History of Solid-State Light Sources

Transcription

1 History of Solid-State Light Sources Oleg Shchekin and M. George Craford Contents Introduction Early Pioneers Evolution of Visible III-V LED Technology GaN/AlInGaN Materials and Devices Continual Improvement in LED Performance Laser-Based Solid-State Sources LED Packaging Conclusion References Abstract The second decade of the twenty-first century has become the period when lightemitting diodes (LEDs) are beginning to reach their promise of being the dominant technology for generating light. For those who have worked in this field, it is gratifying to see the enthusiasm of consumers for the technology and how the rate of adoption often outpaces expert projections. At the time of writing, leading manufacturers of lighting products achieve more than 40 % of their revenue by sales of products that use LEDs as the light source, and this portion is projected to be over 80 % by The rapid adoption of LEDs is largely due to the tremendous improvement in underlining technology which enabled cost reductions and increase in functionality of lighting systems. As novel as LEDs may O. Shchekin (*) M.G. Craford Lumileds, San Jose, CA, USA oleg.shchekin@lumileds.com # Springer International Publishing Switzerland 2017 R. Karlicek et al. (eds.), Handbook of Advanced Lighting Technology, DOI / _63 41

2 42 O. Shchekin and M.G. Craford seem, the related discoveries and technology development have over a centurylong history, with generations of researchers working on fundamental exploration without which the present day successes would not have been possible. At the same time, the present state of the art of LED technology and manufacturing lacks the homogeneity of the more established fields such as conventional light sources or silicon integrated circuits. This chapter will review the pioneering work, with more focus given to the developments of the most recent decades. An outlook to the directions that the field may take will be provided as well. Introduction The second decade of the twenty-first century has become the period when lightemitting diodes (LEDs) are beginning to reach their promise of being the dominant technology for generating light. For those who have worked in this field, it is gratifying to see the enthusiasm of consumers for the technology and how the rate of adoption often outpaces expert projections. At the time of writing, leading manufacturers of lighting products achieve more than 40 % of their revenue by sales of products that use LEDs as the light source (Philips Results Q3 2014), and this portion is projected to be over 80 % by The rapid adoption of LEDs is largely due to the tremendous improvement in underlining technology which enabled cost reductions and an increase in functionality of lighting systems. The technological advances have produced commercial components with luminous efficacy over 200 lm/w. Such high efficiency allows not only for energy savings but also for the reduction in system cost as fewer LEDs and less heat sinking are needed to produce a desired amount of light. The growing production volumes and the beginning of standardization of technology and manufacturing methods are reducing the cost of LED components, further lowering barriers for adoption of solid-state lighting. Besides the energy savings, the basic flexibility of LED technology invites new possibilities such as choice in light color and the natural integration with electronic components enabling control systems such as occupancy detection and communication via modulated light. The long life and high brightness offer design flexibility as well as added safety for automotive applications, while the microelectronic nature of LEDs is even utilized for data transfer and communication (Philips creates shopping assistant with LEDs and smart phone; Elgala et al. 2011). Such growing adoption and awareness of the LED technology by the society motivate reviews of the history of innovation in the field, which is the focus of this chapter. As novel as LEDs may seem, the related discoveries and technology development have over a century-long history, with generations of researchers, often working in challenging conditions on fundamental exploration without which the present-day successes would not have been possible. At the same time, the present state of the art of LED technology and manufacturing lacks the homogeneity of the more established fields such as conventional light sources or silicon-integrated

3 History of Solid-State Light Sources 43 circuits. Therefore, we are writing this manuscript sometime half to three quarters of the way in LED evolution from a laboratory novelty to a common component, depending on application. This chapter will review the pioneering work in the field, where the reader will be referenced to some detailed accounts written previously (often by contemporaries of a period). More focus will be given to the developments of the most recent decades. An outlook to the directions that the field may take will be provided as well. Early Pioneers The first recorded observation of electroluminescence was made in 1907 by Henry Round. The early twentieth century was the time of rapid development in radio, and Round and others were experimenting with solid-state rectifying detectors (crystal detectors). He noticed that applying a voltage between two point contacts on the surface of carborundum (SiC) resulted in the crystal giving off a yellowish light. Round recorded his observations in a note to the editors of Electrical World (Round 1907), in which he inquired about the existence to references of similar observations and also theorized about the origins of the light. Round experimentally identified the dependence of the emission on the bias voltage and asserted that the emission is not connected with heating. Now we know that Round formed a Schottky contact and the light was generated by injecting carriers across the junction at bias. The research on electroluminescence of SiC was independently restarted by Oleg Losev in A very informative account and analysis of Losev s work are given by Loebner (1976). Similar to Round, Losev observed luminescence from a point metal contact against the surface of SiC, but unlike Round, he continued the study of the phenomenon and published four patents and 16 papers related to the topic. Losev made a few very important observations. He showed that the wavelength of light emitted by the diodes was, effectively, the applied voltage minus the resistive voltage drop and postulated that the emission was an inverse of the photoelectric effect. He also understood that the mechanisms for light emission in forward and reverse bias were distinct and showed that the emission of light was coming from a thin layer beneath the surface of the crystal. He characterized the active part of the crystal to consist of the three distinct conductive layers, which, in retrospect, was an observation of a p-n junction, except he described all of the layers as n-type stopping short of discovering hole conduction in silicon carbide. Besides the academic investigations, Losev envisioned practical uses for his discovery. He filed a number of patents outlining methods for the use and application for the light-emitting SiC crystal detectors. One particular patent for a light relay, filed in 1927, is truly visionary (Losev 1929). Here Losev teaches the use of the crystal detector as a high-speed modulated light source to record information by exposing a moving photographic plate to the emitted light. The detector itself is connected to a modulating circuit to which the signal is transmitted with or without a wire, and the fast modulating property of the light-emitting crystal detector is

4 44 O. Shchekin and M.G. Craford leveraged for high-speed information transfer and recording. The invention is many decades ahead of the age of telecommunication and optical storage. It is worth mentioning that Losev did not have a formal university education and was a 19-year-old technician in Nizhegorodskaya Radio Laboratory at the time of his discovery of luminescence in SiC. He continued to work as a technician during his carrier in various Soviet radio laboratories but was given a Candidate of Science degree (Soviet equivalent of a PhD) by the Ioffe Institute in 1938 without a proper dissertation. He passed away from hunger in 1942 during the WWII blockade of Leningrad at the age of 33. At the time, Losev was busy working on a three-contact semiconductor system to replace vacuum tubes (Zheludev 2007). He intended to publish a manuscript on an important silicon device, but due to the blockade, he could not send the document out to the evacuated office of the journal Soviet Physics. If the manuscript is ever found, we may see if Losev was on the path to inventing the transistor. The modern explanation of the electroluminescence in SiC was formulated in 1951 by Kurt Lehovec, almost 20 years after Losev s initial observations. It is important to note, though, that the understanding of light-emitting diodes would not be possible without the discovery and the description of transistor in 1947 by John Bardeen and Walter Brattain (Brattain et al. 1948). The transistor validated the existence and injection of holes and that the injection of current into a semiconductor creates non-equilibrium in electron hole populations, with carriers of each type being injected over the energy barrier (minority carrier injection). Lehovec et al. (1951) revisited Losev s SiC luminescence work and postulated that the process of light emission consists of carrier injection of over a p-n barrier, with a direct reference to the transistor effect, while light emission results from the recombination of these carriers across the bandgap (see Fig. 1). A possibility of non-radiative recombination Fig. 1 Reproduction of Fig. 8 from reference (Lehovec et al. 1951) illustrating the explanation for light emission in silicon carbide as a result of carrier injection over the p-n barrier followed by radiative or non-radiative recombination of the carriers across the bandgap (Copyright (1955) by The American Physical Society)

5 History of Solid-State Light Sources 45 was also stated where the energy would be released as heat. This description of the mechanism of light emission in a diode has served as a guiding principle in the development of LEDs. The decade of the 1950s was dominated by research on electroluminescent ZnS for possible use in flat television panels. Even though luminescent II-VI materials only found use in niche applications, the field did keep enough bright people employed for important progress to be made toward the present-day mainstream LED technology. A number of efforts particularly stand out. One is the work by Michael Schon in 1953, where Schon revisited electroluminescence of SiC and separated the efficiency of the process into injection efficiency of minority carriers and, separately, efficiency of recombination at and near the p-n junction. This distinction structured the way LED and laser designers go about optimizing the electrical efficiency of emitters. Also significant is the demonstration of electroluminescence in GaP by Wolff et al. in This marked the beginning of the work on III-IV semiconductor-based LEDs and motivated further investigations of GaP luminescence in a number of laboratories (Loebner 1976). The 1960s is the decade of the first practical implementations of LEDs using materials and principles which are part of the basis of LED technology in use today. A number of demonstrations of GaAs-based infrared LEDs were made in 1962 (Biard and Pittman 1966; Hall et al. 1962; Pankove and Berkeyheiser 1962; Pankove and Massoulie 1962; Quist et al. 1962). Biard and Pittman, at Texas Instruments, filed a patent on August 8, 1962 in which they detailed the structure of a GaAs p-n junction LED (Biard and Pittman 1966). Biard and Pittman noticed infrared emission in 1961 when working on tunnel and varactor diodes based on GaAs. This led them to think about how to design the semiconductor and diode contact for efficient extraction of light, which was described in the 1962 patent. The same year Texas Instruments released the first IR LED product (TI-SNX-100) which found its first use in IBM punch card readers. GaAs is a direct-bandgap semiconductor which, unlike earlier indirect-bandgap SiC and GaP LEDs, has potential to be a high quantum efficiency emitter. The missing link from the standpoint of lighting was the demonstration of a directbandgap visible emitter. The first direct-bandgap visible (red) LEDs were reported in 1962 by Holonyak and Bevacqua (1962) at General Electric using GaAsP, an alloy of GaAs and GaP. They also demonstrated a red current-injected laser at 77 K. GaAsP had been studied by Ehrenreich at General Electric (Ehrenreich 1960) who documented the bandgap behavior versus the As/P fraction. However, experts in the field had been doubtful that crystal quality suitable for electronic devices could be achieved. Holonyak and Bevacqua demonstrated that by using closed-tube vapor transport, they could achieve suitable crystals. Holonyak had the vision to recognize that this work could lead to a practical light source although much more experimental work must be done (Readers Digest 1963). All of the high-performance LEDs and injection lasers utilized today follow from Holonyak s work and are based on semiconductor alloys (Fig. 2). After this work was demonstrated, Holonyak was visited by R.A. Ruehrwein of Monsanto Company. Ruehrwein recognized that the closed-tube growth technique

6 46 O. Shchekin and M.G. Craford Fig. 2 Image of the first red-emitting GaAsP alloy laser demonstrated by Prof. Nick Holonyak, Jr., at General Electric in 1962 (Image courtesy of Nick Holonyak) Laser Diode Wire Bond Contact Red Laser Light 0.2 mm Scattered Light Heat Sink used by Holonyak could be replaced with an open-tube chemical vapor deposition (CVD) process that would be much more scalable for manufacturing. In the following years, Monsanto implemented the large-scale epitaxial growth of GaAsP and in the 1970s became the world s largest supplier of compound semiconductor materials, particularly GaAsP for applications such as calculators and watch displays, prior to the development of LCDs. Evolution of Visible III-V LED Technology The indirect GaP and phosphorus rich GaAsP semiconductors were used also to generate light of other colors in addition to red. In 1965, Thomas et al. (1965) demonstrated that if GaP is doped with an optically active isoelectronic impurity, the spatial localization of carriers at the impurity spreads the wavefunction in momentum space, which increases the probability of an optical transition with one of the bands. Thomas and his team at ATT Labs used nitrogen-doped GaP to generate green emission and in 1967 Logan et al. demonstrated red-emitting GaP using Zn and O pair emission (Logan et al. 1967). In the early 1970s, the team at Monsanto used nitrogen doping in GaAsP to achieve orange and yellow emitters as well as 10 brighter red than those based on direct GaAsP (Groves et al. 1971; Craford et al. 1972). GaAsP:N had higher performance than GaAsP primarily due to improved light extraction from the chip. The emission from the nitrogen sites was below the energy gap, and the indirect semiconductor had less absorption than direct GaAsP. The GaAsP:N was also grown on a transparent GaP substrate instead of GaAs. The CVD grown nitrogen-doped GaAsP:N and nitrogen- and ZnO-doped GaP, grown using LPE, were the main materials systems for visible LEDs through the early 1980s (Fig. 3). In the early 1980s, AlGaAs heterojunctions became the highest performance red LEDs. AlGaAs had been studied using liquid-phase epitaxy (LPE) growth since the

7 History of Solid-State Light Sources 47 Fig. 3 Examples of some of the first products using LEDs. The Hamilton Pulsar wristwatch was announced in 1970 and used GaAsP LEDs in its display. Hewlett-Packard and Texas instruments offered a series of programmable calculators GaAsP with LED displays. LED use was short lived for both of these applications as the LEDs consumed quite a bit of power and were not bright enough in direct sunlight. Liquid crystal displays eventually replaced LEDs in calculators and wristwatches 1960s (Alferov et al. 1969; Rupprecht et al. 1967). It was seen to have advantages over GaAsP by offering an opportunity to combine direct-bandgap heterostructures in a lattice-matched system. This combination was expected to result in devices with high internal efficiency due to the lattice-matched system and the confinement of carriers by heterostructures and high extraction efficiency since the confining layers would be of higher bandgap and would not absorb light from the active region. Unfortunately, the high-volume CVD could not be used because the aluminum attacked the hot quartz walls of the growth chamber. The main challenge with state-of-the-art LPE was creating high-quality crystals in high volume. The temperature-difference LPE technique developed by Nishizawa in Nishizawa et al. (1977) was used by Stanley Electric to introduce and manufacture highperformance red LEDs. The combination of excellent crystal quality, efficient direct-bandgap recombination, and good light extraction due to the heterostructure yielded devices with a 10 improvement in performance. Stanley Electric demonstrated luminous efficacies up to 10 lm/w, and other manufacturers soon followed. The 10 lm/w barrier may not be impressive today, but it was an important milestone in the evolution of LEDs as this efficacy, along with the low voltage and size, made LEDs attractive for use in a variety of practical outdoor applications such as traffic lights, displays, and car taillights. However, AlGaAs materials, proved to

8 48 O. Shchekin and M.G. Craford have severe reliability issues because the aluminum, at concentrations necessary for establishing a useful heterostructure, became readily oxidized, and LEDs failed. The next improvement in visible LED technology came with the quaternary alloy AlInGaP. (Al x Ga 1-x ) 0.5 In 0.5 P heterostructures can be lattice matched to GaAs and made to emit from red to green. At Al compositions near 53 %, the semiconductor becomes indirect; therefore, the emission near green is inefficient. However, for longer wavelengths, the direct bandgap and confinement of the active region by the heterostructure promised high radiative recombination and photon extraction efficiencies. Additionally, the lower aluminum content was expected to alleviate the reliability issues associated with oxidation in AlGaAs LEDs. In the mid-1980s, the AlInGaP alloy was being studied for use in semiconductor lasers (Kobayashi et al. 1985; Ikeda et al. 1986; Ohba et al. 1986; Itaya et al. 1990) but was not being pursued for LEDs. LPE and VPE were the dominant high-volume, low-cost crystal growth technologies in LED production with epi cost of around $10 per square inch (Craford 2013). For thermodynamic reasons, LPE was found to be unsuitable for high-volume, high-yielding growth of AlInGaP, and as in the case of AlGaAs, CVD had an issue with aluminum attacking the quartz walls of the chamber. The technique which was being used to grow AlInGaP was metal-organic chemical vapor deposition, or MOCVD. MOCVD was pioneered by Manasevit at Rockwell (Manasevit 1968). Unlike in CVD, where the walls of the reactor are kept hot, MOCVD heats the growth substrate to facilitate cracking of metal-organic precursors and growth of epitaxial films. In 1977, R. D. Dupuis and P. D. Dapkus demonstrated room-temperature operation of AlGaAs lasers grown by MOCVD with the performance matching the best of those grown by LPE. The results were obtained by a greatly improved MOCVD process and showed that MOCVD could be used for the generation of high-performance semiconductor devices. In the mid-1980s, MOCVD was still a relatively expensive low-capacity technique thought to be useful for lasers but not commercial LEDs. Hewlett-Packard started a program developing AlInGaP LEDs believing that there were no fundamental barriers to developing MOCVD into a high-volume manufacturing technology. In 1990, the HP team announced AlInGaP LEDs operating at 10 that of existing yellow LEDs and equivalent to that of best AlGaAs devices (Kuo et al. 1990). The first devices had the absorbing GaAs substrate but had a thick, transparent, and lattice mismatched GaP layer on top to give current spreading and improve light extraction. The AlInGaP-based LEDs are the dominant technology for red/orange direct color emitters today. The epitaxial growth is based on AlInGaP lattice matched to GaAs, which creates challenges for maximizing emitter efficiency. The emission wavelength of the heterostructure is varied by adjusting the aluminum-gallium ratio and is limited to aluminum concentrations below 53 %, at which the bandgap becomes indirect. Additionally, the confinement of electrons worsens with higher aluminum content, resulting in strong temperature sensitivity of the internal efficiency. Therefore, for practical applications, the AlInGaP material system is restricted to wavelengths 580 nm and longer. In addition, in the early AlInGaP devices, the extraction of the photons was hindered by the absorbing substrate and the relatively high index of refraction of the AlInGaP semiconductor. A series of

9 History of Solid-State Light Sources 49 Fig. 4 Evolution of high extraction efficiency AlInGaP emitters: (a) thick GaP window layer with absorbing GaAs substrate; (b) diode with thick window layer and transparent substrate; (c) TIP chip architecture; (d) thin-film architecture advances in the optical efficiency was made by teams from Hewlett-Packard. In the 1992 demonstration by Huang et al. (1992), the extraction efficiency of the LEDs was improved by 2 by utilizing an even thicker GaP window, allowing for improved current spreading and extraction of photons before they are absorbed by GaAs substrate. In 1999, Kish et al. eliminated the absorbing substrate all together by etching it away and bonding AlInGaP to the GaP transparent substrate, raising the extraction efficiency to an estimated % (Kish et al. 1994). In 1999, Krames et al. have further improved the extraction efficiency of the die, by introducing tapered inverted pyramid (TIP) chip architecture (Krames et al. 1999). The taper frustrated the TIR rays inside the LED and resulted in further 40 % improvement in efficiency with an estimated photon extraction efficiency of 60 %. The TIP chip was the first LED to achieve 100 lm/w. The TIP chip and transparent substrate architectures have high extraction efficiencies and by design emit photons over a large area and into a wide solid angle. This makes these architectures not ideal for applications where secondary optics in a lighting system are used to collect the light from the LED and shape or focus the light with high flexibility. For such purposes, a two-dimensional surface emitter is much more optimal. The need to address such applications has resulted in the introduction of thin-film architectures by a number of manufacturers. Figure 4d schematically shows a thin-film architecture for a red AlInGaP die. Here an AlInGaP epi structure has a metal contact deposited and bonded to a carrier substrate. The carrier substrate is often GaAs, Ge, or Si. After bonding, the growth substrate is removed, and the exposed AlInGaP surface is roughened for enhanced photon extraction. The thin, high-index film creates challenges for extraction of photons, and significant effort has to be put into die and epi-layer design, but high extraction efficiencies have been demonstrated (Streubel et al. 2002; Broell et al. 2014). GaN/AlInGaN Materials and Devices In 1968, the Radio Corporation of America, the leading television manufacturer, launched an internal program to develop blue LEDs. At that time, the technology for red- and green-emitting LEDs had been demonstrated and was becoming available.

10 50 O. Shchekin and M.G. Craford Developing a blue LED would allow for creation of a flat color television. Herbert Paul Maruska was a scientist at RCAwho was asked by James Tietjen, the director of the company s central labs at the time, to look into gallium nitride as the basis material for blue LEDs. There are a number of reviews of the development by Maruska himself as well as by others (Schubert 2006; Maruska). By then, GaN had been prepared as powder, its crystal structure was documented, and a variety of dopants were experimented with. No one had tried the VPE growth of GaN, so Maruska, who had a reactor for growth of GaAsP, replaced the arsine bottle with ammonia. He chose sapphire as the substrate because it was available and would not react with ammonia. In 1968, after some failed attempts, Maruska succeeded in growing the first monocrystalline film of GaN by turning up the growth temperature to 850 C. The GaN films were strongly n-type without intentional doping, but the initial attempts for p-doping proved unsuccessful. Maruska and Tietjen published the work in 1969 (Maruska and Tietjen 1969). In 1971 Jacques Pankove and Ed Miller, Maruska s colleagues at the RCA labs, demonstrated a GaN-based LED using a metal-insulator-semiconductor (MIS) diode structure where the insulating layer was created by doping GaN with zinc following earlier work by Maruska. These devices emitted green light and were the first GaN-based LEDs (Pankove et al. 1971). At that time, Maruska was enrolled in a PhD program at Stanford, where he continued his work on the GaN-based LEDs, sponsored by RCA. There he experimented with magnesium doping of GaN, which he believed would be a better p-dopant to achieve blue emission. In July 1972 he successfully demonstrated the first blue/violet current-injected GaN-based LEDs emitting at 430 nm (Maruska et al. 1972). The efficiency of the emitters was very low since these LEDs were effectively metalinsulator-semiconductor diodes as the Mg-doped GaN did not show p-type conductivity. The Stanford-RCA team published a number of studies where they attributed the luminescence to impact ionization of the magnesium dopant by electrons injected by tunneling through the insulating regions (Maruska and Stevenson 1974; Pankove and Lampert 1974). Maruska returned to RCA the following year. Unfortunately, due to budget concerns at the time, the blue LED project was canceled. The LEDs created by the RCA effort were not yet ready for commercialization, and more discoveries were needed to make these LEDs practical, but the technology of GaN on sapphire and magnesium doping is the foundation of all of the GaN-based LEDs today (Fig. 5). The work on GaN-based LEDs slowed considerably in the second half of the 1970s and the early 1980s. However, a few groups of researchers continued to focus on two main issues limiting the LED performance: crystal quality and p-doping. In 1983, Yoshida et al. reported on the effectiveness of using AlN buffer for the growth of GaN films on sapphire using reactive MBE (Yoshida et al. 1983). In 1986, Amano et al. (1986) demonstrated and quantified the improvement in the quality of MOVPE GaN films grown on sapphire by means of a thin AlN buffer layer. In the initial demonstration, followed later by a detailed study (Hiramatsu et al. 1991), the team showed that a thin AlN layer, deposited at relatively low temperatures, crystallizes, upon heating of the growth substrate, into columnar structures which serve as nucleation sites for the growth of GaN along the c-axis. As GaN is deposited, the

11 History of Solid-State Light Sources 51 Fig. 5 The first blue LED based on Si- and Mg-doped GaN on sapphire substrate developed by Maruska (Photo: Herb Maruska) multitude of growth islands coalesce, eventually forming a high-quality monocrystalline film. The quality of the crystal was reflected in the much narrower width of X-ray rocking curve peaks and lower background electron concentration. Later, high-quality growth of GaN was also achieved with low-temperature GaN buffer layers by Nakamura (1991) using MOCVD and by Moustakas using molecular beam epitaxy (MBE) (Lei et al. 1991). The breakthrough on p-doping of GaN was announced in 1989 by Amano et al. who discovered that the conductivity of GaN doped with Zn or Mg dramatically increased after the material was irradiated with a beam of low-energy electrons. This demonstration was an important step toward GaN-based p-n junction LEDs which were also reported in the same paper. The mechanism responsible for the improved conductivity was explained by Nakamura et al. (1992) who later showed that a similar effect can be achieved by thermal annealing of Mg- or Zn-doped GaN. The low-energy electrons or thermal treatment break the Mg-H complexes which form during crystal growth in which the dopant is passivated by hydrogen. Annealing is a commonplace, large-volume manufacturing technique, and the demonstration by Nakamura was an important step toward industrialization of GaN p-n junction-based devices. In 1993, Nakamura, who was employed as an engineer at Nichia Corporation, demonstrated, for the first time, high-quality InGaN heterostructures as well as GaN/InGaN double-heterostructure LEDs grown on sapphire with MOCVD (Nakamura et al. 1993). The blue-emitting LEDs had the external quantum efficiency of 0.22 %. A year later, Nakamura and his team followed up with an InGaN/ AlGaN heterostructure LED, which now showed an EQE of 2.7 %. This demonstration of the broad range of nitride alloys allowed for subsequent engineering of GaN-based LEDs to much higher efficiencies. The use of MOCVD, sapphire substrates, and the thermal activation of magnesium enabled the leveraging of the existing manufacturing tools for the rapid development and proliferation of GaN-based blue, UV, and green LEDs. In 2014, Isamu Akasaki, Hiroshi Amano, and Shuji Nakamura were awarded Nobel Prize in Physics for their numerous key contributions to the development of high-efficiency blue LEDs.

12 52 O. Shchekin and M.G. Craford The significance of the development of high-efficiency blue LEDs, and the motivation behind the recognition by the Nobel Prize committee, is due to their direct application to high-efficiency white-light sources. In 1970, Bell Labs was granted a US patent 3,691,482 (Pinnow and Gerard Van Uitert 1970), where the inventors described the use of YAG and various other down-converting materials to generate white as well as colored light. Generation of white light, in particular, was taught as a combination of light from a phosphor excited by blue laser and a portion of the laser light which was not converted. With the demonstration of efficient blue LEDs, the ideas for generation of white as well as different-colored LEDs using down conversion followed (Shimizu et al. 1999; Butterworth and Helbing 1998; Höhn et al. 2001; Baretz and Tischler 2003). Continual Improvement in LED Performance The introduction of white emitters based on blue LEDs started the proliferation of solid-state lighting into various applications which influenced LED designs and motivated improvements in efficiency. To better illustrate the development of LED sources through the 1990s, it is worth introducing the main elements comprising the luminous efficacy of LEDs. The LED luminous efficacy, η L, can be defined as the product of the following efficiencies: internal quantum efficiency (IQE) of carrier recombination in the active region, extraction efficiency (EXE) of photons from the blue LED, electrical efficiency (ELE) of injecting carriers into the active region, and efficiency of converting (CE) blue photon into the desired LED spectrum (Krames et al. 2007): η L ¼ IQE EXE ELE CE The conversion efficiency (CE) can then be broken down further into package efficiency (PE), Stokes shift penalty (QD), luminous equivalent of emission spectrum (LE), and quantum efficiency (QE) of down conversion by phosphors. Figure 6 illustrates the main interactions between the building blocks of an LED (blue die, package, epi, and converter) and how these affect luminous efficacy. The evolution of the GaN-based chip has been driven by applications and a general need for greater LED efficiency. Figure 7a shows the schematic of a blue LED of the type used in the original demonstration of the blue doubleheterostructure LED by Nakamura (Nakamura et al. 1994). The main feature of this structure is the semitransparent Ni/Au contact which is used to facilitate uniform injection of current on the p-side. While helping to spread the current, the contact absorbs photons and limits the extraction efficiency of the structure. In the late 1990s and early 2000s, two directions in LED architectures emerged. One focused on low cost and low power per emitter for use in indicators and mobile displays. The other focused on high-power applications which valued the number of lumens per LED such as traffic and automotive signaling and, later, automotive forward lighting, illumination, projection, and camera flash. The chip in Fig. 7a was well suited for

13 History of Solid-State Light Sources 53 LED building blocks IQE ELE EXE CE PE LE QD QE epi (substrate, emission wavelength, heterostructure) die (type, p-, n- contacts, interconnects) package (type, encapsulants, interconnects) converter (phosphors, encapsulants) Fig. 6 LED building blocks and their impact on the efficacy. The blue cells indicate a component of luminous efficiency affected by a particular LED building block a Semi-transparent Ni/Au contact b Transparent conductive oxide contact p-gan p-gan n-gan Sapphire n-gan Sapphire c Sapphire d e Sapphire n-gan p-gan n-gan p-gan n-gan p-gan Ceramic Submount Ceramic Submount f Cathode contact n-gan g Anode contact Anode contact p-gan Cathode contact Fig. 7 Evolution of the GaN blue die and proliferation of die types: (a) conventional chip; (b) lateral die (LD); (c) flip chip; (d) thin-film flip chip (TFFC); (e) chip-scale package (CSP); (f) vertical thin film (VTF); (g) embedded-contact VTF (EC-VTF) low-power applications. To increase the extraction efficiency of the die, certain changes were adopted and proliferated through the industry during the past decade. Figure 7b shows an example of a conventional architecture, which is similar to that in Fig. 7a, except that the semitransparent p-contact has been replaced with a transparent conductive oxide, such as indium tin oxide (ITO), and the epi is grown on patterned sapphire. Due to lateral direction of current spreading in this architecture, it is often called a lateral die (LD). Often, to reduce the parasitic interaction

14 54 O. Shchekin and M.G. Craford between the die and a package, there is a reflector deposited on the sapphire surface. The patterned sapphire, combined with the lower loss from the transparent oxide, can yield architectures which achieve very high extraction efficiencies, upward of 88 % (Narukawa et al. 2010). The LD in Fig. 7b came to dominate the low- and mid-power needs of the display industry. While low cost and straightforward, the die and the package, which will be discussed in detail later in the chapter, are limited in how much power density an LED can handle. Therefore, the high-power LEDs and the die followed different architecture directions. The flip-chip architecture (FCLED), shown in Fig. 7c, allowed for excellent current spreading and heat extraction while enhancing extraction efficiency by utilizing a large area, highly reflective, silver-based p-contact to reflect photons out of the LED and also allow a large fraction of the photons to avoid any interaction with the metal contacts (Wierer et al. 2001). The extraction of heat was facilitated by means of metal interconnects between the die and a submount. The proximity of the p-contact to the quantum wells in FCLEDs was later used by Shen et al. to further enhance the extraction efficiency by tuning the dipole radiation pattern to maximize the energy coupled into the GaN/sapphire escape cone (Shen et al. 2003). Roughening of the high-/low-index interfaces had been proposed and utilized for IR LEDs as the means to break the total internal reflection which was limiting the extraction efficiency of LEDs (Stern 1964; Joyce et al. 1974; Schnitzer et al. 1993). In 1998, Wong et al. demonstrated the separation of a sapphire substrate from a GaN film using a UV (excimer) laser pulse (Wong et al. 1998) and applied this technique to demonstrate thin-film LEDs (Wong et al. 1999). In 1996, Minsky et al. (1996) demonstrated that a GaN surface could be photo-electrochemically etched. Laserassisted lift-off (LLO) and PEC etching of the consequently exposed n-gan were then used to significantly enhance the extraction efficiency of GaN-based LEDs. In 2004, two teams, from UCSB and from OSRAM, demonstrated this approach in a so-called vertical thin-film (VTF) die architecture, shown in Fig. 7f. The VTF architecture was fabricated by bonding the metallized p-side of the GaN to a conductive submount while the substrate on which GaN has been grown is removed, followed by roughening of the exposed n-gan and deposition of a patterned n-contact. At that time the teams reported an almost doubling of the extraction efficiency of un-encapsulated chips (Fujii et al. 2004; Haerle et al. 2004). Application of lift-off and PEC to a flip-chip LED enabled the realization of the thin-filmflip-chip architecture (TFFC) with an extraction efficiency over 80 % reported by a Philips Lumileds team (Shchekin et al. 2006; Krames et al. 2007). The TFFC parts had a higher extraction efficiency than VTF because the emitting surface was not occluded by the n-contact on top of the n-gan as was the case with VTF. Also, because of the absence of wire bonds, TFFC chips could be combined into compact high-brightness arrays and were readily suited for use with high-performance ceramic phosphors. The need for higher extraction efficiency in the VTF architecture motivated introduction of what could be called an embedded-contact VTF (EC-VTF) shown in Fig. 7g. In this architecture, the n-contact features resemble those previously found in flip-chip and TFFC LEDs but still feature a wire bond

15 History of Solid-State Light Sources 55 similar in appearance to VTF. It is important to note that in the industry, the thin-film architectures are also made in high volume with epi grown on substrates which can be chemically removed or thinned such as silicon, silicon carbide, and GaN. Due to the excellent heat dissipation ability and high extraction efficiency, flip-chip and the thin-film architectures (TFFC, VTF, and EC-VTF) have become the dominant die architectures for the high-power LEDs. The drawback of the thin-film based emitters is that there is need for an additional substrate to provide mechanical support. As discussed, in case of the VTF or EC-VTF, the support comes from an intermediate substrate carrier and ultimately requires a separate substrate which provides electrical contacts via wire bonds. In the case of TFFC, there is a separate submount/substrate which provides electrical connections and also mechanical support. Figure 7e shows an example of a chipscale package (CSP) approach, where a flip-chip architecture is fabricated from epi grown on patterned sapphire. The sapphire is left on and provides mechanical stability such that the die itself can have solder pads and be handled as a standalone package. Such a CSP approach combines the high-performance attributes of flip chip, robustness of patterned sapphire, and compactness without a need for an intermediate substrate. In the late 1990s, it became apparent that for the InGaN/GaN LEDs grown on sapphire substrates, the internal quantum efficiency (IQE) of the radiative recombination of charge carriers peaked at fairly low current density and then monotonically decreased with current drive (Kim et al. 2001; Nakamura and Fosol 1998). The origins of this droop have been a rich topic of research and discussion. The phenomena has been attributed to spillover of carriers (Kim et al. 2007; Bochkareva et al. 2010; Özg ur et al. 2010), recombination at dislocations (Monemar and Sernelius 2007), and Auger recombination (Shen et al. 2007). Although the origin of the droop is still an open question, the Auger theory is finding growing support (Kioupakis et al. 2011). A number of directions are being pursued by researchers in the field to reduce droop, with the efforts focused on alternative growth substrates, such as nonpolar and semipolar GaN (Chakraborty et al. 2005; Zhao et al. 2011) and optimized active region design. YAG:Ce has been one of the earliest and most enduring choices of phosphor materials for making white LEDs. It is efficient, stable, relatively low cost, and excitable in blue and has emission broad enough to produce light of reasonable quality. Introduced in products by Nichia Corporation around 1996, it is still found in LEDs for a broad range of applications. However, using YAG alone for making white LEDs allows only CCTs of 4000 K and above with CRI 70. For warmer CCTs and higher CRI, approaches were investigated using combinations of at least two phosphors: green and red (Mueller and Mueller-Mach 2000). Some of the combinations considered early on included (Ca, Se)S:Eu for red and thiogallate for green. While these materials could produce spectra with excellent color rendering and high lumen equivalent, there was a need for robust red phosphors without some of the shortcomings of thiogallates and sulfides, such as poor performance at high drive and stability in humid environments. In 1995, Schlieper et al. (1995) reported the synthesis of nitride-silicate Sr2Si5N8 and Ba2SiN8, (258) europium-activated

16 56 O. Shchekin and M.G. Craford Fig. 8 Illustration adapted from Ref. (Shchekin et al. 2010) showing how the yellow gap in the directemitter efficiency can be bridged by fully converted phosphor LEDs nitride phosphors. In 2006, Uheda et al. (2006) reported nitride CaAlSiN3:Eu red phosphor for use in white LEDs. These nitride reds have proven to have higher quantum efficiency and greater stability than either sulfides or orthosilicate (Tasch et al. 2001) phosphors while offering a broad range of emission wavelengths accessible by tuning of the stoichiometry of the material. The nitride phosphors have also created opportunities for new applications and further improvements in lm/w. In 2009, a team from Philips Lumileds demonstrated and productized a fully converted amber-color-emitting pcled (Mueller-Mach et al. 2009). The device used a 258-nitrido-silicate phosphor sintered into a ceramic plate placed onto a TFFC blue pump LED. The resulting amber LED, using the higher efficiency blue pump, outperformed direct-emitting amber LEDs based on AlInGaP. Since the nitride can be prepared as a ceramic, it is possible to reduce the scattering of blue light from the phosphor, making this approach more efficient and resulting in color with greater saturation than fully converting blue using powder phosphors. Figure 8 shows how the yellow gap in the efficiency of direct-emitting color LEDs can be spanned by fpcleds utilizing various nitrides (Bechtel et al. 2010). Currently, the combinations of nitride reds and aluminum/gallium garnet yellow and green phosphors are the state of the art for efficient phosphor conversion in white LEDs. Quantum efficiency of conversion is over 90 % for 1W power LED emitters and higher for low-power emitters. The remaining quantum efficiency losses come from photon absorption in the package and are worsened by mechanisms impeding photon extraction out of the package, such as scattering and Fresnel reflections. Figure 9 shows the breakdown of efficiency components for a representative, as of this writing, warm white LED in comparison to practical limits. Here the practical limits for epi IQE and phosphor quantum efficiency can be debated, as the limiting mechanisms are not universally agreed on, while the quantum deficit and lumen equivalent are for LED spectra with CCT of 3000 K, CRI of 80 and red rendering index (R9) greater than zero.

17 History of Solid-State Light Sources 57 Internal Quantum Efficiency Electrical Efficiency Extraction Efficiency Package Efficiency Quantum Deficit QE of Downconversion 0% 20% 40% 60% 80% 100% Lumen Equivalent Efficacy Lm/W Practical Limit values listed for 3000K 80CRI phosphor converted LEDs at J=350mA/mm 2 and Tj=85C Fig. 9 Efficacy breakdown for a typical warm white phosphor-converted LED at J = 350 ma/mm 2 and Tj = 85 C For phosphor-converted LEDs, the most significant areas for improvement in lm/w are in epi IQE and the shape (lumen equivalent) of the white emission spectrum. The limitations of the current red nitride phosphors lie in the width of the emission spectra. At nm FWHM, these phosphors generate a considerable amount of light at wavelengths to which human eye has low sensitivity. Figure 10 shows the human eye photopic response curve against a comparison of the emission spectrum of a typical warm white LED with common nitride red phosphors and a hypothetical spectrum for the same CCT, but with the red phosphor with FWHM of 30 nm. The difference in lumen equivalent between the two spectra is 17 %. The anticipated lumen gain depends on CCT and color rendering specification and is generally %. The optimal narrow red phosphors, with high efficiency, stability, and emission wavelength for maximum lumen equivalent, do not exist yet, but there are some candidates emerging. Quantum dots, which, have been

18 58 O. Shchekin and M.G. Craford Spectral Power Distribution (noramalized) CCT ~ 2700K Photopic Response 30nm FWHM red, 90 Ra, LE ~ 370 lm/wopt today ~ 80 Ra, LE ~ 315 lm/wopt Lumen loss of current Phosphors Wavelength (nm) Fig. 10 An illustration of lumen benefit from reducing the width of the red phosphor spectrum considered for use in LEDs for some time (Bawendi et al. 2002), have been introduced in display applications in configurations where the converting element is not in immediate contact with the blue die. Mn 4+, activated fluorosilicates (Paulusz 1973; Setlur and Radkov 2010) have been proposed for use in lighting. Recently, a Eu2+-activated nitride phosphor has been announced with FWHM of 50 nm (Pust et al. 2014). This material is particularly interesting as it has potential to be readily usable in all LED emitter architectures similar to the commonplace broad-band Eu2+-activated nitride phosphors. Other ways to generate white light with LEDs, besides phosphor conversion, is to mix direct red, blue, and green emitters or utilize so-called hybrid light engine, where a phosphor-converted LED with a color point slightly off-planckian is combined with a direct-emitting AlInGaP-based red LED. These approaches potentially allow the reduction or elimination of energy losses associated with phosphor conversion (Fig. 11). The RGB solution with direct color emitters is severely limited by the IQE of the AlInGaN material system in the green wavelengths. Additionally, for high color rendering, one would need to add a direct yellow emitter, for which there is no high efficiency option. Because of the limited efficiency in the yellow/green colors, the direct-emitter solutions are not used in the general lighting applications where high efficiency and color rendering are specified. However, the direct RGB emitters are widespread in architectural lighting applications, where the decorative properties of the three color LEDs offer significant reduction in power consumption over filtered light from conventional light sources. The hybrid approach, however, has been very successful in achieving high luminous efficacies, especially in applications which allow use of multiple LEDs and where there is need for significant red content and high color rendering. Recently, Philips reported a TLED prototype (LED replacement of fluorescent T lamp) with 200 lm/w at CCT K and uncompromised color rendering CRI > 80 and R9 > 20 (Details of the 200lm/W

19 History of Solid-State Light Sources 59 Phosphor-converted LEDs Blue pump LED + phospho Direct-emi ng LEDs 3 or 4 colors Hybrid LEDs Combining phosphor-converted and direct-emitting LEDs Photopic response 3000 K Planckian pc-led 3000K Photopic response 3000 K Planckian Blue Green Red Photopic response 3000 K Planckian Off-white Direct red Wavelength (nm) Wavelength (nm) Wavelength (nm) Fig. 11 Types of SSL sources for maximum luminous efficacy Efficacy (lm/w) Cool White LED Warm White LED Qualified Data, Warm Qualified Data, Cool Fig. 12 US DOE white-light LED package efficacy projections for commercial product (US Department of Energy 2014) TLED lighting technology breakthrough unraveled). The Hybrid approach has also been used in many commercial products, for example, in the DOE L-Prize winning 60 W equivalent LED bulb from Philips (Rice 2011). Figure 12 shows the lm/w efficacy projections published in the 2014 edition of the US Department of Energy Solid-State Lighting Research and Development Multi-year Program Plan (MYPP). According to the MYPP, the maximum efficiency for pcled and mixed/hybrid sources are expected to be similar at 250 lm/w with the assumption of 25 C and 35A/cm 2. Even though the hybrid light engines allow reduced phosphor conversion

20 60 O. Shchekin and M.G. Craford losses, the IQE of the AlInGaP material system is limited, especially as operating temperatures increase. A similar projection but at 85 C junction temperature and warm white taken at 3000 K CRI 90, R9 50 puts both pcleds and hybrids at a practical maximum of 225 lm/w (Soer et al. 2014). The hybrid sources are expected to reach near-maximum efficacies sooner than pcleds as the direct-emitting red LED sources are readily available, while the suitable narrow red phosphors for pcleds require basic materials development and discovery. Ultimately, with 100 % internal, electrical, and package efficiency, the luminous efficacy would be limited by the spectrum and the desired quality of color rendering. Per estimate by Phillips et al. (2007), one can place the ultimate limit for lm/w of a white LED at 400 lm/w. To move past the lm/w limit of present-day technologies, it is necessary to exclude phosphors and emit a white spectra directly, which would necessitate substantial, long-term investment into the existing compound semiconductor systems or introduction of new ones. Examples of technologies which may be capable of bringing greater efficacies are nitride-based micro- and nano-wires (Wang et al. 2014) or new materials systems based on quantum dots (Mashford et al. 2013) or perovskite crystals (Tan et al. 2014). Laser-Based Solid-State Sources Even though solid-state lighting is conventionally associated with LEDs, lasers have always been considered as possible options for sources of white light through mixing of directly emitted color light or utilizing blue light to pump phosphors. As has been mentioned (Pinnow and Gerard Van Uitert 1970), some of the initial ideas for phosphor conversion detailed the use of blue lasers and YAG phosphors. Lasers offer possibility of relatively small emitting surface and high efficiency at highinput power density, making them attractive for use in high-brightness applications. At the same time, the high price and low peak efficiency have kept semiconductor lasers out of use in general lighting. The examples of successful application of lasers include overhead and cinema projectors (Beck) and highend car headlights, as demonstrated by BMW (Hanafi and Erdl 2015).Inthecaseof overhead projectors and car headlights, the underlying technology is similar to LED-based lighting where a blue source, in this case a blue diode laser, is used to excite remote phosphor to produce blue and green/yellow light. For the cinema projection, the technology is more exotic, where red and blue light is delivered by red and blue laser diodes, while green colors are obtained by frequency doubling infrared diode light (Fig. 13). LED Packaging LED packaging has evolved driven by target applications, reliability requirements, and cost. Through the 1970s and the 1980s, LEDs were used primarily as indicators. The typical package of that time is the, very familiar, 5 mm lamp which schematic is

21 History of Solid-State Light Sources 61 b LD Assembly Phosphor Assembly a Fiber Cable LED Secondary Optics Fig. 13 Examples of products using semiconductor lasers: (a) laser-based cinema projector; (b) laser-based headlight (illustration: Compound Semiconductor) (Credit: Barco) Power handling per LED (W) mm lamp Superflux LUXEON K2 LUXEON I Power Handling per LED LUXEON Altilon LUXEON Z LUXEON Rebel LUXEON S LUXEON Flip Chip Year Fig. 14 High-power LED package evolution with examples of emitters introduced by HP/Philips Lumileds shown in Fig. 14a. Such lamps were designed for 0.1 W input of electrical power. A small die was housed and connected to two leads held together by the hard epoxy lens. The lens provides an optical function, comes in a variety of shapes, and could include color filtering. As LED efficiency improved, higher-power LED packages were designed for emerging applications. An example of this is the 0.2 W SuperFlux

22 62 O. Shchekin and M.G. Craford 5mm Indicator Package Example LUXEON High-Power Package Example LED Chip Conductive Epoxy Die Attach; Ball Wire Bond Onto Top Contact Cathode (-) Die on Ceramic Package (LUXEON Rebel) Silicone Lens Wedge Wire Bond Lens Diffuser) Anode (+) TVS L0 Assembly (LED+submount) CTE Buffer Layer Leadframe/ Kicked-up WB Leads Slug Carrier Frame Note: Direct Assembly to Slug Pick-and-Place and Inspection Feature Note: Feature is Part of Slug Carrier Frame Chip in a Frame Package Cu Slug Silicone Lens LUXEON Flip-Chip High Performance Chip-Scale Package Cathode LED Chip Bond Layer Ceramic Substrate Metal Interconnect Layer Thermal Pad (electrically isolated) Substrate PSS PSS Sapphire InGaN EPI Redistribu on Layer Base Metal AuSn Solder Bump Fig. 15 The trend in high-power LED packages is toward smaller size and greater flexibility LED introduced by Hewlett-Packard for the automotive signaling market. In 1998, Hewlett-Packard introduced a 0.5 W red power emitter designed for traffic lights. In 2001, Lumileds (formerly Hewlett-Packard) introduced the LUXEON I white highpower LED which used a 1 mm 2 flip-chip die with the package designed to dissipate the heat from the die through a dedicated heat sink. The design allowed up to 3 W electrical power input and junction temperatures up to 120 C. To increase the power handling further, the LUXEON K2 package was released where a number of highperformance technologies were implemented, allowing the junction temperature to be raised up to 150 C for white and 185 C for direct emitters and drive current up to 1.5 Amps. In 2007, Philips Lumileds introduced the LUXEON Rebel package which started the LED industry switch to die-on-ceramic (DoC) technology. DoC allowed significant simplification and cost reduction over previous high-power packages while offering all of the benefits in a more compact form factor. The technology in LUXEON K2 and DoC is leveraged for high-power array emitters used for illumination and automotive forward lighting. The improvements in efficiency, packaging materials, and the pressure to reduce component cost led to further miniaturization of the high-power LEDs. The use of aluminum nitride allowed reduction in the attach footprint as well as a choice of whether L1 optical elements were used. This flexible approach is illustrated in Fig. 15 where the chip in a frame concept offers the option to have a flat package instead of a dome for a lower cost emitter with smaller source size (LUXEON ZES). The chip-scale package (CSP) is the emerging concept in high-power LED architecture and is illustrated in Fig. 7e. Here the solder pads are on the die itself, and the optical elements can be

23 History of Solid-State Light Sources 63 Fig. 16 (a) Schematic of a typical mid-power LED, (b) illustration of the scalability of the mid-power architecture from 10 lm to 5000 lm emitters shaped into the die or molded around it. The heat management of CSP devices is achieved partly by the high efficiency of the emitter and by the board to which the CSP is attached. A CSP can be used by itself or as an emitter in a separate package with the latter adding thermal or optical functionality. Another type of LED architecture, often referred to as low or mid power, is illustrated in Fig. 16. At the heart of the LED is, usually, a blue lateral die as shown in Fig. 7b, which is attached to an overmolded lead-frame package. Phosphor mix or silicone is then dispensed into the cavity ( cup ) formed in the package. This architecture originates from display applications, where similar packages were first used for backlighting low-power mobile displays. As their efficiency improved, the LEDs could be driven harder, allowing for use in large screen displays and ultimately in lighting applications. The primary attraction of this architecture is its low cost, since the die and the package leverage the high-volume manufacturing used for display applications. Also, such emitters are quite efficient. The efficiency of the blue die has been mentioned earlier in the chapter. The high efficiency of the package comes from a relatively large volume of phosphor, reflective package materials and lower current density at which the die is driven compared to a typical high-power emitter. The architecture is highly scalable as multiple dies can be placed in packages of various sizes, as is shown in Fig. 15b. The disadvantage of the mid-power architecture is that the power density cannot be as high as that for specifically designed high-power emitters. While high lumen output has been demonstrated for the large CoB arrays, the surface brightness is lower than that of high-power emitters. Figure 17 maps the increasing diversity of the LED packages across source luminance and lumen output. This is a useful view to help link the solid-state source to an application and also see where a need can be addressed with more than

24 64 O. Shchekin and M.G. Craford 1.E+09 Laser sources High brightness emitters Imaging, Automotive Forward Lighting 1.E+08 Luminance (cd/m2) 1.E+07 1.E+06 Mid-power single emitters Illumination High-power single emitters Illumination High-power arrays Illumination Mid-power arrays Illumination Low power emitters Indicators, small display 1.E lumen output Fig. 17 Lumen output versus luminance map of the various LED sources one type of emitter. For high luminance applications, for example, automotive forward lighting, mobile camera flash, directional outdoor illumination, spot lights, and down lights, which demand >10 7 cd/m 2, high-power LEDs dominate. Here the emitting surface (source) has to be small and bright to allow for narrow and/or well-defined beams. This usually requires power densities at which the reliability of mid-power architectures is compromised. For applications such as nondirectional lamps, the mid-power and high-power emitters and arrays strongly overlap in the region of to cd/mm 2 (the dashed area in Fig. 16). Here cost and intended application determine the selection. Indeed, in applications where the luminance is not critical, high-power, mid-power, and, in some cases, low-power LEDs compete. In this case, the key metric becomes the light output at a given cost, expressed as lm/$, and now performance (lm/w and W/mm 2 ) and cost ($/mm 2 ) are inherently linked. As an example, during the last few years, the mid-power LEDs have almost entirely displaced high-power LEDs in A19 lamps, and low-power LEDs are displacing mid power in tube lights. The application landscape for the various LED architectures will continue to be dynamic as the efficacies increase and costs decrease. The low cost of the low- and mid-power emitters based on lateral die will continue to be attractive. As the luminous efficacy of the LEDs increase, these architectures will start addressing applications where the high brightness requirements of the source could previously be fulfilled only by high-power emitters. It is likely that the highend camera flash, projection, and automotive forward lighting will continue to be