Technical Insight
Getting the GaN VCSEL to market
The output power of the GaN VCSEL hits a new high with the introduction of epitaxial lateral overgrowth
BY TATSUSHI HAMAGUCHI FROM SONY CORPORATION
GaN is a great material for making optoelectronic devices. Global sales of GaN-based blue, green and white LEDs are netting billions of dollars every year, and there is also a substantial market for in-plane lasers emitting in the blue, blue-violet and green. This makes the GaN VCSEL the only obvious omission in the GaN optoelectronic portfolio. But if this device can follow in the footsteps of that made from GaAs, it should enable the production of a class of GaN laser that combines a very low threshold current with the capability to operate at high frequencies and form an emitter array.
Armed with these attributes, GaN-based VCSELs are destined to replace conventional LEDs and lasers as light sources in many applications, including optical storage, laser printers, projectors, displays, solid-state lighting, optical communications and biosensors. And if green and blue forms of this device are united with red-emitting GaAs VCSELs, this could spawn incredibly small, wearable projectors and high-power light sources for full-colour displays.
Figure 1. In a VCSEL (right), light travels in the vertical direction to the active region, while in an in-plane laser diode (left) it travels parallel to the active region. This key difference comes from their structures. In a VCSEL, all layers, mirrors and active regions are parallel to each other, so that the direction of photon propagation is vertical to those layers.
However, fulfilling this dream will not be easy. Peruse through the scientific literature and you'll soon realise that making a GaN-based VCSEL is far harder than producing a GaAs-based cousin. Multiple issues plague the production of mirrors and cavities, and it is tough to realise sufficient current confinement in the active region to ensure lasing. However, significant breakthroughs are being made, including those by our team at Sony. We are taking output powers to new highs by turning to epitaxial layer overgrowth for the production of high-quality active regions close to the mirrors.
Many of the challenges associated with GaN VCSEL production are absent with GaAs because it can be paired with AlGaAs, a ternary with a negligible difference in lattice constant. Together, these arsenide alloys can form low-defect-density, distributed Bragg reflectors (DBRs) that sit either side of the active region and ensure sufficient optical gain for lasing. What's more, because these materials can be made conductive by doping, carriers can be injected into the active region via deposition of an electrode on each DBR.
Another attribute of the GaAs-based material system is that partial oxidation of AlAs can transform it from a semiconducting to an insulating form. So, when AlAs is inserted next to the active region, it can be oxidised to create a current-confined aperture. This structure is essential, because it enhances induced emission, a pre-requisite for lasing.
Figure 2. In a GaAs-based VCSEL, the active region is sandwiched between two DBRs. All of the layers are formed by epitaxial growth, permitting both DBRs to be placed near the active region. AlAs can be partially oxidized to leave a current convergence spot that confines carriers.
With GaN, the situation is markedly different. First and foremost, it is very difficult to produce semiconductor DBRs. When GaN is paired with AlGaN "“ an approach that is similar to that employed in GaAs-based VCSELs "“ cracks appear, due to the large lattice mismatches associated with changes in aluminium composition. One promising solution is to pair GaN with Al0.8In0.2N, which has a very similar lattice constant. This can ensure the production of crack-free mirrors. However, the price to pay for this success is a low throughput. Reports suggest that the growth rate for AlInN is so low that it would take 12 hours to form a 40 AlInN/GaN DBR with a peak reflectivity exceeding 99 percent. Such a long time is unacceptable for the growth of just one section of a commercial device.
Figure 3. Several groups, including a team from Nichia, have fabricated GaN-based VCSELs by thinning the GaN substrate to a few microns, prior to n-side DBR deposition. There are several papers on this type of VCSEL, for lasing wavelengths from the UV to the green. However, almost all were published before 2012. The commercial success of this method appears to be hampered by difficulties associated with polishing a GaN wafer to a thickness of just a few microns
Figure 4. At Sony, GaN-based VCSEL fabrication involves an epitaxial lateral overgrowth process. This assists the placement of the n-side DBR just a few microns beneath the active region, as it does not involve a severe process. Prior to this breakthrough, similar structures could only be formed with a delicate thinning process.
Even if such long times could be tolerated, there is still a major issue with the GaN-based DBR: severe thickness control is required. With AlInN and GaN, the small difference in refractive index between these two alloys produces a sharp reflectance spectrum. This is awkward, because a thickness variation in the DBR layer of just ±3 percent can shift the reflectivity peak out of the lasing wavelength and prevent lasing. As a ±3 percent fluctuation in growth rate is very likely in an MOCVD reactor for GaN growth, it is clear that low yields would plague mass production of GaN VCSELs with AlInN/GaN DBRs.
Based on this reasoning, we believe that dielectric materials are a better alternative to wide bandgap alloys for GaN-based VCSEL production. The likes of SiO2, Ta2O5 and SiN can be deposited by common methods, such as electron-beam deposition and sputtering, and there are no concerns relating to growth rates.
One advantage of taking this route is that it allows the pairing of two materials that have large refractive indices and produce broad reflectivity spectra. A thickness range in excess of 10 percent can then be tolerated, which is within the process capability of those conventional deposition methods named above.
If dielectric mirrors are to be a success, they must be incorporated into the fabrication process of the GaN VCSEL. This means that these mirrors must be positioned close to the active region, within a distance of a few microns. Failure to do this will prevent the device from lasing, due to considerable diffraction loss and material absorption in a cavity with a length of 10 µm or more.
At Nichia, researchers have tackled this challenge by thinning a GaN wafer, which contains an active region, to just a few microns thick, and then sandwiching it between two dielectric DBRs. As the thin wafer cannot be free-standing, it is bonded to a support wafer, such as silicon, prior to the thinning process.
Difficulties associated with this production process include the development of a highly sophisticated technique for thinning the wafer, and avoiding damage to the active region "“ according to reports from Nichia, damage within this region appeared after just 10 minutes of device operation. We think that this might be due to mechanical stress in the active region of the thinned structure. Note that Nichia has not published any papers on GaN VCSEL development since 2012, suggesting that the issues associated with this class of VCSEL were serious enough to stifle the project.
We are pursuing an alternative approach to wafer thinning that involves epitaxial lateral overgrowth, with dielectric DBRs forming masks for selective growth. This technique begins with deposition of dielectric DBR islands directly on the GaN substrate. These islands are engulfed in n-type GaN by epitaxial lateral overgrowth.
At Nichia, researchers have tackled this challenge by thinning a GaN wafer, which contains an active region, to just a few microns thick, and then sandwiching it between two dielectric DBRs. As the thin wafer cannot be free-standing, it is bonded to a support wafer, such as silicon, prior to the thinning process.
Difficulties associated with this production process include the development of a highly sophisticated technique for thinning the wafer, and avoiding damage to the active region "“ according to reports from Nichia, damage within this region appeared after just 10 minutes of device operation. We think that this might be due to mechanical stress in the active region of the thinned structure. Note that Nichia has not published any papers on GaN VCSEL development since 2012, suggesting that the issues associated with this class of VCSEL were serious enough to stifle the project.
We are pursuing an alternative approach to wafer thinning that involves epitaxial lateral overgrowth, with dielectric DBRs forming masks for selective growth. This technique begins with deposition of dielectric DBR islands directly on the GaN substrate. These islands are engulfed in n-type GaN by epitaxial lateral overgrowth.
Figure 5. With a maximum CW output of 1.1 mW at room temperature, Sony's VCSEL has broken the power output record for a GaN VCSEL. This 453.9 nm laser has an 8 μm aperture.
With this approach, seed crystals grow in the window area, where the GaN substrate is exposed between DBR islands. Further growth embeds the DBR islands in an n-type film, which provides the foundation for depositing the active region, p-type GaN layers and p-side DBRs.
The cavity, created by placing the n-side dielectric DBR just a few microns beneath the active region, is formed without the need for delicate processes, such as polishing. This is advantageous, because no undesirable mechanical stress is imposed on the active region. The epitaxial layer overgrowth process is uniform over the entire wafer, making it suitable for mass production.
Our VCSELs have set a new benchmark for optical output power of more than 1 mW. This result is very encouraging, and even higher output powers may follow through refinements to the epitaxial layer overgrowth process "“ remember that it was developed to reduce threading dislocations in GaN grown on sapphire, for the production of GaN-based in-plane lasers. Epitaxial layer overgrowth may also enable the growth of GaN VCSELs on cheaper substrates, such as silicon. This could drive greater deployment of these devices.
With this approach, seed crystals grow in the window area, where the GaN substrate is exposed between DBR islands. Further growth embeds the DBR islands in an n-type film, which provides the foundation for depositing the active region, p-type GaN layers and p-side DBRs.
The cavity, created by placing the n-side dielectric DBR just a few microns beneath the active region, is formed without the need for delicate processes, such as polishing. This is advantageous, because no undesirable mechanical stress is imposed on the active region. The epitaxial layer overgrowth process is uniform over the entire wafer, making it suitable for mass production.
Our VCSELs have set a new benchmark for optical output power of more than 1 mW. This result is very encouraging, and even higher output powers may follow through refinements to the epitaxial layer overgrowth process "“ remember that it was developed to reduce threading dislocations in GaN grown on sapphire, for the production of GaN-based in-plane lasers. Epitaxial layer overgrowth may also enable the growth of GaN VCSELs on cheaper substrates, such as silicon. This could drive greater deployment of these devices.
Controlling current flow
Another big challenge for GaN VCSEL developers is to realise lateral current control, so that carriers are confined. With GaN VCSELs this is not trivial because there are no equivalents of the conductive DBRs and AlAs layers found in GaAs-based cousins. The good news is that a team led by Tien-Chang Lu from National Chiao-Tung University, Taiwan, has shown that it is possible to control the injection of holes into GaN-based VCSELs by making a transparent electrode from indium tin oxide, and using SiO2 for insulating regions. Lateral current spreading results from placing ITO between a p-side dielectric DBR and p-type GaN, while current confinement at a limited part of the active region is realised from partially inserting a thin SiO2 layer between the ITO and p-GaN layers.
Care is needed when taking this approach because ITO could absorb enough light to prevent lasing. The trick is to suppress absorption to negligible levels by adjusting the vertical configuration in the device so that ITO is located at the null point of the resonating mode standing wave.
Another concern is electrical stability at the ITO/p-GaN interface. Although ITO is widely employed as a p-contact in GaN-based LEDs, the current density in these devices is hundreds of times lower than that expected in a VCSEL. But we have shown that ITO contacts are sufficiently durable in aging tests at current densities as high as 60 kA cm-2. This indicates that ITO can be used in GaN VCSELs.
An alternative approach, for example, pursued by a team from the University of California, Santa Barbara, is to use a tunnel junction to inject holes. With this architecture, a transparent layer is formed by highly conductive n-GaN laid over p-GaN. An appealing aspect of this technology is that it mirrors the approach used in GaAs and InP. However, because GaN has a far higher bandgap than GaAs, it is difficult to reduce the operating voltage of the GaN-based tunnel-junction. When GaN VCSELs incorporate this, the operating voltage is higher by 1 V or more, due to the large built-in voltage at the tunnel-junction. It is our understanding that to ensure reliable operation in a laser diode, the voltage drop at the tunnel-junction should be no greater than 0.1 V at the operating current density. So the tunnel-junction must be improved by an order of magnitude before it can be considered as a replacement for ITO in GaN VCSELs.
Multiple markets
Thanks to the recent breakthroughs, it is now appropriate to consider potential applications for the GaN VCSEL. There are many, as this device can span the UV through to the green, which is a spectral range that has not been reached with GaAs-based VCSELs.
At the shorter end of this range is the 375 nm VCSEL, which could be paired with ytterbium ions to deliver greater accuracy in chip-sized atomic clocks "“ they currently employ GaAs-based near-IR VCSELs and rubidium gas. At the slightly longer wavelength of 405 nm, the GaN VCSEL could replace the GaAs-based VCSEL in laser printers, where it would enable a finer resolution; while at 488 nm, the VCSEL could be used in bio-sensing applications; and in the green, it could expand the use of optical communication in plastic optical fibre. These low-cost waveguides have a high optical loss in the red and IR, but not in the green.
Even more exciting is the combination of the classic red GaAs VCSEL and its blue and green GaN variants. Working together, they can provide full-colour light sources for displays. Arrays allow an increase in the output of VCSEL sources, and a path to watt-class red-green-blue modules.
At far lower powers, these modules are good candidates for retinal-scanning wearable displays. As it is easier to maintain small peak powers with VCSELs than in-plane devices, this form of laser is ideal for keeping retinal projection systems safe for eyes. With an in-plane laser, the addition of a neutral-density filter ensures eye-safety; however, such a configuration drives up energy consumption, resulting in a need for higher-capacity batteries. This is highly undesirable, because it increases the dimensions of wearable displays and reduces the usability of the device.
Further opportunities for GaN VCSEL arrays include reinforcing functions of electronic devices, where single-beam scanning of in-plane lasers is currently used. For example, arraying a 405 nm VCSEL could speed optical storage and 3D printing.
Some of the other virtues of the GaN VCSEL should enable improvements in applications currently using 445 nm LEDs. They include visible light communication, where the higher frequency operation of the VCSELs provides a broader band; and the light source for car headlamps, where efficient operation under large current injection improves high brightness.
The market for VCSELs is already worth hundreds of millions of dollars, and analysts are predicting further growth. We believe that this could be spurred by the introduction of GaN VCSELs, following successful efforts to address issues that have held back the performance of this very promising device.
