03 Aug 2007
As high-power LEDs start to deliver enough light output to challenge traditional light sources, Susan Curtis finds that device manufacturers are working on strategies to make LEDs more cost-competitive with other lighting solutions.
One of the biggest obstacles to widespread adoption of LED solutions for lighting applications remains their comparatively high cost. Using the most common cost metric in the lighting industry, the cost per lumen of light output, LEDs are around $100 (€75) per kilolumen, as compared with just a few dollars per kilolumen for more established light sources.
Of course, LED manufacturers are quick to point out that solid-state lighting should be judged on lifetime costs rather than just the initial expenditure. Indeed, LEDs have long been recognized as offering the most cost-effective solution in some applications. Take traffic lights for example, where coloured LEDs deliver energy savings of up to 90% compared with white incandescent lamps combined with filters to achieve red, amber and green output. LEDs also need replacing less frequently, cutting down on maintenance costs.
Solid-state light sources are also becoming the favoured solution in architectural installations – particularly for hard-to-reach locations, such as bridges – where their longer lifetimes dramatically reduce the costs associated with maintenance and replacement. Other outdoor lighting applications, such as streetlights, look set to follow, while LEDs could also yield lasting cost savings for public spaces where the lights are always on.
Such long-term cost advantages can be encapsulated in alternative economic metrics that take account of complete life-cycle costs, rather than the initial investment. The so-called "cost-of-light" formulation, which has been adopted by the Illumination Engineering Society of North America, incorporates the upfront installation costs, the energy costs needed to operate the light source, and maintenance costs that are incurred as the lighting system ages, such as the cost of lamp replacement.
Comparing the cost of light for different energy sources reveals that halogen and incandescent lights have relatively high values, because they consume a lot of energy, while fluorescent and LED light sources have relatively low values (see figure 1). By 2010, continuing improvements in the light output and efficacy of LED light sources is expected to bring the cost of light below $10 per million lumen hours.
However, LED manufacturers would be fooling themselves if they thought that cost-of-light metrics will sway the price-sensitive consumer market. As a result, reducing the cost per lumen remains an essential priority for LED manufacturers.
One key strategy is to continue the trend towards producing higher-power devices that deliver greater luminous flux. Just four years ago, a prototype white LED module developed by Toshiba required 1300 LEDs to produce the same luminous flux as a fluorescent tube. Now, however, white LEDs, such as Philips Lumileds' Luxeon K2 have a maximum output of 140 lm, which would enable the same luminous flux to be produced from just 22 LEDs.
Improvements in the luminous efficacy of these devices, which currently stands at 50–60 lm/W, will enable devices to produce more lumens for the same input power level. Indeed, LED manufacturers such as Philips Lumileds believe that continuing innovations at the chip and package level will yield a 3–4-fold increase in efficiency, especially for green LEDs.
At the same time, better thermal management will enable LED chips to operate reliably at significantly higher currents, some three to four times higher than today. Together with higher efficiencies this could yield LEDs that deliver up to 1000 lm per chip, which would reduce the cost per lumen by at least a factor of 10.
Focus on manufacture
Other cost reductions are likely to come from improvements in production efficiency, particularly for nitride-based blue and green LEDs. This presents a real challenge for LED manufacturers since growing these devices using metal–organic chemical vapour deposition (MOCVD) remains a relatively immature process that is also much more complex than for other material systems.
The key reason for this is that GaN-based devices must be grown on foreign substrates, either sapphire or silicon carbide, and the difference in material properties between the substrate and the epitaxial structure makes it difficult to fabricate high-quality layers. What's more, new applications for LEDs are demanding even greater control over their light emission characteristics, which places more stringent limits over the quality and uniformity of the deposited layers.
The upshot is that GaN-based chip production generally suffers from low throughput and low yields. LED manufacturers are working closely with equipment makers to increase the number of devices that can be produced in a single process run, and to increase the percentage of those devices that can be sold to end users.
One solution is to increase the substrate size from the 2-inch wafers that are commonly used to produce GaN-based LEDs. Cree, which manufactures its LEDs on silicon carbide substrates, has already switched most of its LED output to a 3-inch line, while Showa Denko plans to open a 4-inch facility for GaN-on-sapphire production.
However, further increases in substrate size will not be easy, since the disparity in material properties between the GaN layers and the substrate generates strain in the epitaxial structure during growth. This strain build-up can cause the wafer to bend, which leads to variations in the composition of the epilayer. Pronounced warping of the wafer – which can occur as the wafer cools after epitaxial growth – can even cause the deposited films to crack.
One solution is to use an in situ reflectance sensor to monitor the curvature of the wafer during a process run. This can significantly reduce the time and expense needed to adapt and optimize the growth process before production begins, and can improve product yields during full-scale production. Equipment maker Veeco already produces reflectance sensors for use with its own reactors, while the EpiCurve product from German company LayTec can be fitted to the majority of Aixtron's single-wafer and planetary MOCVD machines.
Another way to increase throughput is to increase the wafer capacity of the MOCVD reactor, and both Veeco and Aixtron have been working to introduce machines that process more than 40 2-inch wafers at the same time. A particular challenge for these MOCVD machines, as well as for smaller reactors, is to ensure uniform deposition across each wafer and between different wafers, since any variation in composition or thickness has a direct impact on the emission characteristics of the LED.
One factor that affects the uniformity of the deposited layers is the flow of ammonia through the reactor. Ammonia is the most common source of nitrogen for GaN growth, but its relatively high viscosity – combined with the high flow rates needed to ensure adequate nitrogen supply – makes it difficult to maintain the laminar flow that is crucial for uniform deposition.
To address this issue, Aixtron's "yield+" system – available as standard on its newer machines, and as a hardware upgrade to earlier planetary reactors – incorporates a new gas injector nozzle that features two separate inlets for the group V gases, in addition to the single inlet for group III material. This, Aixtron claims, produces a more laminar flow, delivers greater control over the gas flows within the reactor and halves the amount of ammonia consumed, which reduces the overall bill of materials.
The central region of the reactor is also kept cooler to prevent deposited material from building up on the injector nozzle. This helps to improve the run-to-run reproducibility by minimizing changes to the reactor's thermal profile. As a result, the peak photoluminescence wavelengths measured for three different epiwafers show average standard deviations of 1.20, 0.81 and 0.95 nm over three consecutive runs, as compared with 2.70, 2.91 and 2.70 nm using a standard planetary reactor.
Stringent process control is also essential to maintain product yields. Understanding what happens inside the process chamber during material and device development helps to optimize performance within the smallest number of process runs, and in high-volume production facilities helps to detect and correct any problems as they arise. The complexity of MOCVD growth of GaN-based LEDs makes proper process control even more important, since there is greater scope for variations in the growth conditions that can affect the optical performance of the LED.
Ideally, measurements should be taken inside the reactor during epitaxial growth, but some of the most useful techniques are not compatible with the harsh conditions inside a MOCVD reactor. In practice, in situ measurements of optical properties and temperature must be combined with ex situ imaging and analysis of wafer samples. Greater automation and more rapid inspection techniques are being developed to improve throughput and yields.
LED manufacturers believe that the production costs of nitride-based LEDs could be reduced by a factor of three to six as yields are increased, manufacturing techniques become automated and larger substrates are introduced. But these companies must continue to innovate to achieve the sort of cost reductions required for high-power LEDs to mount a serious challenge in the general illumination market.
The issues that are raised in this article are discussed in more detail in the latest LED Quarterly Insights report from Technology Tracking, entitled "Cost and manufacture of LEDs". For more information see www.technology-tracking.com.
• This article originally appeared in the July/August 2007 issue of Optics & Laser Europe magazine.