Pyramids pave the way to monolithic white LEDs

LEDs will utterly dominate the display backlighting market in the next year or so. This has spurred LED chip manufacturers to hunt for new, lucrative applications, and by far the biggest of these is general illumination. Producers of LEDs are already starting to tap into this market, but they will only make significant inroads when they can substantially cut the cost per lumen of this emitter, a goal that can be fulfilled through increases in luminous efficacy. This revolution in lighting will be driven by white LEDs, which combine a blue-emitting chip with a phosphor that is pumped with blue light and emits yellow. In this device, the luminous efficacy of the white emission resulting from colour mixing is limited – even if the efficiency of the blue LED and the phosphor are very high, there is still an unavoidable energy loss due to the difference between the energy of the photons used to pump the phosphor and the energy of those emitted. Using a phosphor also has a practical downside – coating this material onto the LED chip adds to manufacturing costs. The ultimate approach is a monolithic white LED that emits multiple spectra, such as blue and yellow or the three primary colours. However, fabricating such a device is challenging due to the plummeting efficiencies of InGaN blue LEDs at longer wavelengths and the rapid fall-off in the efficiency of red InGaP LEDs at shorter wavelengths. Acting together, these weaknesses lead to an absence of efficient green and yellow emitters – the so-called ‘green-gap’ – and they hamper the realisation of a monolithic white LED. At Samsung Advanced Institute of Technology (SAIT) we have being developing a novel, alternative LED architecture for adressing the green gap and enabling the fabrication of monolithic white LEDs. Our technology is based on nano-scale pyramids. Why phosphor-free? An affordable, high-quality LED light bulb is the holy grail of solid-state lighting. This type of luminaire with good colour quality and very high efficacy is already available today, but the cost is far too high for many consumers. Scaling-up chip manufacturing is the obvious way to cut costs, with large size silicon wafers offering the best returns. We have developed a technology for this and have recently fabricated blue LEDs delivering 510 mW at a 350 mA drive current on 4-inch and 8-inch silicon substrates. But when it comes to reducing the cost of ownership, it is more effective to increase luminous efficacy than trim manufacturing costs. That’s because this efficacy-centred approach delivers three separate benefits: A cut in the cost per lumen at the chip level, a reduction in packaging cost for the luminaire, and a fall in electrical usage. In the labs of leading chipmakers the luminous efficacy of the best white LEDs can exceed 200 lm/W. At a drive current of 350 mA, the US firm Cree holds the efficacy record with a 231 lm/W device. Further gains in efficacy will undoubtedly follow, but there is not much headroom left because the maximum theoretical efficacy for a phosphor-converting white LED is 263 lm/W. One factor restricting efficacy to this theoretical maximum is the Stokes shift energy loss. This is about 20 percent for blue pumping of yellow phosphors, and the loss is even higher when blue light is used to excite longer wavelength phosphors, such as orange ones, which are needed to form the warm-white light that is desired by the residential lighting market. The combination of colours used to create a white-light source also places a limit on efficacy. The human eye is most sensitive in the green, and the absence of this colour in a conventional white LED has a big impact on this source’s efficacy. Compounding the issue, the broad emission spectrum associated with a phosphor produces photons at wavelengths where the human eye is unresponsive. And making matters even worse, the proportion of invisible photons gets more severe in warm white. Eliminating wavelength conversion gets around these issues and increases the theoretical maximum efficacy to more than 400 lm/W. But building such a device is tricky because it requires the fabrication of a single chip that not only produces polychromatic wavelengths, but also delivers efficient green, yellow and possibly even red emission. Barriers to monolithic white LEDs The green gap in nitride LEDs stems from a combination of poor crystal quality of indium-rich InGaN and the polar characteristics of III-Nitrides on the cplane. Macroscopic polarization occurs in this material, giving rise to a piezoelectric field perpendicular to the plane of the quantum well. This field pulls apart electrons and holes in the well, leading to a decline in the radiative recombination rate (called quantumconfined Stark effect). As the wavelength increases, this efficiency reduction becomes more severe. A hike in indium content is needed to reach these longer wavelengths, and this also increases strain in the quantum well, leading to higher piezoelectric fields that hamper radiative recombination. On top of this, the larger strain and lower growth temperature required to incorporate more indium deteriorate emission efficiency, due to the generation of many non-radiative recombination centres, such as point and line defects. One extensively studied, experimentally proven technique for sidestepping the piezoelectric fields is to turn to GaN grown on semi-polar or non-polar substrates. These can be made from either sapphire or GaN. However, stacking faults plague epitaxial non-polar and semi-polar films grown on sapphire, and free-standing GaN substrates of any orientation are too expensive to be used for making LEDs. One promising alternative that is under investigation by many groups, including National Taiwan University, is to reduce the electric field through strain control, such as pre-straining of the multi quantum wells (MQWs). A more radical idea that has great potential is to build LEDs from InGaN nanostructures. Emission from the blue right through to the red has already been demonstrated with such structures, which have received much attention thanks to their promise to close the green gap and realise polychromatic white LEDs. Strengths of the nanostructures include facets for semipolar and non-polar GaN growth, enhanced light extraction, and the promise of increased crystal quality, thanks to reduced strain that stems from their small features. We have used this class of structure – specifically nanoscale pyramids with InGaN layers – to produce epilayers delivering very efficient green, yellow and red emission. In addition, we have fabricated a monolithic LED that produces white light through colour mixing from different quantum wells. Closing the green gap To produce structures with efficient green to red emission we have used MOCVD to grow InGaN/GaN MQWs or a double heterostructure (DH) on nano-size GaN hexagonal pyramids. These are formed via selective growth on patterned c-plane GaN templates featuring circular openings in a 100 nm-thick SiN film. After patterning the wafer, we form un-intentionally doped GaN hexagonal pyramids with a proprietary growth process. This growth step concludes with the addition of three InGaN QWs with GaN barriers or an InGaN/GaN DH structure, with the growth condition for the ternary layer carefully selected to control emission wavelength and efficiency. The SiN films are not removed after growth. The six facets of these arrayed pyramid structures are clearly visible in scanning electron microscopy and transmission electron microscopy (TEM) images (see Figure 1). According to high-resolution X-ray diffraction, all of these facets are semipolar {1122} planes. One of the benefits of this approach is that the threading dislocations are terminated before propagating into the InGaN layer – see the cross-sectional TEM image in Figure 1 (b). This dislocation filtering that occurs when carrying out selective-area growth through nano-scale openings arises due to the thermal mismatch between GaN and the dielectric mask. We are able to produce a wide range of colours with efficient emission using our nano-pyramid structure. Photoluminescence (PL) measurements reveal green, yellow and red emission (see Figure 2 a), with corresponding internal quantum efficiencies of 61 percent, 45 percent and 29 percent, respectively, according to Arrhenius plots of the normalized integrated PL intensity over a 10 to 300 K temperature range (see Figure 2b). To identify the origin of this high efficiency, we excite the yellow MQW on nano-pyramids and compare its emission with that produced by another structure – a blue MQW grown on the c-plane of another wafer. PL spectra generated by pumping both structures at a range of energies uncovers a linear relationship between excitation power and PL intensity that kicks in at lower incident powers in the yellow-emitting structure, indicating that this one has fewer defects than the blue MQW (see Figure 3). We have determined the strength of the piezoelectric field through low-temperature measurements of the shift in the emission peak as a function of excitation power density (see Figure 4). Photo-generated carriers screen the piezoelectric field, so it is possible to estimate the field strength from blue shifts in the emission peak with excitation power. Again, we compare the yellow and blue-emitting structures: A blue shift of 29 meV occurs in the blue MQW when the excitation power is increased from 1 to 10 mW, indicating the presence of the piezoelectric field; in comparison, the blue shift in the yellow quantum well is negligible. To reveal whether this strongly suppressed piezoelectric field in our yellow emitting pyramids stems from the semi-polar plane or results from strain relaxation in these nanostructures we scaled this structure, building equivalent pyramids with a bottom diameter of about 2 μm. In this case the blue shift was 47 meV. This is a relatively small shift for emission centred around 570 nm, indicating significant influence from the semi-polar growth plane (In comparison, variations in excitation power of one order of magnitude have been reported to produce a 143 meV blue shift in MQW structures emitting at 500 nm). However, the blue shift associated with the micron-sized pyramids is still far, far larger than that occurring in its nano-scale cousin. Our conclusion: Growth of MQWs on {1122} facets of nano-size pyramids effectively suppresses the piezoelectric field via the semi-polar growth plane and strain relaxation. Building a white source One of the benefits of using selectively grown InGaN is that its composition can be varied through changes in both the growth condition and the type of selective growth employed. This has very important implications: The wavelength of an InGaN layer on nano-pyramids is different from that on a planar substrate, and it is consequently possible to realise multiple wavelength emission. We have grown an InGaN MQW on a nanopatterned structure that has regions with 20 μm x 20 μm openings for planar growth. Micro-PL measurements reveal that the resultant epistructure produces white PL emission via colour mixing of blue light from the planar area and yellow emission from the nano-pyramids (see Figure 5). The longer wavelength that results from a higher indium composition in the MQWs in the nanopyramids is not due to simply different planes for deposition. Instead, it is caused by enhanced presence of indium species, which undergo lateral vapour diffusion and surface migration from the dielectric mask to the GaN pyramids. In fact, we have found that when we grow an InGaN MQW after removing the SiN layer, the emission peak from the pyramids shortens by 15 nm. We have also learnt that it is possible to control the emission wavelength from the MQWs on the nano-pyramids by adjusting the separation of these pyramids. Through variations in selective epitaxial patterns we have made multiple wavelengths LEDs (see Figure 6, which shows blue, cyan, yellow, and white LEDs from the same substrate). The white variant combines yellow emission from InGaN MQWs on nano-pyramids and blue emission from micron-sized planar areas. The forward voltage for this device at 20 mA is 3.69 V, which is higher than ideal due to the high resistivity in the p-contact layer on the nano-structures. However, the turn-on voltage is lower than that for a blue LED, thanks to the lower bandgap of the yellow MQW. The colour temperature for this LED is 7100 K, but lower values are possible by adjusting the area ratio between the nanopyramids and planar area. Our efforts highlight the tremendous potential of LEDs built from nano-pyramid arrays, which can deliver high luminescence efficiency over a very wide spectral range thanks to the combination of reduced defects, relaxed strain and a suppressed piezoelectric field. Challenges still remain, however, and one of the biggest is realising uniform current flow in the InGaN layer – today it crowds through the shortest current paths due to the three-dimensional geometry of the structure. Although the crowding is not as bad as it would be in nanorods, uniform current injection is essential for highpower LEDs. We will work to overcome this issue, and also try to develop higher p-type doping on the {1122} plane, a step that is needed to lower the LED’s operating voltage.