High brightness LEDs continue to make inroads into various markets, including displays, signage, traffic signals, backlights for cell phones and automotive dashboards, and exterior automotive lighting. While there has also been penetration of various niche lighting markets, the "Holy Grail" of replacing existing (incandescent, fluorescent etc.) technologies for general illumination remains in the distant future. A U.S. effort to seek federal funding for an industry-wide development program has been gaining momentum recently. A similar effort, entitled "The Light for the 21st Century", is already under way in Japan. In the U.S., the Optoelectronics Industry Development Association (OIDA) and the Department of Energy (DOE) recently organized two workshops, the first to examine the technical issues, roadblocks and challenges facing the LED community [1], and a similar workshop to examine the potential of organic LEDs (OLEDs). This article reports only on the LED workshop. As described in the Agilent/Sandia white paper [see CS 6(2), p.32], the potential benefits of solid state lighting (SSL) are enormous. By 2025, it is estimated that SSL could decrease by 50% the global amount of electricity used for lighting, and reduce total global electricity consumption by 10%. This would result in savings totaling around $100 billion per year, as well as environmentally significant reductions in the levels of carbon dioxide emission from power generation. These savings assume a 50% market penetration by SSL, which in turn requires light sources capable of operating at a luminous efficacy of 200 lm/W. There are many hurdles to overcome before this daunting level of performance can be achieved, both for AlInGaP and InGaN devices (see page 53) and in using these devices to make white light sources. What Does the Lighting Industry Want? One of the issues facing LED manufacturers is that they don t necessarily appreciate the requirements of the lighting industry. While excited by the potential energy savings, controllability and other attributes of LEDs, the lighting industry also expects that LEDs will have the same properties as existing light sources in terms of features such as distribution, flux density and color quality. At the OIDA workshop, Steve Johnson, head of the lighting research group at Lawrence Berkeley National Laboratory, discussed some of the most important attributes of light sources: 1. Distribution. Light distribution in a uniform, controllable, efficient manner is perhaps the primary concern of the lighting industry. Although LEDs are in theory an extremely good point source, the spatial and power distribution of light from packaged LEDs is not well defined by the manufacturers specifications. Also, the color of white LEDs can shift from the center of the device to the sides, due to the color mixing that occurs to create white. 2. Light output/flux density. The lighting industry works with the total luminous output of a lighting system. A 100 W incandescent bulb provides 1700 lumens, while the best LED gives about 60 lumens. Methods to combine many LEDs to get the desired aggregate output, and the effect this has on light distribution, also need to be considered. 3. Color rendering index (for white light). Color rendering refers to how a color appears when illuminated by different light sources, compared to the color under natural daylight. A source is assigned a color rendering index value Ra between 1 and 100 by comparing the appearance of various colors under illumination by the source in question and by various standards. As shown in , the application-specific requirements for Ra vary a great dealfor example, Ra must exceed 80 for general illumination. The table also shows the potential performance of various LED lighting technologies. 4. Controllability. The potential for fully controllable LED light sources is very exciting for the lighting industry. The ability to vary the source intensity or switch a lamp on and off without affecting the lifetime is very attractive. Also, combinations of different colored LEDs allow any color to be produced. 5. Color temperaturesee footnote [2]. Fluorescent lamps are sold with different (fixed) color temperatures for different applications, and LED light sources will need to replicate this. There is at present a huge difference in the acceptable variations of color accuracy in fluorescent lamps, and the color variations found in white LEDs. 6. Lifetime. LED light sources offer very long lifetimes, with the potential to reduce replacement and maintenance costs and therefore overall operating costs. Although low cost is obviously of paramount importance in the lighting industry, many of the other requirements, such as long lifetime and high efficiency, also have a large cost element. 7. Efficacy (lumens/watt). The potential to greatly increase the efficiency of LEDs could lead to considerable savings in electricity usage, which of course lowers the overall cost of operation. Challenges for Solid State Lighting There are two main approaches to achieving white light using LEDs. One is to combine the output from two or more LEDs of different colors (the multi-chip LED approach), the second is to use a phosphor or other material to down-convert the emission from a blue or UV LED (the phosphor converted or PC-LED approach). These methods have many pros and cons, and are both likely to remain in use for several years. Ensuring color quality and control is essential to all applications, and can prove difficult with either white LED approach. Color accuracy in PC-LEDs is strongly affected by the emission wavelength and other parameters of the blue pump LED, as well as by the amount and distribution of the phosphor around the LED chip. Flux variations in any of the components of a multi-chip LED can shift the color accuracy, and because two different material systems are currently involved (AlInGaP and InGaN) there is unlikely to be a uniform, predictable degradation of LED properties over time. Active control is therefore required in multi-chip LEDs to maintain the color accuracy. "The color accuracy of PC-LEDs is fixed by the manufacturing process, so this step is critical," notes Mike Pashley of Philips Research. "RGB multi-chip LEDs can theoretically be adjusted to give the correct white point; the challenge is to do that cost-effectively." An example of the difficulty of achieving good color quality with multi-chip LEDs was given by Mike Krames of LumiLeds Lighting, who described a multi-chip LED combining 460, 530 and 630 nm devices. The combination potentially offers high efficiency but has a poor color rendering index. A better CRI value can be achieved using 460, 550 and 610 nm LEDs, but this introduces further problems. Using a 610 nm AlInGaP LED causes both the color temperature and wavelength to vary with temperature, while both the efficiency and color stability with drive current of the InGaN green LED suffer at wavelengths above 540 nm. Phosphor-converted LEDs do not require active control or color-mixing optics, and have simpler (less expensive) fabrication than RGB LEDs. Potential disadvantages are the lack of available converter materials and the inherently lower efficiency. "Conversion efficiency is still a problem with PC-LEDsthe power conversion efficiency from blue to white is about 0.58 for a single phosphor," says Mike Krames. As shown in , the combination of a blue LED and a yellow YAG phosphor does not give a sufficiently high CRI value for indoor lighting applications. Other combinations, such as blue LEDs with green and red phosphors, have also been investigated. Fluorescent lamps use a tri-color phosphor optimized for white point and color rendering, which is excited at 254 nm. One potentially optimal solution for white LEDs is to use a three-color phosphor and a UV LED. Since the LED does not contribute to the visible light output, the white point is independent of the LED characteristics. "Future solid-state lighting is likely to combine high quantum efficiency UV LEDs plus an optimized three-color phosphor plus low-cost packaging," says Steve DenBaars of Cree Lighting. His company has reported excellent results for short wavelength LEDs, namely 28% QE for 405 nm InGaN LEDs, compared to 17% for 520 nm devices. However, a great deal of work is necessary to create phosphors that can absorb at violet/near-UV wavelengths and provide the correct emission characteristics. According to Mike Pashley, while PC-LEDs in particular provide a low profile light source, many LEDs operating at 30 lm/W are required to get the desired total lumen output, meaning that the total area of the LED light source can be large. Even so, because the chip size is very small, collimation and beam-shaping optics can be positioned close to the source, allowing easy control of the light to generate complex light distributions. "LEDs have the potential to illuminate a designated area with much greater efficiency than existing sources," says Mike Pashley. "A good example is street lighting, where low pressure sodium lamps throw light everywhere in a very inefficient manner." The Next Steps One of the main aims of the OIDA Workshop was to reach an industry-wide consensus on the targets and challenges for solid-state lighting, some of which are listed in . Based on the target of eventually achieving 200 lm/W sources, a tentative roadmap of parameters for white LEDs was put forward at the OIDA workshop [see . These numbers reflect expectation of general trends and have not yet been accepted by OIDA or DOE. OIDA s intention is to generate a framework that will assist in forming partnerships between industry, government, academic institutions and national laboratories to accelerate the effort for achieving commercial solid state lighting. The process will also assist industry in strategic business and technology planning, identify opportunities for government-industry collaboration, and highlight issues that are more suited for research programs in national labs and academic research programs. Specific recommendations will be given to both industry and the Department of Energy to advocate the need for a major industry-government cooperation to put the U.S. into a leading position in solid state lighting. The stakes include (i) substantial energy savings, (ii) reduced carbon dioxide pollution and (iii) the creation of a new U.S. lighting industry. Challenges for AlInGaP- and InGaN-based LEDs In any potential market for high brightness LEDs, from traffic signals to white lighting for general illumination, there remains a great deal of work to be done to increase market penetration of LED technology. Compound Semiconductor spoke to Frank Steranka, R&D Manager at LumiLeds Lighting, who outlined some of the challenges raised at the recent OIDA Workshop. A general feature of LED-based lighting applications is that wider market acceptance requires brighter devices at lower cost. Competing technologies, such as incandescent bulbs, provide a large number of lumens per dollar, and LEDs will eventually have to offer competitive cost-performance figures. "For red or amber devices, penetration of markets such as automotive is largely determined by the cost benefit of LED technology," says Frank Steranka. "There are certainly other advantages to using LEDs, but a lot of auto companies look only at the bottom line." "In general, the more LEDs start to penetrate the lighting market as a whole, the less the unique features of LEDs are going to count, and the more the cost of the solution compared to competing technologies becomes important." The driving forces are a little different for nitride-based LEDs, where energy savings are an important factor in a larger fraction of the market. "For AlInGaP, energy savings are important for traffic signals, but much less so for other red lighting markets such as signage or automotive," says Frank Steranka. "For nitrides, cost is important but it can be somewhat offset by energy savings, especially in general lighting applicationswhen lights are on a lot of the time, energy usage quickly adds up." The general target for high brightness LEDs is therefore to increase the number of lumens per dollar, using techniques such as increasing the internal quantum efficiency, improving the extraction of photons, and reducing manufacturing costs by raising yields and improving epitaxial growth. AlInGaP LEDs One target for AlInGaP LEDs is to improve the internal quantum efficiency, particularly at shorter wavelengths, where reduced conduction band offsets result in poor carrier confinement. Bandgap engineering of the device structure could result in significant improvements in both performance and temperature sensitivity. A number of techniques have been employed in AlInGaP semiconductor lasers to try to improve confinement of the carriers in the light-generating region, but these have not been widely employed to date in LEDs. "These require tight control over the growth process, and are possible approaches for the future as the quality of production-scale MOCVD reactors improves," notes Frank Steranka. There is also a need to be able to operate the devices at higher current densities. "The more current you can drive through one of these LEDs while maintaining high reliability and efficiency, the more lumens you get out: the cost is roughly the same so the lumens per dollar rises." However, operation at very high current densities generates heat and causes the efficiency to fall. Bandgap engineering can make devices less sensitive to temperature, and improved heat sinking also helps by getting the heat out more effectively. The development of low thermal-resistance packaging has been a major focus at LumiLeds. Another key area for AlInGaP LEDs is light extraction. LumiLeds truncated-inverted-pyramid chip structure has proved successful in extracting over 50% of the photons generated, and results in improvements in lumens per dollar. However, further gains are possible. For example, the use of higher index encapsulants could also potentially increase light extraction. AlInGaP LEDs have an index of refraction approaching 3.5, compared to 1.5 for standard epoxy encapsulants. For such a large refractive index step, there is a small critical angle and a lot of light gets reflected back into the chip. The challenge is to raise the index of refraction of the encapsulant without reducing its transparency or adding cost. Considering LumiLeds wafer-bonded transparent substrate approach, it may be that the introduction of an "epi-ready" transparent substrate (e.g. a GaP substrate with a high quality GaP-to-AlInGaP graded layer) could reduce manufacturing costs. "On the other hand, the wafer bonding process is fairly well understood, so huge savings are not necessarily going to arise," says Steranka. "Larger wafer sizes might help: if 4-inch GaP wafers were available, we would consider switching [from 3-inch] and that would likely help reduce the cost." Nitride Devices Probably the biggest target for nitride devices is to develop improved or new substrate materials to replace sapphire and enable the growth of lower defect density material. Many alternatives, including bulk GaN, AlN and ZnO, are being considered. Commercially available MOCVD reactors also require improvementproblems relate both to the design of reactors and the lack of fundamental understanding of the nitride growth processes. "I think everybody would agree that the key to getting low cost nitride devices is improved yield," says Frank Steranka. "One major route to achieve this would be to improve the MOCVD epitaxy process, allowing deposition of InGaN layers with more uniform composition." Yet another key issue is the efficiency of nitride devices at high current densities. "Today s devices fall in efficiency as the current density is increased," says Steranka. "Eliminating this problem could also result in significant improvements in lumens per dollar." White lighting requires the development of improved phosphors to better match the LED emission characteristics. The two main approaches are combining blue LEDs and yellow phosphors, or combining UV LEDs and three-color (red, green, blue) phosphors. The OIDA meeting suggested that it could take about 5 years of work to develop the latter type of phosphor. The blue LED + yellow phosphor approach requires further improvements in efficiency. It is difficult to control the color temperature of the white light, partly due to variations in the emission of the blue LEDs, partly due to the manufacturing control required to ensure the right amount of phosphor around the chips. Also the color rendering index needs to be improved if this approach is going to be used in a bigger fraction of the indoor illumination market. At shorter wavelengths, performance appears to be more sensitive to defect density. Gains in performance and lifetime in short wavelength blue lasers have come about as a result of reduced defect density, using ELOG and then ELOG on GaN. Defect density seems to be a bigger issue for short wavelength violet/UV emitters than for longer wavelength devices. A lot of the packaging materials commonly used for LEDs are not suitable for UV emitters. "LumiLeds uses a silicone encapsulant, which has fewer issues in the deep blue to UV range. This is why we have been able to demonstrate long-term stability from our white high-power LEDs," says Frank Steranka. "However, other encapsulants that won t degrade or yellow, or won t absorb UV light at all, are also desirable for other types of devices." For nitride devices, light extraction is less of a problem since the sapphire substrate is reasonably transparent. "LumiLeds recently introduced a flip-chip structure [see CS 6(9), p. 19], which does a pretty good job of enabling the photons to escape," says Frank Steranka. "Most of the focus for nitrides in the future will be on the development of alternate substrate materials, further improvements in the MOCVD epitaxy process, and improvements in high-current density efficiency. The resulting improvements in luminous flux and yield will drive the improvements in lumens per dollar that are required for solid state lighting to become a reality." [1] OIDA Workshop on LEDs for Solid State Lighting, October 2627, 2000, Albuquerque, NM [2] Any color of light can be expressed by its chromaticity coordinate x,y on the CIE 1391 chromaticity diagram. The boundaries of the diagram (the so-called Spectrum Locus) represent monochromatic colors. The Planckian Locus traces the chromaticity coordinates of a blackbody radiator at temperatures in the 100020,000K range. Color Temperature refers to any point on the Planckian Locus. A typical incandescent lamp has a color temperature of around 2850K, and the typical color of daylight is 6500Kmost white lamps fall between these color temperatures. Any shift of color temperature along the Planckian Locus is usually acceptable depending on the application; shifts away from the Planckian Locus are often unacceptable.