It is far more challenging to make a bright, cheap ultra-violet LED than one emitting in the blue. But success is promised with a transparent contact layer, reflective electrodes, photonic structures and growth on silicon, says Hideki Hirayama from RIKEN.

The performance of the nitride-basedLED continues to improve, opening-up new, lucrative markets. Billion dollar annual revenues began for this class of chip when it lit the keypads and backlights of mobile phones. More recently, unit shipments have surged as this solid-state source has seen widespread deployment in TVs, tablets and computer screens, and now it is starting to be deployed in lighting, in replacements for incandescent and fluorescent lamps.

Figure 1. The ultraviolet spectrum can be sub-divided into the UVA, UVB, UVC and vacuum UV ranges. Emission at 270 nm can be used for sterilization, due to the damage that it can cause to DNA 

One of the next big challenges for the LED is to replicate its performance in the blue – which can be converted to white by combining the chip with a yellow phosphor – in the deep ultra-violet (DUV) region of the optical spectrum, which is wavelengths of about 280 nm and below (see Figure 1). A solid-state source emitting in this spectral region could provide a replacement for mercury lamps used for sterilization, which are not ideal, because they are not monochromatic and they generate heat. A UV LED chip emitting monochromatic radiation at 270 nm, the ideal wavelength for sterilization, could address both these issues and find deployment in the home, where it would be used in refrigerators, water and air purifiers, air conditioners, foot curing systems, and vacuum cleaners. What’s more, a UV LED could be used to aid the transportation of foods; help to make bank notes more difficult to forge; and be deployed in sterilization equipment in hospitals and large systems, such as water tanks and air conditioners found in office blocks.
There are also other commercial opportunities for ultraviolet, solid-state sources. They could be used for medical treatments, such as skin care; enable high-speed dissociation of pollutant materials; play a role in high colour-rendering illuminations; and form the heart of a new generation of high-density optical storage devices.

But to make significant inroads into every one of these promising markets will require a significant improvement in the efficiency of UV LEDs operating at around 280 nm and below (see Figure 2). Back in 2010, several groups announced efficiencies for these DUV LEDs of more than 1 percent, and recently values of 5-14 percent have been reported (see Figure 3). If this level of improvement is maintained, the DUV LED market could take off in 2015.

Figure 2. The efficiency of LEDs emitting in the deep UV is significantly less than that for the blue

Figure 3. Some of the leading deep UV developers, including US firms Sensors Electronic Technology and Crystal IS, plus UV Craftory and RIKEN from Japan, reported EQE results at the International Workshop on Nitride Semiconductors in 2012 and ICNS 2013

Our team at RIKEN, a research institute just outside Tokyo, Japan, is one of the leading developers of these devices. Working in partnership with Panasonic, we are developing 270 nm LEDs that will be launched on the market that combine efficiency in excess of 2 percent with a lifetime of more than 10,000 hours.

Results in our lab for this class of LED include an external quantum efficiency (EQE) of 7 percent. This compares to reports at the International Conference on Nitride Semiconductors in 2013 (ICNS-10) of an EQE of about 7 percent for Crystal IS’ DUV LED that was fabricated on a single-crystal AlN substrate, and efficiencies of 14 percent and 11 percent from UV Craftory and Sensor Electric Technology Inc. So, at this stage, we are slightly behind the state-of-the-art values for external quantum efficiency. However, we have a program in place that could enable us to overtake the leaders.

Just reaching 7 percent EQE was not easy, and required improvements to many aspects of the device, including the internal quantum efficiency (IQE), electron injection efficiency and light-extraction efficiency – it is the product of these three that determines the EQE.
Now we are taking radical steps in LED design, such as the formation of the device on an array of AlN hexagonal pillars, to take device EQE to double-digit efficiencies. In addition, we have started a programme to slash the cost of DUV LEDs by growing the devices on silicon substrates. 

Increasing internal efficiencies

Between 2006 and 2010, we focused our efforts on increasing the internal quantum efficiency (IQE) of our devices. When we started, this was below 1 percent, and it has been increased through improvements in material quality. The key has been the development of a low threading-dislocation density AlN buffer on sapphire, which is produced with ammonia pulse-flow multi-layer growth. With this deposition technology, we have formed AlN layers with atomically flat surfaces and threading-dislocation densities of just 3×108 cm-2.
Diminishing this threading-dislocation density increased the IQE from the AlGaN quantum well to over 60 percent. Even higher values are possible with InAlGaN wells: We estimate that they can produced an IQE in excess of 80 percent, thanks to indium segregation effects in the quaternary alloy. Segregation causes carrier localisation, and this supresses non-radiative recombination. Further improvements to DUV LED performance have resulted from increases to the electron injection efficiency through the introduction of a multi-quantum barrier, electron-blocking layer. This electron-blocking structure is especially beneficial for DUV LEDs emitting below 250 nm.

Despite these significant improvements, devices still produced a low output power due to poor light extraction efficiency. There are two reasons for this: All the light that exits the quantum well towards the top surface of the chip is absorbed by the p-GaN contact layer; and much of the light that exits towards the substrate is reflected back into the device, due to significant refractive index differences at both the sapphire/air and AlN/sapphire interfaces (see Figure 4). A result of all of this is that light extraction efficiency is restricted to just 8 percent.

Figure 4. To enable efficient light extraction, deep-UV LEDs need to have a markedly different architecture from their blue cousins. The conventional p-type contact, GaN, absorbs UV light, which is partially reflected back into the device at AlN/sapphire and sapphire/air interfaces

Figure 5. Light extraction in deep UV LEDs can be improved by: switching from p-GaN to p-AlGaN, a transparent contact layer; using a highly reflective mirror for this wavelength range; and forming the device on connected AlN pillars that channel light out of the chip

To address the absorption of light by the p-GaN contact layer, we switched to a transparent p-AlGaN contact (see Figure 5). Making this adjustment in isolation can increase light extraction efficiency to more than 40 percent, but at the expense of hole injection efficiency. When high-aluminium-content p-AlGaN is used, the deep acceptor level of the magnesium dopant can limit the hole density to less than 1015 cm-3. This meant that when we fabricated 265 nm DUV LEDs with a high-aluminium-content p-AlGaN contact layer, the EQE of the device was no better than its predecessors. However, the EQE was not significantly worse either, and as we shall see in the next paragraph, that held the key to a generation of brighter devices. To first determine what the appropriate compositional wavelength of p-AlGaN is, we varied this between 290 nm and 270 nm, which corresponds to aluminium compositions of approximately 48-60 percent. This short study revealed that p-AlGaN with aluminium composition as high as 60 percent is useful for the contact layer of a DUV LED (see Figure 6).

Figure 6. The aluminium composition in AlGaN can be reduced without a large impact on EQE. This indicates that an AlGaN layer, which is less absorbing than GaN, could lead to higher efficiencies in a modified device. The downside of p-AlGaN is a lower hole density, but this can be compensated with a better electron blocking layer. When researchers at RIKEN followed this path, their devices delivered far higher EQEs than before  

Validation of this choice of aluminium composition came from another set of experiments, where we fixed the compositional wavelength of p-type AlGaN at 270 nm, corresponding to an aluminium composition of 60 percent, and changed the emission of the quantum well from 265 nm to 282 nm. This revealed that the 270 nm p-AlGaN contact layer is transparent for emission at wavelengths longer than 280 nm. Our final step in this particular effort was to address the aforementioned reduction in the hole density that resulted from the switch from p-GaN by p-AlGaN. We compensated for this by adding a higher electron-blocking structure: A higher-aluminium-content AlGaN, multi-quantum barrier electron-blocking layer. The other issue that we faced – that the substrate reflects light back into the device – was addressed by introducing a new device architecture with a highly reflective p-electrode mirror. Why did we do this, rather than simply turning to a transparent electrode? Well, although that is a good approach for the blue LED, thanks to the availability of transparent ITO, no material can fulfil that role in the deep UV.

One candidate material for making the mirror is aluminium, which has a reflectivity of 92 percent at UVC wavelengths (280 nm to 100 nm). However, aluminium cannot form a p-type ohmic contact on p-type AlGaN, so we first insert a sub-nanometre-thick layer of nickel. Reflectivity with this combination is 64 percent, which is far superior to the conventional pairing of nickel and gold, which produces a reflectivity of just 30 percent. Note that another option for our highly reflective p-contact is a mesh-pattern electrode. If we decide to go down that road, we will need to insert a p-type, AlGaN hole-spreading layer beneath the electrode, which could be formed with a short-period superlattice. This structure could also help to combat an increase in forward voltage of about 5 V, which resulted from replacing p-GaN with p-AlGaN.

Our efforts at switching from p-GaN to p-AlGaN and adding a reflective contact were rewarded. With the new blocking layer in place, switching to the p-AlGaN contact layer increased EQE by about 50 percent, and a further efficiency gain of 70 percent resulted from the transparent p-AlGaN contact layer and highly reflective p-electrode (see Figure 7).

Figure 7. To take the EQE of its deep UV LEDs from about 3 percent to 5.5 percent, RIKEN’s researchers used p-AlGaN contact layer and switched from a Ni/Au contact to Ni/Al 

Increasing light extraction

One route to making further improvements in LED light extraction is to introduce photonic nano-structures, such as two-dimensional photonic crystals or moth-eye patterns. We are investigating this possibility and introducing a connected-pillar AlN buffer beneath our devices. An array of AlN pillars should increase device efficiency by allowing light to propagate vertically along the array, and it should also enhance material quality, because the threading dislocation in the pillars should be quite low.

Our efforts in this direction begin by taking patterned sapphire substrates and growing connected, hexagonal-shaped AlN pillars on them by controlling the V/III ratio and growth temperature (see Figure 8). To reduce the threading dislocations in these structures, an ammonia pulse-flow method is employed in the initial stage of AlN pillar growth. Once the array is formed, we reduce the V/III ratio, so that the pillars merge to form a flat surface. The threading dislocation density in the pillars is low, according to cross-sectional images provided with a transmission electron microscope.

Figure 8. The array of hexagonal AlN pillars was formed by ammonia pulse-flow growth on patterned sapphire

These connected pillars have formed the foundation for 265 nm LEDs that deliver a continuous output of over 5 mW and have an EQE of a few percent. This work is still in its infancy, and we know that the external quantum efficiency of our devices will be far higher when we optimise the surface roughness of a connected pillar AlN buffer. We expect that by combining a transparent p-AlGaN contact layer with a connected-pillar AlN buffer, we will be able to increase the external quantum efficiency of our DUV LEDs to several tens of percent.

Silicon foundations

Silicon is a very attractive substrate for UV LEDs. It is cheap, it is available in large sizes and it is easy for it to be separated from the LED epitaxial structure with wet chemical etching. That step is essential, because silicon absorbs the device’s emission, and once it has been removed, it is possible to fabricate vertical LEDs. Such structures can also be made with LEDs formed on sapphire, but the substrate has to be removed with laser lift-off, and this damages the material.

We have developed a fabrication process for making DUV LEDs on silicon (see Figure 9 for the device structure). It consists of AlGaN LED growth on silicon, wafer bonding to a heat sink, silicon wafer removal, dry etching of AlN and fabrication of a mesh electrode. With this approach, the biggest challenge is to grow crack-free AlN on silicon, due to the significant difference in the thermal expansion coefficient of these two materials. One way that we have addressed this is to fabricate an AlN buffer on silicon by ammonia pulse-flow growth. Within 1 µm of buffer growth, we obtained a crack-free film with a low threading dislocation density. This provided a platform for the growth of a range of DUV LEDs on silicon that featured an InAlGaN quantum well and produced emission from 284 nm to 300 nm.

Figure 9. Moving from a sapphire substrate to one made with silicon aids the development of vertical LEDs. A damaging laser lift-off process has to be used to separate the epitaxial stack from sapphire, but with silicon a wet chemical etching approach can be employed to yield a pristine chip 

We have also explored epitaxial lateral overgrowth for the deposition of crack-free AlN on silicon. Again, we began by depositing an AlN layer on silicon using an ammonia pulse-flow method, before fabricating a stripe pattern on this binary and then growing the ELO layer. With this approach, we realised a low threading dislocation density, 3 µm-thick AlN buffer layer on silicon and demonstrated a series of DUV LEDs featuring AlGaN quantum wells emitting between 256 nm and 278 nm.

We are now planning to remove the substrate and develop vertical DUV LEDs. This should lead to devices with low operating voltages and high light extraction that will help the UV LED industry to serve many applications that would benefit from an affordable solid-state, portable source of light.

Further reading
Sachie Fujikawa et al. Appl. Phys. Express 4061002 (2011)
Takuya Mino et al.  Appl. Phys. Express409210 (2011)

QCLs at RIKEN

In addition to RIKEN’s development of DUV LEDs spanning 220 nm to 350 nm, the leading research institute has a programme devoted to the development of terahertz quantum-cascade lasers.
One notable success is the record for the highest temperature, stable operation of lasers emitting in the 2-4 THz range. Thanks to the use of a novel quantum-cascade structure, the researchers fabricated a 1.9 THz device that can operate at 160K.
The team is also developing QCLs based on the pairing of GaN and AlGaN, which have the potential to cover the 5 Thz to 12 THz range and operate at relatively high temperatures. The GaAs-based material system is not ideal for operating at these frequencies, due to phonon interactions.

RIKEN’s researchers are the first group to report spontaneous emission by current injection from an AlGaN-based QCL structure. Their device has a 150 period active region based on GaN and Al0.2Ga0.8N and emits at 1.37 THz.