RPI Starts To Extinguish The Green Gap
Who would not want to go green these days? Not the green of the dollars that were lost on Wall Street, more the green of ecologic responsibility.
Among LED developers, green is a goal in terms of efficiency and color. Bright-red tail lights adorn today s automobiles and electronic gadgets are covered with blue LEDs. But it s much trickier to make a true green that efficiently emits at 550 nm – the wavelength to which our eyes are most sensitive.
A fascination with green might seem misplaced. Doesn t the future lie with white LEDs – well touted replacements for inefficient incandescents and ugly fluorescent lamps? Yes and no. These white-emitting chips are highly efficient, as anybody with a battery-powered LED flashlight will testify. But skin tones produced by flashlight photography look awful because they are skewed by the light source – a mixture of a blue LED and yellow phosphor.
Switch to a red-green-blue source, such as that used for display backlighting, and color reproduction improves. But even this has its limitations. If you illuminate food or jewelry, such as diffracting diamonds and mother of pearl, some colors will still be missing. That s because colors cannot be seen unless they are already in the light source. What s needed is a full color spectrum that includes efficient greens, which requires a solution to the green gap.
At Rensselaer Polytechnic Institute (RPI) in Troy, NY, we have embarked on this quest and made significant progress with growth techniques that produce the smooth quantum well and barrier layers required for efficient greens. LEDs that are built on non-polar crystal faces are also showing promise.
We decided to avoid the combination of green-emitting phosphors and blue LED chips because down-conversion reduces efficiency by at least 20%. Instead, we have looked to extend the spectral coverage of conventional nitride devices.
In one sense, it s obvious how to increase the wavelength of a nitride LED – add more indium to the quantum wells. However, Ga1–xInxN layers with more than 20% InN can suffer from strain that leads to defects and low radiative efficiency.
In addition, III-nitride crystals have a hexagonal crystal structure, and heteroepitaxial growth on conventional sapphire leads to a complementary pair of static two-dimensional sheet charges at every interface. Large internal electric fields result from these, which are equivalent to +/–1014 cm–2 electrons. They pull apart the electrons and holes in the quantum wells, causing wavelengths to shift with drive current and cutting the electron–hole recombination rate.
We have extracted a value for the electric field strength from spectroscopic features seen in measurements on nitride structures with 3 nm thick quantum wells. However, this value for electric field strength cannot explain other experimental results on these structures, such as the large variation in radiative recombination rates and the large Stokes shift between the absorption edge and emission band.
But we can make sense of everything when we describe the electric field as a "Stark ladder" with step sizes equal to the polarization dipole (figure 1). The Stokes shift accurately tracks this dipole energy (the product of electric field strength, well width and electron charge) in ultraviolet, blue and green nitride quantum-well structures. It also accounts for the blue shift in emission wavelength with increasing drive current.
We are not the only team of researchers with an explanation for wavelength variations in these structures. Other groups have put forward qualitative theories based on fluctuations, separation and ordering in GaInN wells, which are based on observations from X-ray diffraction and high-energy electron-beam experiments. We don t dispute their claims that GaInN can show compositional variations, because spinodial decomposition of GaInN is to be expected under thermodynamic equilibrium. However, MOCVD processes can take place away from equilibrium conditions. Horst Strunk s group at the University of Erlangen, for example, has identified material with high and low alloy uniformities.
Our studies have involved the fabrication of a range of LEDs on c-plane sapphire that emit at various wavelengths. We have found that device output and efficiency fall at wavelengths beyond 500 nm, due to an increase in quantum-well roughness. This results from V-defects that decorate threading dislocations, which are far more detrimental to green LEDs than to blue ones.
Over the course of a couple of thousand epitaxial growths we have learned to reduce the active region s roughness. Progress has been assessed through measurements of the smoothness of the barrier layer with an atomic force microscope. We have established a growth regime that eliminates the decoration of threading dislocations with V-defects.
X-ray measurements reveal the excellent thickness uniformity of improved barrier and well structures. This has allowed us to double the number of quantum wells in the active region, resulting in an even greater gain in the light output of our green and dark-green LEDs. Pulsed measurements reveal that 350 µm × 350 µm bare die, which don t have the benefits of light-extraction technologies, produce 45 mW output at 537 nm and 700 mA (figure 2). And 700 µm × 700 µm die can deliver 70 mW at 545 nm under the same driving conditions.
Using the same production process, 530 nm structures can hit a peak electroluminescence internal quantum efficiency of 40% and have an efficiency of 20% at 35 A/cm2 – the typical current density for 1 W lamps. We believe that these are the best values ever reported for deep-green LEDs.
Substantial additional gains should result from improvements in crystalline quality, and in particular a reduction in threading dislocations. HVPE can deposit thick GaN layers with dislocation densities of 5 × 106 cm–2. Companies such as Kyma Technologies are producing polished, epi-ready bulk GaN substrates with this technology in sizes of up to 2 inches in diameter.
This material is a great starting point, but its use can only be justified when its high quality is carried through to the epitaxial structure. Polishing damage, even below the surface, can cancel out the benefits.
Armed with this material, we have studied the effects of piezoelectric polarization on the LED s efficiency, its performance droop and, in particular, its role in the green gap. By slicing along different crystallographic orientations of thick GaN layers grown in the c-direction, we have prepared 5 mm stripes of a- and m-plane material and used them for the growth of non-polar devices.
The LEDs that we have grown on the a- and m-planes are not hampered by polarization fields that are associated with strained layers. They are also free from the threading and misfit dislocations that plague nitride heterostructures grown on cheaper platforms, such as r-plane sapphire and m-plane SiC. Defects are very scarce, with cross-sectional transmission electron micrographs (TEMs) showing an upper defect density limit of 107–108 cm–2 for our LED structures grown along the c-, a- and m-axes (figure 3). This compares favorably to non-polar structures produced by other groups.
Identical process conditions for these three types of LED are no guarantee for identical layer structures and optimal performance in every case, due to alloy composition and layer thickness variations. However, TEM and X-ray diffraction spectra reveal well width and alloy composition variations of less than 10%. This level of uniformity is not actually that surprising – first principles calculations by John Northrup from Palo Alto Research Center, CA, suggest that at an atomic level the growth mechanism is the same for GaInN grown on the c- and m-planes.
The differences in polarization fields in the c-, a- and m-plane devices have a big impact on emission wavelength. Our c-plane is red shifted by 70 nm compared with its non-polar m-plane equivalent (figure 4). However, this is to be expected – our calculations indicate that the polarization dipole in the polar LED is 250 meV, and if the fields could be turned off it would be blue shifted by 58 nm.
Red shifting of LED emission by polarization fields aids the development of polar LEDs spanning the shades of green. Non-polar structures don t share this benefit, so they need much higher indium content in the wells, or a larger well width. We have now adopted these approaches and started to produce green a-plane LEDs on r-plane sapphire and on the a-plane side wall of bulk GaN (figure 5). As expected, threading dislocations plague the former structure, but are almost absent from the latter.
LED powers from these non-polar, green-emitting structures are currently less than 1 mW, but we expect substantial improvements through better p-type doping and optimized junction placement. One encouraging sign is that these devices deliver the predicted wavelength stability with drive current. Unlike with polar structures, there s no blue shift in emission – in fact, there s just a tiny red shift as the current is cranked up.
At the University of California, Santa Barbara (UCSB), researchers have reached even longer wavelengths with semi-polar GaN LEDs. The polarization dipole in their device is weaker than that in a c-plane LED, but it still pushes the emission to longer wavelengths and causes a significant blue shift with drive current – UCSB s 600 × 450 µm chip emits at 579 nm at 1 mA, and 550 nm at 100 mA.
The recent availability of polar and non-polar LEDs will help us to unravel the role of piezoelectric polarization in the green gap. Questions that we can start to address include: does light output efficiency benefit from the improved wave-function overlap in the non-polar structure, and are blue LEDs highly efficient as a result of piezoelectric polarization? The roles of electron overflow into the p-type layers and non-radiative Auger processes – popular theories for the cause of droop in blue LEDs – also have to be established in relation to the green gap.
Although there are fundamental issues still to be resolved, it is clear that reduced threading dislocations and improved polarization control will play a huge role in bridging the green gap. This could lead to the manufacture of a product with green credentials in all aspects - a white LED lamp with a great color balance and unprecedented efficiency.
T Mukai et al. 1999 Jpn. J. Appl. Phys. 38 3976.
C Wetzel et al. 2000 Phys. Rev. B 62 R13302.
H P Strunk et al. 2006 J. Microscopy 224 79.
T Detchprohm et al. 2008 Appl. Phys. Lett. 92 24119.
H Sato et al. 2008 Appl. Phys. Lett. 92 221110.
C Wetzel et al. 2008 J. Cryst. Growth 310 3987.
View pdf of article