the highest figure for planar GaAs microcavities – 11,000.
As mentioned, VCSEL mirrors have a second purpose – providing electrical injection into the active region. Unfortunately, the Al1-xInxN and AlyGa1-yN layers hamper electron and hole transport due to the large band-offsets between these alloys. In addition, the layers with high aluminum content have a low conductivity that leads to a fairly high series resistance and poor vertical injection through the DBRs.
Switching to dielectric DBRs is not a solution because this structure is not conductive. However, success is possible by turning to a VCSEL design that injects carriers from intracavity contacts, which are metallic rings inserted between the mirrors and the active region (figure 3). Producing such a device is not straightforward because sufficient hole injection for lasing cannot be produced at acceptable levels of optical loss from either tunnel junctions or semitransparent contacts, such as ITO or thin metallic films. It is also tricky to implement a ring-shaped geometry that prevents contacts from reabsorbing the emitted light, because the nitride s poor hole transport prevents carriers moving from the rings to the center of the active region. This is a major concern because lasing requires high current densities in the very center of the device where optical confinement is optimized.
However, we can get round all of these problems by replicating the current-confinement technology found in arsenide VCSELs. In our case, we oxidize part of an AlInN layer that is inserted close to the quantum wells. This confines the carriers in the cavity s non-oxidized region, which enables the use of electrical contacts with a ring geometry.
The first device that we built with this technology was an LED that had a circular emitting area with a 3 µm diameter. It had a current density of 20 kA/cm2 and an output power of 400 kW/cm2 (figure 4). We followed this up with a crack-free VCSEL structure on sapphire, which featured an active region with three InGaN/GaN quantum wells and a hybrid microcavity – a 40 pair lattice-matched AlInN/GaN bottom DBR and a 16-pair SiO2/SiN top DBR mirror. A fully epitaxial structure was not used, even though it would have offered better optical quality, because processing would have been challenging. Hole injection through a nitride DBR is not possible, so mirror etching would be required before the contacts were defined, but this step demands an unattainable precision without an etch-stop layer.
Before we processed our wafers, we measured the microcavity s quality factor and found that it peaked at more than 3000. This high value enabled a lasing threshold for pulsed optical pumping of just 300 kW/cm2 at room temperature and allowed us to observe lasing under CW excitation with a power density of 10 kW/cm2 at temperatures of more than 50 K (figure 5). The emission from this structure is a high-quality singlemode beam with a linewidth of just 0.025 nm and a divergence angle of 6° (figure 6).
We then measured the electrical characteristics of our lasers, which have a thinner cavity than our LEDs. This adjustment increases the device s resistance and cuts its typical maximum current density. Under pulsed operation the p-type contact of these VCSELs fails at less than 10 kA/cm2. Unfortunately, this is just below the theoretical value for the onset of lasing. However, we did see some electroluminescence from this device, which had a 0.3–0.35 nm linewidth that corresponds to a quality factor of 1200–1600 (figure 7).
The values for the quality factor are lower than those obtained on unprocessed wafers, which we attribute to "edge effects". These impact the deposition of the top dielectric DBR over the structured surface, which takes place after mesa etching and the deposition of the electrical contact. Nevertheless, these quality factors are the highest reported values for a nitride VCSEL under current injection and we believe that lasing is within our grasp. Typical threshold currents for edge-emitting laser diodes on sapphire substrates are a few tens of kiloamps per square centimeter. Taking into account the vertical geometry of our VCSELs, we expect to reach the lasing threshold at a similar current density when the microcavity quality factor approaches 3000.
We have definitely made a great deal of progress, but there are still several obstacles to overcome before we make the first GaN VCSEL that delivers lasing via electrical injection. Refinements in the process used to form the intracavity ring contacts are needed because the existing approach seems to reduce the microcavity s quality. In particular, deposition of a more uniform top dielectric DBR is needed to cut optical losses, boost the cavity s quality factor and ultimately decrease the threshold current density to a sustainable value. A switch from sapphire to free-standing low dislocation GaN substrates should complete the list of improvements required, as this will increase VCSEL quantum efficiency and thermal management capability.