Mari Holcomb, Pat Grillot, Gloria Hfler, Mike Krames and Steve Stockman from LumiLeds Lighting explain how they overcame the challenges involved in developing amber, orange and red AlGaInP LEDs with impressive efficiency and power characteristics.

Incandescent lighting sources, first developed by Edison in the 1880s, still represent roughly a third of the energy consumed for residential and commercial building lighting in the United States. For monochromatic lighting applications such as automotive lighting or traffic signals, light from the incandescent bulb is typically filtered through a plastic lens of the desired color. This results in poor efficiency, since most of the visible light produced is thrown away. Visible LEDs, as introduced commercially in the 1960s, offer the advantage of efficient direct monochromatic emission. However, until recently the commercial use of visible LEDs has been largely confined to indicator and display applications. The recent efficiency improvements of MOCVD-grown AlGaInP (red, orange and amber) and GaInN (blue and green) LEDs have enabled their use in a wide range of applications such as exterior automotive lighting, traffic signals and full-color outdoor signs. Further efficiency improvements and manufacturing cost reduction will enable LED-based systems to compete in the $40 billion lighting market with conventional technologies such as incandescent bulbs, fluorescent lighting, sodium vapor lamps and neon. The highest-efficiency LEDs demonstrated to date come from the quaternary AlGaInP material system, which encompasses the amber through red color regime. Four generations of AlGaInP chips are shown in , highlighting the major development stages of these high-efficiency and high-power devices. In this article, the key technologies responsible for the record performance of AlGaInP LEDs will be reviewed, including epitaxial growth, wafer bonding, chip shaping and LED packaging. Device design strategies for optimizing LED efficiency and the resulting state-of-the-art lamp performance characteristics will be examined, along with the advantages AlGaInP LEDs provide compared with conventional lighting sources. Finally the limitations of AlGaInP technology are discussed, along with future challenges. Growth and device structure (AlxGa1x)0.5In0.5P is lattice matched to GaAs for Al compositions ranging from = 0 to = 1, and has a direct bandgap for = 0 to 0.53. In this direct bandgap compositional range, (AlxGa1x)0.5In0.5P emits over the red (1.9 eV) to yellow-green (2.26 eV) spectral regime; however, the radiative efficiency drops significantly with higher Al composition as the alloy approaches the directindirect bandgap crossover. As such, commercial AlGaInP LEDs are primarily limited to red, orange and amber emission. Commercial AlGaInP LEDs are grown by low-pressure MOCVD. Alloy ordering, hydrogen passivation of acceptor atoms and unintentional oxygen incorporation are key issues in the MOCVD growth of high-quality AlGaInP material. Spontaneous AlGaInP ordering occurs when the AlGaInP material forms an unintentional superlattice of InP/AlGaP layers. This superlattice results in a lower bandgap (longer emission wavelength) than the random alloy. Since AlGaInP efficiency decreases as Al composition increases, it is generally desirable to minimize the emission wavelength (bandgap) for a given Al composition. Reducing or eliminating this alloy ordering, which is also important for wavelength control, may be accomplished by choosing appropriate growth parameters such as substrate orientation and growth temperature, and by post-growth annealing. The latter is also used to fully activate acceptors, which are often initially inactive in as-grown wafers due to hydrogen passivation. Typical p-type dopants used in AlGaInP epitaxy include Zn and Mg, while Si and Te are the most common n-type dopants. Minimizing the incorporation of oxygen is also important, especially in layers with a high Al content, since oxygen tends to introduce defect states in the semiconductor layers that decrease radiative efficiency. Oxygen incorporation depends strongly on Al content, as well as growth parameters such as substrate orientation, growth temperature and input V/III ratio. The careful optimization and control of growth parameters are thus used to help control material properties, resulting in AlGaInP with high radiative efficiency and enabling the commercialization and high-volume manufacturing of AlGaInP LEDs and laser diodes. A schematic of two common AlGaInP LED structures is shown in . For both structures the active layer is sandwiched between two carrier-confining layers which are typically composed of (AlxGa1x)0.5In0.5P, with > 0.7. The top contact layer, usually GaP or AlGaAs, serves as both a current-spreading layer and a window layer to improve extraction of light directed toward the side of the chip. The structure in figure 2(a) is an AS (absorbing substrate) LED. In AS devices a distributed Bragg reflector is often grown below the lower confining layer to increase on-axis light emission. Figure 2(b) shows a TS (transparent substrate) structure in which the absorbing GaAs substrate is replaced with an optically transparent GaP substrate by solid-state wafer bonding. Wafer-bonding technology is discussed in more detail below. Measuring performance The primary figure of merit used to evaluate visible LED performance is luminous efficiency, expressed in lumens per watt (lm/W). This is the wall-plug efficiency (optical power divided by the input power) weighted to match the human eye response. LED luminous efficiency is determined primarily by three factors: the internal quantum efficiency (QE), the extraction efficiency and the operating voltage. The internal QE characterizes how many photons are generated for each electron passing through the LED. The extraction efficiency is the percentage of the generated light that escapes out of the semiconductor chip. Due to the high refractive index mismatch between the semiconductor and the surrounding medium, most photons are internally reflected multiple times, during which they may be absorbed. The product of the internal QE and extraction efficiency is referred to as the external QE. The operating voltage is determined by the active region bandgap and the LED electrical series resistance. The internal QE of LEDs is determined by the material quality and the active region design. The active region of the LEDs is typically composed of multiple thin AlGaInP layers, or quantum wells, separated by higher-bandgap barrier layers. For AlGaInP LEDs, the internal QE increases with the thickness of the active layer, due to reduced electron leakage current to the confining layers, especially for active regions with higher Al composition. However, the extraction efficiency decreases as the thickness of the active layer increases, due to increased absorption during multiple passes of the light through the active region. As a result of these opposing trends, there is an optimum total active layer thickness for a given wavelength so that the external QE is maximized. The total active layer thickness is therefore chosen through experimental optimization for a given target emission wavelength. The extraction efficiency is limited by photon loss during multiple passes within the chip. The primary loss mechanisms are absorption in the GaAs substrate (in AS devices), the active layer and the ohmic metal contacts, and free carrier absorption in the window and substrate (in TS devices). The escape cone for a given point source in the active layer is defined by the critical angle according to Snells law. Increasing the top window layer thickness significantly improves extraction efficiency, because the light directed towards the sides of the chip can escape before being internally reflected or absorbed at either the top or bottom LED surface. The thickness of the top window layer typically ranges from a few micrometers to 60 m. Devices with a 60 m GaP window have demonstrated efficiencies twice that of 10 m windows (Huang et al.). In high-volume manufacturing, thick GaP window layers are typically achieved by regrowth using hydride vapor phase epitaxy. For AS devices a large source of photon loss is the absorbing GaAs. The extraction efficiency can be significantly improved by replacing the absorbing substrate with an optically transparent substrate. TS AlGaInP LEDs were developed at the Hewlett-Packard Optoelectronics division in 1994 using compound semiconductor wafer bonding technology (Kish et al.). After epitaxial growth, the GaAs substrate is removed by selective etching. The remaining epitaxial wafer and the bonding GaP wafer are shown in . The wafer bowing evident in the epitaxial wafer results from the coefficient of thermal expansion mismatch between the AlGaInP layers and the GaP window layer. The epitaxial layers are placed in contact with the GaP so that the crystallographic orientation of the wafers is matched. Robust chemical bonds are created at the bonding interface upon application of uniaxial pressure to the wafers at elevated temperatures. Despite the 3.6% lattice mismatch between the two wafers, good optical transparency and electrical conduction are obtained across the interface. TS devices have at least twice the flux of AS devices made from the same epitaxial material. Interface states at the bonding interface do result in a small forward voltage penalty of 50100 mV compared with AS devices. Nevertheless wafer-bonded TS LEDs are still the devices with the highest luminous efficiency available to date. Furthermore these good bonding characteristics can be achieved uniformly across wafers as large as 75 mm in diameter, enabling TS LEDs to be manufactured in high volumes. Lamp performance Epitaxial growth or wafer bonding are followed by deposition of top and bottom ohmic contacts. The wafers are subsequently diced into squares, typically 210300 m each side. The chips are then attached to a lead frame with silver-loaded conducting epoxy. After wirebonding the top contact to the lead frame, both the die and lead frame are encapsulated in a hard transparent epoxy. A typical 5 mm lamp is shown in (a). The highest luminous efficiency measured from conventional 5 mm lamps made with TS wafers is shown in as a function of peak wavelength (Gardner et al.). The lamps have luminous efficiencies exceeding 50 lm/W at a current density of 40 A/cm2 over the color range used for commercial AlGaInP LEDs. The highest luminous efficiency for the 5 mm lamps is 74 lm/W for lamps emitting at 615 nm. The external QE of AlGaInP LEDs improves with increasing wavelength (lower Al composition) due to better carrier confinement, higher relative electron population of the direct minimum, reduced non-radiative impurity incorporation and reduced absorption. For these lamps an external QE of 32% at 632 nm was observed. However, the luminous efficiency drops due to the decreasing response of the human eye with increasing wavelength, as indicated by the CIE curve. The maximum DC output flux from 5 mm lamps driven at 50 mA is limited to 510 lm per LED. High power and efficiency Typical illumination systems require photometric output power of several hundred lumens. Simply increasing the number of conventional LEDs is often impractical for such systems. To solve this a new generation of high-power LEDs and packaging technology have been developed and commercialized by LumiLeds Lighting. The simplest way of increasing the flux per LED is to make the chip bigger. One example of a chip with increased die area is shown in (c). The junction area of these devices is approximately five times that of the conventional die packaged in 5 mm lamps. Therefore driving the larger die with five times the current of the conventional die (equivalent current density) should in principle increase the flux fivefold. However, simple scaling of the device area significantly reduces extraction efficiency. This results in part from a decrease in the fraction of the light within the critical-angle escape cone that reaches a chip side wall before being internally reflected at the die top or bottom surface. Photons must traverse a much longer path before encountering a surface, making internal absorption much more likely. For a fixed die height, modeling and experiments show a 25% decrease in external QE when the die area is increased from 210 210 mm2 to 500 500 mm2 (Hfler et al.). In addition, conventional LED packaging used for 5 mm lamps is not sufficient to dissipate the heat associated with higher input powers to the larger chip. Thus a high-power package had to be designed in addition to a larger chip. One such high-power package is shown in (b). The die is solder-attached to a submount to compensate for the coefficient of thermal expansion mismatch between the copper heat sink and the LED. These features, combined with the reduced thermal resistance of the chip itself, result in a lamp with thermal resistances of between 12 and 15 C/W, some fifteen times lower than that of conventional 5 mm lamps. Another feature of the package is a soft encapsulating gel that minimizes stress on the die, which is injected through a hard outer protective lens. As a result of all these improvements, the maximum DC output flux from these high-power lamps is more than five times higher than that achieved by conventional 5 mm lamps made of the same epitaxial material. The efficiency-scaling penalty can be overcome by shaping the die. The ideal geometry for maximum extraction efficiency is a sphere with an emitting volume in the center, but this is not practical to manufacture. The truncated inverted pyramid (TIP) shape shown in (d) was chosen as a manufacturable alternative. This geometry experimentally increases extraction efficiency up to 50% compared with square large junction devices with the same 500 500 mm2 active area (Krames et al.). This is due to higher photon emission from both the side and top surfaces. Specifically, compared with conventional dice with orthogonal side walls, rays that are internally reflected from the top (beveled side) surface subsequently hit the beveled side (top) surface with a lower angle of incidence, hence increasing the number of rays within the critical-angle escape cone. The die is oriented junction side down for improved heat sinking and is packaged in lamps similar to those described for square high-power chips. Peak performance AlGaInP TS TIP-LEDs are the most efficient visible spectrum LEDs demonstrated to date in any material system. The DC luminous efficiency at 40 A/cm2 of several TIP-LEDs with peak wavelengths from 594 to 652 nm is plotted in . The highest luminous efficiency of 102 lm/W was achieved for 611 nm lamps. The external QE for these devices ranges from 15% for amber lamps to 55% for deep-red 652 nm lamps. For the deep-red lamps, 60% external QE was observed under pulsed operation. As the internal efficiency is the highest for long wavelength devices, this 60% efficiency represents the lower bound of extraction efficiency for the TIP-LED structure. More than 60 lm of output flux has been measured from single TIP-LED lamps over the wavelength range of 590640 nm at DC drive currents less than 500 mA. This corresponds to roughly 100350 mW of optical power. This is about 15 times higher than the output power that can be obtained from conventional 5 mm lamps. Mechanical and accelerated life testing indicate that the TIP-LED devices are just as robust and reliable as square chips. Lighting applications Commercial AlGaInP LEDs are expected to account for at least 30% of the high-brightness LED market, which is forecast to exceed $3 billion per year by 2005 (see page 53). Several examples of AlGaInP lighting applications are shown in . In the automotive lighting market, AlGaInP LEDs are currently used as center high-mount stop lights on most new European automobiles. With increasing power per LED, LED utilization is expanding to include turn signals, tail lights and daytime running lights motivated by increased styling and design flexibility, faster turn-on times for improved safety and lifetimes exceeding that of the vehicle. High-power AlGaInP and InGaN LEDs are also enabling increased adoption in traffic signals, where the power savings, increased safety (LEDs are more reliable) and reduced maintenance/replacement cost are major advantages over conventional incandescent bulbs. For example, a 12 inch traffic ball using 12 high-power red LEDs requires less than 15 W of input power, an order of magnitude lower than the 150 W incandescent lamp-based traffic balls currently prevalent in the US. LED-based systems are also beginning to replace neon signs, which require hazardous high-voltage operation and are not as mechanically robust as LEDs. There is also growing interest in using LEDs for non-lighting applications such as dental curing and skin treatment. Finally, high-power AlGaInP red and InGaN blue and green LEDs are expected in the future to be incorporated into projection displays and LCD backlighting applications. For these applications, the color purity and color range exceed those that can be obtained by conventional lighting sources. The largest potential market for high-efficiency LEDs is in the area of solid-state illumination. The demonstrated high efficiency of AlGaInP LEDs makes them well suited to these applications, either in the form of monochromatic amber illumination or as a part of a red-green-blue combination for the efficient generation of white light. Future challenges How quickly the market adopts LED lighting solutions will depend in part on the speed of reduction in system costs. This will result from further increases in flux per LED and from lowering the costs of manufacturing, so that the LED dollar/lumen cost is closer to those of conventional lighting sources. Although the AlGaInP material system is relatively mature, there are fundamental physical properties that put a limit on high-power performance. The internal efficiency of AlGaInP LEDs is primarily limited by active-region electron loss to the surrounding confining layers, and the indirect conduction band minima. Carrier population of the indirect minima is especially severe for shorter wavelength devices such as amber and green. This poor carrier confinement results in decreased efficiency at higher current densities and at high temperatures. For example, the shaped TIP-LED devices discussed earlier can have luminous efficiencies exceeding 100 lm/W for drive current densities equivalent to conventional 5 mm lamps (40 A/cm2). However, for most applications these TIP-LED devices are typically driven at current densities two to three times higher than conventional LEDs. At these operating current densities the luminous efficiency can drop by 20% or more, depending on the wavelength and how efficiently heat is removed from the active region. Further increases in the drive current density are necessary to increase the output flux per LED, despite the efficiency penalty. However, there is a maximum drive current for AlGaInP LEDs above which the flux decreases, as junction heating effects dominate due to the increased drive power dissipated in the LED. In addition to the electron loss mechanisms described above, the decrease in luminous flux with increasing temperature also results from the decrease in the human eye response as the wavelength shifts away from the CIE peak with increasing temperature. Because of the high flux sensitivity to elevated temperatures, it is essential to ensure good heat sinking when designing a system based on AlGaInP LEDs. Thus band-gap engineering and the modification of active region design to improve carrier confinement could potentially greatly enhance LED system performance, by improving internal QE and by increasing the current and ambient temperature range of LED operation. For some applications the maximum drive current of AlGaInP LEDs is limited by the operating lifetime of the LED. The rate of flux degradation of AlGaInP LEDs over time increases with both increasing junction temperature and current density. Typical operating current densities of AlGaInP LEDs are 50200 A/cm2. Enhanced LED performance can also come from further improvements in packaging. For example, improved heat sinking would enable operation at both higher temperatures and higher current densities. The LED extraction efficiency could be significantly increased by developing alternative higher index encapsulants. For example, the typical refractive index of the current commercial LED encapsulant is 1.5, less than half the value of the LED chip. Increasing the encapsulant refractive index by 30% could result in a 4560% improvement in extraction efficiency, depending on chip geometry. Ways forward AlGaInP LED lamps have the highest efficiency demonstrated to date of all visible spectrum LEDs, and are commercially manufactured in large volume. The high efficiency results from technical developments in epitaxial growth, wafer bonding, packaging and device engineering. However, despite the recent breakthroughs in visible LED performance, the system dollar per lumen cost is still prohibitively high to meet requirements for general illumination. Reduction of the system cost will come through further fundamental breakthroughs or engineering development in device performance, as described above. In addition, manufacturing costs can be reduced by improvements in uniformity and reproducibility, and by identifying less costly manufacturing methods, while maintaining device performance. As these two objectives are met, the high efficiency and other inherent advantages of visible LEDs will enable large market opportunities, and have a significant impact on reducing worldwide energy consumption. Further reading Gardner et al. 2000 Appl. Phys. Lett. 74 2230. Hfler et al. 1998 Electron Lett. 34 1781. Huang et al. 1992 Appl. Phys. Lett.61 1045. Kish et al. 1994 Appl. Phys. Lett.64 2839. Krames et al. 1999 Appl. Phys. Lett.75 2365.