Incandescent lighting is on the way out. Around the world, governments are mandating use of lighting that works with efficiencies unattainable with ordinary tungsten-filament bulbs. The long-term energy-efficient choice looks to be solid-state lighting based on light-emitting diodes able to give off as much light as their incandescent counterparts.
The high cost of illumination-grade LEDs, though, is problematic. Several studies have shown that superefficient LEDs often don’t pay for themselves despite lifetimes exceeding 10,000 hr, compared with the 1,000 hr typical of incandescent bulbs.
So makers of solid-state illumination are on a quest to bring down the cost of these LEDs. But they’ve got their work cut out for them. The process of making high-brightness LEDs is not as mature as that of making high-volume integrated circuits. For example, there is no commonality among LED makers regarding many of the basic test methods used to measure wafer qualities. And the automated wafer handling that is taken for granted in high-volume IC fabs is generally absent in most LED fabrication facilities.
Not your father’s IC process
One reason it costs more to fabricate LEDs than ICs is that LED chips are made on smaller wafers and, thus, entail higher handling costs. Most high-volume ICs are cut from silicon wafers that are 300 mm in diameter. In contrast, high-brightness LEDs are on sapphire wafers, not silicon, and the wafers themselves have 6-in. diameters at most, with many production processes still using 2-in.-diameter wafers. (At least one supplier has demonstrated a 12-in. sapphire substrate, but not as a production item.)
LEDs also use a fabrication process unlike that of ordinary ICs. The layers of LEDs are formed through use of metalorganic chemical-vapor deposition (MOCVD). It is employed for producing thin films of compound semiconductors, as typically used in LEDs, through chemical reactions on the surface of the substrate. The reactions take place at temperatures above about 800°F at pressures up to about 100 kPa in the absence of oxygen.
The typical procedure for growing films with MOCVD starts with positioning the wafer on a platform called a susceptor, inside a reactor chamber. The surface of the wafer gets bathed in a vapor formed by bubbling a carrier gas (often nitrogen or hydrogen) through a metalorganic liquid. The gas picks up metalorganic molecules and exposes the wafer surface to them through special applicators. A chemical reaction on the surface of the wafer forms a layer of diode material such as InP or GaN and reaction by-products which are exhausted from the reaction chamber.
According to the market research firm Yole Développement, Lyon, France, more than 65% of the MOCVD reactors shipped in the latter half of last year were set up for 4-in.-diameter wafers or larger. The firm expects that for the first time in 2011, 2-in. wafers will represent less than half of all LED substrates. The firm also thinks this could tighten the supply of sapphire wafers because larger wafers are thicker than their 2-in. counterparts and tend to produce lower yields. (Volumetrically, a 6-in. sapphire wafer is the equivalent of seventeen 2-in. wafers.)
Last year, the semiconductor industry organization SEMI recorded a big jump in spending on such equipment for LED fabs. SEMI expects that the trend will continue this year with a 40% year-over-year increase in sales of LED fab gear. A lot of these systems are going to China, thanks to subsidy programs there. SEMI recorded 19 new LED fabs starting up in China last year, with another 27 new fabs expected in 2011. Though high-brightness LEDs get most of the headlines, SEMI expects strong demand from LCD backlighting to drive capacity growth for the time being.
There are two main suppliers of MOCVD systems, Aixtron SE in Germany and Veeco Instruments Inc. in Plainview, N. Y. Both are devising enhancements for their equipment aimed at boosting production throughput. An example is a system called the TurboDisc MaxBright GaN MOCVD Multi-Reactor that is made by Veeco. Specifically designed for high-brightness LEDs, it positions up to four reaction chambers around a robot arm that moves wafer carriers between reaction chambers and docking stations for wafer caddies. It works with wafers having up to 8-in. diameters.
Similarly, Aixtron recently devised a MOCVD system called the AIX G5 HT, which simultaneously deposits GaN and related alloys on eight 6-in. wafers. Aixtron says it can be configured to handle 8-in. wafers when the time comes. The equipment includes what Aixtron calls a novel gas injector that introduces perfectly laminar gas flows into the reactor so metalorganic molecules react at about the same rate throughout the chamber. Also, the disks that hold the wafers spin individually on a rotating planet disk to keep temperature deviations to less than 1°C. The whole apparatus is heated by an RF coil. A pyrometric device measures the temperature at the top of each satellite platform and keeps the wafer temperature at desired settings.
One difficulty facing LED makers trying to automate is that there are few standards for carrier and equipment interfaces on LED fab equipment. That makes it tough to implement any kind of material-handling automation because each fab is liable to handle wafers with slightly different hardware. Partly to overcome such obstacles, a group called the SEMI North American HB LED Standards Committee formed to help set form factors for widely used manufacturing equipment. Observers say the outcome could be greater use of robotic wafer handling, automated glove boxes, interbay automation, and Standard Mechanical Interfaces (SMIFs). Because LED wafers aren’t as big as the 300-mm wafers used for ICs, the material handling equipment involved needn’t be as heavy.
The process of ringing costs out of LED fabrication continues to be a research topic. One indication of the trend was the recent awarding by the California Energy Commission of $500,000 to Applied Materials for a research project to develop more-economical LED manufacturing equipment and processes, in particular the MOCVD step. The project’s total cost is just under $8.7 million of which Applied Materials will provide $4.2 million. The company also received an award of just under $4 million from the American Recovery and Reinvestment Act from the U. S. Dept. of Energy.
Getting the light out
It has been a challenge to get high levels of light from LEDs, partly because the materials used to form the diodes have high refractive indices. This causes much of the light the LED generates to reflect back into the diode at the interface between the material and air. These reflections heat up the diode material as well as reduce light output. And even with today’s high-brightness LEDs, light-extraction efficiencies are generally below 20%. So the topic of light extraction has gotten a lot of attention both by LED developers and makers of LED production equipment.
Most of the light from an ordinary LED emerges within a few degrees of perpendicular to the surface of the chip. Internal reflections can emerge from the LED chip from other crystalline faces if the angle of incidence of the reflections can be kept sufficiently low and if the crystal faces can be oriented so as not to act as mirrors to the light rays. Therefore, a lot of the research into boosting LED light output targets the shaping of the facet angles in a manner analogous to the facets of a Fresnel lens.
One trick used to minimize reflections is to just add a slight texture to the chip surface. The features added are on the order of nanometers and can be random or done in regular and periodic arrays.
LED makers are also exploring the idea of making chips so that the edge facets are not parallel or orthogonal to each other and, thus, are less likely to act as mirrors. One effort in this area at the University of Hong Kong has investigated LED chips cut with pentagonal, hexagonal, and several other polygonal shapes. The shaping takes place with lasers similar to those used for dicing chips from wafers, but with custom stages that move the wafer as the laser beam cuts the LED chips.
Researchers used this setup to make LEDs shaped like a truncated pyramid, an inverted cone, and other similar shapes. They say the resulting structure brought about 89% better light output than that emitted by ordinary square-shaped LED dies with vertical facets.
Another approach for boosting the amount of light transmitted from the LED is the use of a potting compound. The material, usually a plastic, does double duty as both a protective physical support for the LED chip and as a refractive interface between the high-index semiconductor and the low-index open air. The latter function lets light emerge at a higher angle of incidence than would be possible from a bare chip.
Materials used for these potting media is another topic of research, because there is a possibility that the potting material coating the LED can have a nonuniform thickness. Different thickness can make the LED color change, depending on the viewing angle. This effect is particularly deleterious for white LEDs, because humans can perceive even slight changes in the color.