By MARK DURNIAK and CHRISTIAN WETZEL from RENSSELAER POLYTECHNIC INSTITUTE
The solid-state lighting revolution is now upon us. Businesses and homeowners are trimming their electricity bills, and by 2030 an ever-increasing uptake of LED bulbs could lead to savings as high as $30 billion, according to the US Department of Energy.
Most consumers that are currently investing in solid-state lighting are buying bulbs that employ a highly efficient blue LED to excite a phosphor, which then re-emits photons of lower energy, usually in the yellow spectral region. Colour mixing of blue and yellow creates a white-emitting luminaire.
This approach has the merit of simplicity, but it is not ideal. One weakness is the inherent conversion efficiency ceiling, stemming from the Stokes loss in the down-conversion of high-energy blue photons to lower energy yellow photons. In addition, the colour temperature or ‘warmness’ of the white light is fixed, governed by the mix of phosphors.
These limitations can be lifted with luminaires based on the colour mixing of several LEDs, such as those emitting in the red, green and blue. With this approach, which can involve three or more LEDs, it is possible to construct a high-quality, colour-tuneable light source. The Philips HUE bulb is an example of this, which allows users to pick the colour they want by running an app on their smart phone.
Another advantage of the colour mixing approach is that it can achieve higher theoretical efficiencies than phosphor conversion. The efficacy of this form of lighting is then governed by the efficiencies of the LEDs.
Blue and red LEDs excel in this regard, but the same cannot be said for their green-emitting cousins. At modest currents, such as 20 mA, their efficiency is significantly inferior. And when the current is cranked up, they are more prone to droop, so their efficiency plummets faster.
These weaknesses have several origins, all associated with flaws in material characteristics. To produce a green LED, engineers use a similar growth process to that employed for making a blue emitter, but increase the indium content in the InGaN quantum well. Unfortunately, this increases the crystalline strain in the well, leading to the generation of crystalline defects that peg back the efficiency of the green LED.
This device is fabricated from the hexagonal, or wurtzite, phase of the wide bandgap material, which features strong internal fields. These fields cause a shift in wavelength as the current is increased, due to a phenomenon known as the quantum confined Stark effect. This is particularly severe in green LEDs.
As well as the shift in emission wavelength, there is a pulling apart of electrons and holes, which pile up on opposite ends of the well. This impairs the radiative recombination efficiency. Turning to narrower wells can bring the charge carriers closer together, but may have very damaging side affects associated with the increase in carrier density. Auger recombination is known to increase with the cube of carrier density, and if, as many believe, this is the primary cause of droop, turning to thinner wells will fail to improve the performance of a green-emitting LED at meaningful current densities.
So, as you can see, the green gap is a very tricky problem. Researchers are trying to solve it, however, with some taking radical steps that will allow a lowering of the carrier concentrations in the quantum wells. Since it is difficult to see how one could do this with polar material, groups are trying different approaches, such as growing devices on non-polar faces of hexagonal GaN. And at the NSF-funded Smart Lighting Engineering Research Center at Rensselaer Polytechnic Institute we are taking a similar, but distinct tack, by turning to the cubic phase of GaN.
Attributes of cubic GaN
One of the great strengths of the cubic phase of GaN is that it has no internal electric fields in the typical growth direction, and thus no quantum confined Stark effect. The benefits of this are twofold: bands no longer move with increasing current, so the emission wavelength is fixed; and the overlap of electrons and holes in the quantum wells increases, so these carriers are more likely to recombine radiatively. Removal of the electric fields also opens up the possibility to turn to wider quantum wells that enable efficient radiative recombination while lowering Auger recombination rates.
Another great virtue of cubic GaN is its bandgap: At 3.2 eV, it is 0.2 eV lower than that of the hexagonal phase. This difference means that cubic GaN has almost a 30 nm head start prior to growth for making long-wavelength emitters, which is a great benefit when aiming for green LEDs.
What’s more, cubic GaN theoretically has better transport properties than the hexagonal variant due to higher crystal symmetry. Holes are a necessary player in producing light from LEDs, but p-type hexagonal GaN is notorious for its low concentration of holes and their poor mobility. Since holes do not move around as freely as electrons, most of the emission from an LED comes from the last quantum well − the one closest to the p-side. Better hole mobility would increase its population in many quantum wells, culminating in more light from the device. At low hole concentrations, hole mobility in cubic GaN has been reported at 350 cm2 V-1 s-1 compared to less than 200 cm2 V-1 s-1 in hexagonal GaN.
We are not the pioneers of cubic GaN, and our contribution to this field is to increase the size of this material. We are following in the footsteps of the likes of the groups of Donat As at the University of Paderborn, and Tom Foxon and Sergei Novikov at the University of Nottingham – teams that produced cubic GaN layers and devices, starting mainly with 3C-SiC and GaAs cubic substrates.
The MBE growth technique that both of these groups have used is highly versatile, but limited to growth rates of the order of 100 nm/hr. This means that the deposition of an LED structure can take
10 hours or more, an impractically long time for LED manufacture, which is better served by MOCVD. Groups led by Heber Vilchis from Cinvestav and Shigefusa Chichibu from the University of Tsukuba have shown that this deposition technology can form cubic GaN on GaAs and 3C-SiC, respectively, but crystal size is small and phase purity poor. We are now addressing these shortcomings with an approach that offers production scalability, and begins with the most common form of silicon. By using this form, silicon (001), we have the potential to reach wafer sizes of 300 mm.
To address the well-known mismatch problem, we do not directly deposit a film of cubic GaN on a flat silicon wafer, but instead initiate growth in narrow stripes, which can eventually coalesce (see Figure 1). By employing a micro-patterned substrate, we can ‘trick’ GaN into forming the cubic phase in micro-stripes created from two separate hexagonal crystals.
Figure 1. Growth of hexagonal GaN in ‘V-shaped’ trenches leads to the formation of cubic GaN
One of the biggest challenges that we face is that the cubic phase of GaN is metastable – from a thermodynamic perspective, it would rather be hexagonal. However, the energy difference between the two phases is only 10 meV/atom, and if this wide bandgap material forms the cubic phase, it will probably stay that way. The challenge is to get the atoms into the cubic configuration for long enough to repeat the pattern. The difference between these polytypes is merely one of stacking: in the hexagon phase the layers alternate ABABAB, whereas in the cubic phase, every third layer is the same, so the pattern is ABCABC.
Our starting point is a micro-patterned silicon (001) substrate, which is prepared by our collaborator, the group of Steven Brueck at the University of New Mexico. This team uses interference pattern lithography to define long stripes parallel to the (110) direction. These stripes are then etched into the silicon to produce an array of V-shaped grooves with (111)-type sidewalls running the length of the substrate.
The grooves have an 800 nm opening and are spaced at a 4 µm pitch.
There is a lattice mismatch of about 20 percent between hexagonal GaN and silicon (111). So, if thick layers are required, complicated strain management structures are needed to prevent cracking. However, our application only requires a thin layer, so we avoid such complexities.
To prevent alloying of silicon and gallium, we begin by growing an AlN/AlGaN buffer layer. This is followed by deposition of hexagonal GaN: Two (0001) planes of hexagonal GaN grow from the (111) sidewalls of the grooves towards one another (see Figure 2). It is this growth that is the key to our approach, because, atomically, the (0001) plane of hexagonal GaN looks exactly like that of the (111) face of cubic GaN. The two (0001) planes are offset by the natural angle between (111) faces, due to the geometry of the substrate. So, when the next adatom wanders down into the groove, it doesn’t see two hexagonal GaN crystals growing out at one another – it sees the cubic phase, and settles in a cubic lattice position.
Figure 2. Model of the cubic/hexagonal GaN interface
Following this nucleation of the cubic phase, growth proceeds rapidly upward, filling the groove in the new (001) direction. The result is a long, 2-3 µm wide triangular prism of single-crystal cubic GaN that is nestled in the stripe. It features a hexagonal GaN cladding on two of its sides.
With the cubic phase present, it is now possible to grow LED structures, just as one would for the hexagonal case. Forming a high-quality green LED is much easier with this phase of GaN, however, because: less indium in needed in the wells, thanks to the smaller bandgap of cubic GaN; and there is less strain in the active region, because the InGaN layer is only grown in long, thin stripes. A more relaxed layer is beneficial, because it is less likely to form dislocations and more likely to incorporate more indium.
Scrutinising our sample with electron-backscatter diffraction confirms the presence of cubic GaN in the centre of the stripe, and hexagonal material at the edges. Meanwhile, cathodoluminescence reveals that our cubic material has a band-gap of 3.23 eV. According to cross-sectional transmission electron microscopy, the cubic regions are virtually free of line defects and without wurtzite inclusions. This is a breakthrough, because other groups have reported wurtzite inclusions within their cubic GaN. These imperfections may have a detrimental effect if present in the active region of an LED.
Figure 3. LEDs formed in the stripes of a silicon wafer produce green emission
We have fabricated a fully functioning, green-emitting cubic GaN LED on our engineered substrate (see Figure 3). It’s electroluminescence spectra are stable, even when the driving current is changed by an order of magnitude (see Figures 4 and 5). In comparison, subjecting a similar hexagonal GaN LED to changes of drive current of this order can induce a blue-shift in emission wavelength of up to 20 nm. We have also carried out preliminary electrical measurements with nano-probes, finding good electrical performance for p-doped cubic GaN (see Figure 6).
Figure 4. Unlike a conventional green LED, cranking up the current does not lead to a shift in emission wavelength
Figure 5. In a hexagonal InGaN quantum well (left), electron and hole wave-functions are spatially separated due to internal polarization fields. In a cubic InGaN quantum well of the same Indium composition (right), carriers overlap and the emitted photon is of longer wavelength
Figure 6. Nano four-point-probe and transmission line experiments indicate that resistivity is about 0.28 Ωcm Cubic GaN grown on micro-patterned silcion (001). Direct GaN on silicon technology could lead to fully integrated LED displays
Our proof-of-concept devices built of cubic GaN are promising, wavelength-stable green LEDs with good electrical properties. But there is more work to do. First, we want to propel the LED to longer wavelengths: Why should we stop at green, when we could reach yellow and even red?
In addition to reaching deeper into the colour spectrum, we will continue our study of the widely unexplored electrical properties of the cubic phase. We also plan to fabricate higher-performance cubic LEDs by removing the silicon substrate, so light extraction is increased through the underside of the stripes. A longer-term goal of ours is to expand and coalesce the separate cubic GaN stripes into a single, large epitaxial film. Ultimately, this technology has the potential to be scaled up to larger silicon wafers. Handling and processing equipment is already in place for this, thanks to the silicon IC industry.
If we could bring this technology to 300 mm or larger wafers, this could drastically increase the number of LEDs that could be formed from a single growth run. What’s more, these cubic GaN templates don’t just have to be the ideal platform for making green LEDs – they could be a great foundation for blue and red emitters too. If they could arm the makers of colour-tuneable lighting systems with efficient LEDs spanning the entire visible spectrum, it would enable them to offer products that set a new benchmark for performance.
This work was supported primarily by the Engineering Research Centers Program of the National Science Foundation under NSF Cooperative Agreement No. EEC-0812056. This work was also supported by New York State under NYSTAR contract C090145. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect those of the National Science Foundation or New York State
J. M. Stark, et al.Appl. Phys. Lett. 103232107 (2013)