Will Silicon Substrates Push LED Lighting Into the Mainstream?

LEDs have many advantages over traditional lighting including efficacy, longevity, and robustness, but price is not one of them.

The reason why LEDs are expensive is partly because the manufacturing process used to fabricate the wafers from which the individual chips are cut is difficult and employs exotic materials such as gallium nitride (GaN) deposited on sapphire or silicon-carbide (SiC) substrates.

But recently, some manufacturers have proposed using silicon, the material routinely used to fabricate billions of integrated circuits (ICs) every year, as an LED substrate. Apart from potentially reducing the cost of LEDs, the use of mature complementary metal oxide semiconductor (CMOS) IC technology would allow fabrication in conventional wafer fabs that have spare capacity.

However, the idea is not new. Silicon has been tried before, but large mismatches between its crystal structure and coefficient of thermal expansion, and those of the GaN used for the LED proper created faults that lowered the end product’s efficacy and dramatically shortened its life. Now new developments promise to overcome these drawbacks.

This article describes recent advances in manufacturing technology for silicon-substrate LEDs with reference to the work done by major makers, and compares the performance of the experimental chips with today’s commercial LEDs.

Substrate wars

“White” LEDs, typically blue LEDs working in conjunction with yttrium aluminum garnet (YAG) phosphor (see the TechZone article “Whiter, Brighter LEDs”), used for lighting applications have only been commercially produced for two decades, and as such the technology can be said to be still in its infancy.

A commercial white LED like Philips Lumileds’ Luxeon T chip, a device that produces 111 lm/W at 700 mA, exemplifies the state-of-the-art in solid-state lighting. Further gains in efficacy are likely, as Philips Lumileds and others major suppliers such as OSRAM and Cree continue to develop their products.

Efficacy has risen from 15 lm/W in 2000 to around 200 lm/W for devices currently on the lab bench. In the same timeframe, the price of a group of white LEDs required to generate 1,000 lm has dropped from $302 to $7.

Despite these impressive gains, LEDs are still too expensive for most consumers as a replacement for traditional lighting. An LED-based luminaire that can be screwed into a standard light socket is about 15 times as expensive as an incandescent bulb, and 5 times the cost of a compact fluorescent light (CFL). Nonetheless, in 2011, LEDs made up around 12.3 percent of the global lighting market and are predicted to rise to over 64 percent by 2020 (Figure 1).

But before this forecast becomes reality, manufacturers need to bring down the cost of LED lighting to near parity with existing technologies.

Today, the majority of LEDs are constructed from a combination of GaN, which features a band gap suitable for emitting photons in the visible part of the spectrum, and sapphire substrate. GaN thin films are grown by a process known as epitaxy, which builds up the LED’s active regions, the multiple layers sandwiched between the p- and n-contact, by depositing successive layers on the substrate. A simplified schematic of the structure of one type of commercial high-brightness LED is shown in Figure 2.

A key disadvantage of sapphire substrate is the relatively large mismatch between the crystal lattice of GaN and the sapphire itself. The mismatch results in microscopic flaws in the GaN thin film, known as threading dislocations, which affect both LED luminosity and lifetime.

This has led to the rise of SiC as an alternative substrate. SiC has a crystal structure that is much more closely matched to GaN than sapphire, reducing the defect density, and improving efficacy and longevity, by at least one and sometimes two orders of magnitude.² (See the TechZone article “Material and Manufacturing Improvements Enhance LED Efficiency.”)

Another problem with both sapphire and SiC is that they are expensive to produce. Worse yet, the material can only be manufactured on wafers typically measuring 4 in. in diameter (although some fabricators such as OSRAM now produce 6 in. wafers).

Some manufacturers have suggested that a cheaper production technique could be to grow LEDs on larger silicon wafers. Silicon is the mainstay of the chip fabrication industry, and foundries are well versed in producing high-yield, high-volume 8 in. wafers for IC manufacture. If silicon could be used as a substrate for the GaN epitaxy process, the price of LEDs should fall.

A 4 in. wafer has a surface area of around 12.6 in.², compared to an 8 in. wafer’s surface area of over 50 in.². Because it takes almost as long to process a 4 in. wafer as it does a 6 in., the same factory could quadruple production by switching to the larger size with little increase in cost.

Moreover, using current manufacturing techniques, sapphire or SiC wafers are diced into individual devices before testing and packaging. Silicon lends itself to a shorter dicing operation because a manufacturer could leave 16 or 32 LEDs together in a unified block, and subsequently test and package as a single device. The packaging and testing phase account for close to 35 percent of the cost of manufacturing today’s LEDs, so single packages comprising multiple dies would further drive down unit costs, while increasing throughput.³

Overcoming the disadvantages of silicon

Silicon was commonplace as a substrate for silicon chips well before the first white LED winked into life. So why was it not used as a substrate in the early days of solid-state lighting?

The answer is that silicon has some major drawbacks. Silicon’s crystal structure is an even worse mismatch with GaN than sapphire. Moreover, it has a very different coefficient of thermal expansion to GaN. These two factors lead to severe tensile stresses being incorporated into the wafers during fabrication that result in cracking on cooling. Cracked LEDs function poorly, if at all.

Moreover, silicon is a very good absorber of the very photons that should escape from the device and contribute to the luminosity of the LED. (Light extraction from experimental LEDs on silicon substrates is a quarter to a third of that from comparable devices built on sapphire.)

These disadvantages mean that sandwiching GaN and silicon together results in wafers with very-low yields and LEDs with low efficacy and short life.

In LED technology’s formative years, producing reliable, efficient devices was more important than driving down costs. Sapphire and later SiC produced good results so there was little incentive to try out such an unpromising material as silicon.

But the economic imperative discussed above, particularly the requirement to increase production by using larger wafers, something that is very challenging to achieve with sapphire, has caused a rethink.

One technique that has been successfully employed is to compressively strain the wafer during the fabrication process such that the tensile strain is cancelled out on cooling. The resulting wafers crack much less resulting in much higher yield.⁴

Another technique, developed by Bridgelux and now being commercialized by Toshiba, solves the tensile strain problem with a proprietary buffer layer that the companies claim deliver crack-free 8 in. GaN-on-silicon wafers. Bridgelux has not revealed details on how the technique works, but they say that by using the technology they have manufactured white LEDs on an 8 in. wafer with an efficacy of 135 lm/W. Toshiba says it is commercializing the method but has yet to release any GaN-on-silicon LEDs into the market.

Making stress-free wafers is only part of the problem. Another is preventing the silicon from absorbing photons. One solution, which has been pioneered by OSRAM and others, is to grow the GaN thin films on stress-relieved silicon then prepare the topside of the LEDs with electrical contacts and a mirror surface (to facilitate internal reflection).

The original silicon substrate is then removed and another silicon carrier added to the top of the assembly. The wafer is then flipped and light extraction techniques such as surface roughening and epoxy, filled dome lenses with a refractive index close to that of GaN, are employed to improve efficacy. (See the TechZone articles “What’s Next for High-Power LEDs?” and “Improving the Efficiency of LED Light Emission.”) Figure 3 illustrates OSRAM’s technique and Figure 4 shows how the performance of the company’s GaN-on-silicon LEDs compare with conventional devices.

Sapphire and SiC fight back

Many manufacturers are experimenting with GaN-on-silicon technology, but sapphire and SiC are far from obsolete as substrate materials. Promoters of silicon cite lower cost as the key driver for the technology, but there are three counter arguments.

The first is that the wafer-production cost is a relatively small proportion of the overall cost of the luminaire that the customer buys. French consultant Yole estimates that while wafer fabrication makes up 55 percent of the cost of the LED (Figure 5) the device itself makes up only 45 percent of the cost of the luminaire. In other words, wafer fabrication is responsible for only about 25 percent of the cost of the end product.

The second debating point is that sapphire wafers are becoming cheaper and bigger, negating some of the advantage of silicon. Six inch sapphire wafers are now available, and while 8 in. versions would be very difficult to produce, there is no physical limit that stops them from being manufactured in the future. The price of the 6 in. sapphire wafers is forecast to drop by two-thirds between 2010 and 2014.

Third, the manufacturing process for GaN-on-sapphire or -SiC is established, whereas while the CMOS-production part of GaN-on-silicon is mature, the processes for managing strain and thermal mismatch are new and potentially costly. Moreover, the yield for GaN-on-sapphire or -SiC is better than that for GaN-on-silicon, further eroding the cost differential. If the yields from the two processes were similar, and the GaN-on-silicon LEDs were manufactured on 8 in. wafers processed in fully-depreciated CMOS fabs, the components would likely be cheaper, but both of these factors are currently assumptions.

A third option

GaN-on-sapphire and GaN-on-SiC chips continue to get brighter, exhibit higher efficacy, and become cheaper. The latest sapphire-based LED from OSRAM, the OSLON SSL 150 white, is capable of producing 136 lm at 350 mA and a nominal forward voltage of 3.1 V (1.09 W) with an efficacy of 125 lm/W. The devices cost less than $1.80 each (for a batch of 100). Similarly, Cree’s SiC-based XLamp XM-L2 LEDs offer 153 lm/W (at 700 mA, 2.9 V, 2.03 W) at a price of $4.91 each (for a batch of 1,000). Based on long-term trends, these types of LED will only perform better and become cheaper in the near future.

Meanwhile, GaN-on-silicon technology will continue to play catch-up. It is not certain that this alternative technology will ever outperform contemporary devices. Nonetheless, GaN-on-silicon offers a practical third option for the fabrication of LEDs that could eventually be less expensive for specific families of devices. This third alternative is likely to be complementary rather than disruptive to existing technology, and that can only be a good thing for driving solid-state lighting into the mainstream.

References:

1. “Lighting the way: Perspectives on the global lighting market,” McKinsey & Company, second edition, August 2012.
2. “Reduction of threading defects in GaN grown on vicinal SiC (0001) by molecular-beam epitaxy,” M. H. Xie,a) L. X. Zheng, S. H. Cheung, Y. F. Ng, Huasheng Wu, and S. Y. Tong, Applied Physics Letters, August 2000.
3. “How Silicon Will Spur A Boom In Solid-State Lighting,” Bill Watkins, Bridgelux.
4. “Epitaxy of GaN LEDs on large substrates: Si or sapphire?,” A. Dadgar et al., Universität Magdeburg, Germany.