GaN-on-silicon LEDs and light extraction technology enable cost-effective lighting

Traditionally, gallium nitride ( GaN ) LED components are usually based on sapphire or silicon carbide (SiC) substrates, which are lattice-matched to GaN, and are commonly available in 2" or 4" sizes. The industry has been working to develop GaN using more abundant silicon wafers (6" or larger), because silicon substrates can significantly reduce costs and can be manufactured on automated IC production lines. It is reasonably estimated that such substrates can save 80% of costs compared to traditional technologies.

　　However, the problem with silicon substrates is that they are severely mismatched mechanically and thermally with GaN, which can cause severe warping of the wafers that make up the LED components and deterioration of the quality of the crystal material. Now, the silicon-based GaN technology of Cambridge University spin-off CamGan (acquired by Plessey in 2012) has solved this mismatch problem and has been successfully applied to its wafer processing plant in Plymouth, UK. As a result, the industry's first low-cost, entry-level commercial silicon-based GaN LED is now on the market. The primary products are mainly aimed at the indicator and accent lighting markets, with a luminous efficacy of 30-40lm/W. In the third and fourth quarters of this year, 70lm/W products will be launched to supply more general lighting markets.

Figure 1: Vertical LED production flow chart.

　　GaNonSiGrowth: GaN on Si Growth

　　Mirrorlayeradded: Added mirror layer

　　Wafer: Use wafer

　　Flipbondedwafer: flip-chip bonded wafer

　　Substrateremoval: Substrate removal

　　Metallisation and surface texturing: spraying metal layers and surface textures

The production of LEDs　　on silicon substrates requires several process steps to overcome the inherent light absorption of silicon in the architecture and to produce highly efficient components. In the wafer processing process (as shown in Figure 1), a vertical LED component is designed on a GaN architecture (based on a 6" silicon wafer, grown by MOCVD ). Next, a highly reflective contact (typically 95% reflectivity) is deposited and attached, and then some metal layers are made to attach the wafer to a replacement substrate.

　　Next comes the wire bonding, where an electrically and thermally conductive fusible gold-tin layer (remelting point temperature is about 280°C) is used together with other metal layers to serve as a carrier between the solder metal and the component or substitute when casting the solder layer. After the wire bonding is completed, the parent wafer is removed to expose the seed layer for epitaxial growth of the GaN layer. The wafer is flipped for the next step of patterning the LED components. The metal coating is patterned on the wafer and placed on the barrier layer to minimize the coverage of the light-emitting area. Most of the current is carried by the top metal (usually 2m). Finally, light extraction patterning is performed, etching into the GaN layer (exposed behind the wire bonding) and removing the parent wafer. The last step is particularly critical for remote phosphor applications because it enables control of the emission pattern of blue LEDs.

　　Since the reflectivity of GaN semiconductors is very high (about 2.45 for 445nm blue light), very little light escapes into free space. According to Snell's law, its narrow light escape cone is about 25°. If we assume that the light emitted inside the semiconductor has a uniform spatial distribution and the reflectivity of the mirror is greater than 90%, then only 8% of the total light can escape from the top surface of the semiconductor, and the rest is confined to the interior by total internal reflection and is eventually absorbed by the component materials.

　　To improve light extraction, a simple design was used that involves coupling the semiconductor to a large dome lens (whose radius is 1.5 times larger than the size of the semiconductor's light-emitting area). Ideally, the dome lens should be made of a material with a reflectivity index (n ~ 2.45) similar to that of GaN, which allows more than 90% of the light to escape into free space.

　　In reality, however, there is no cost-effective material that can be made into a dome lens with a reflective index that matches GaN, so LED manufacturers typically turn to readily available epoxy or silicone materials with a reflective index of around 1.5. However, adding a dome lens with a reflective index of 1.5 only brings the light extraction rate to 12%. To overcome the poor light extraction performance caused by total internal reflection, it is necessary to optimize the optical path of the light to increase the probability that it will appear within the escape cone.

　　Most traditional cost-effective methods for efficient light extraction are based on surface roughness. This surface technology is critical because it defines the angle at which the final light is emitted from the LED component. This is very suitable for remote phosphor applications, especially for blue LED emission pattern control. Other forms of patterned reflectors are used to scatter light, which can further improve light extraction. In essentially similar micro-architectures, such as fireflies, the jagged shape of the internal structure of the firefly can enhance the emission intensity.

　　Most LEDs emit light in a spatial pattern where the intensity varies as a function of the cosine of the angle of incidence, showing a standard Lambertian distribution. When these standard LEDs are used in an array to form a lighting patch panel, the light propagation results in some abnormal patterns of light spots that are not within the desired range (as shown in Figure 2), which we call "hot spots." In Figure 2, the brightness of the LEDs has been kept low to help illustrate this problem.

Figure 2: Example of an existing lighting patch panel.

　　To extract the light and create a more uniform spatial pattern, presenting a more aesthetic effect to the consumer, the light emitted by the LED should not be distributed in a Lambertian distribution, but in a batwing shape. This allows the light to reach a wider side area, thereby maximizing the pumping efficiency of the phosphor and reducing losses through improved blue light conversion. Once this goal is achieved, the spacing between LEDs equipped with this lighting distribution board can be set larger, ultimately reducing the overall production cost of the required lighting system.

　　It is estimated that this light path extraction project can save 10% of energy and also save assembly costs if the number of LEDs required is reduced. Whether the number of LEDs is reduced will depend on the level of light intensity pattern produced. If the number of LEDs is halved, the brightness variation in the phosphor can be maintained or improved while the light is distributed in a good batwing shape. The light path design can be completed by detailed surface patterning and embossing, and computer simulation technology is used in the design process to achieve the purpose of optimizing spatial layout and light extraction.

　　Combining low-cost GaN- on-Si technology with light extraction techniques in LED design makes it possible to design cost-effective, anti-glare luminaires using optimal low-power LED array configurations.

　　“Smart lighting systems” will continue to apply LED technology to systems with sensors and user interfaces, not just for energy saving purposes. Examples include ambient light monitoring for more efficient energy use, better user detection, and even maximizing the potential of optical communications, using the high conversion capabilities of LEDs to transmit data.