Materials and reactors... It all starts, not surprisingly, at the bottom. For those coming from a higher level of the food chain, the simplest analogy the industry offers is that making semiconductors is like making a pizza. You have a crust, called a substrate, that everything is layered on. Then comes the sauce, which is a blend of just the right main ingredients, and little added "spices" that make it unique to the particulars of the kind of pizza you're making. That sauce is the "epitaxial layers" or simply "epi". In this case, you cook it while you add the secret ingredients that make up the sauce, and what you get at the end is an "epi-wafer". Some of the ingredients manufacturers blend include gallium, indium and arsenic (called "source metals"), along with other ingredients, which are basically vaporized and then showered very precisely over the sapphire or silicon-carbide substrate in big things called epitaxial reactors. The most common of the volume production techniques is to use MOCVD, or metal oxide chemical vapor deposition. Taken one word at a time, the name is actually pretty sensible.

Those machines are not cheap, running probably $1.5M to $2M+ each, nor are they simple. They use a lot of electricity and take a fair amount of time to get the layers just right. The rocket science in the machine itself is how to get exactly the right amount of everything even blended, across the whole substrate, on multiple substrates at a time, to tolerances in the range of hundredths of a millimeter. The objective is uniform coverage that minimizes the "defects" which may be holes, or cracks, or shortage or overages of elements in the material that's supposed to be there. How well you do at this step will set the stage for the overall yield, or "percentage of good devices" you get from a wafer. More is better, since you go to all the trouble, time and expense of getting the materials on there, you want every square millimeter to be useful. The reactors take time to do their job, take time to finish one run and set up for the next, and also need maintenance (as you can imagine, flowing a bunch of hot metals at high pressure take their toll on the equipment). There is also a need to purge out anything that's not part of the formula for any particular run, so changing from one color LED, or efficiency level of a solar cell, to another, takes time to clean the previous formula's leftovers out.

Improvements are happening, and while incremental, they are noticeable. A few years back, at one of our Blue conferences in Taiwan, currently the larger of the "Big 2" when it comes to our world of non-silicon epi-reactors, Aixtron, was sharing the migration path to larger wafer sizes. In the simplest context, edges are useless for putting devices on, and the larger the wafer, the lower the ratio of "useless" edge to "useful" interior. A move from 2-inch to 4-inch, and then 4- to 6-inch wafers can provide a substantial increase in the yield per square millimeter from each run if (big if) you can maintain the uniformity. Veeco has made a big push recently to clearly communicate its intention to drive the fabrication costs, from the substrate through a device ready to packaged, down by a factor of 4 by 2015. According to Jim Jenson, Veeco's VP of Marketing for their MOCVD business, these reactors, and their accessories, currently make up about 50% of the capital expense of an LED fab. Their model K465i, introduced in January, has brought in a new approach to the deposition nozzle (technically, their "uniform flow flange") that has enabled a whole bunch of things to get better all at once. Jensen claims that their customers have seen yield improvements from what has traditionally been in the mid-70% range to something more in the 90's with this update. That represents just a yield-based cost reduction of 20-25%. Yield improvements ripple through the whole LED manufacturing process, as a higher percentage of good devices means that for the same amount of work at each step (such as fabrication of the chips and testing), more LEDs get produced. Changes to the line have also shortened the time it takes to get a new reactor up to speed, with recent results being customers having being able to take delivery of one of the reactors, and fully qualify their process on it in just 2.5 months.

LEDs, the other rocket science... It wasn't that long ago that packaged "lighting quality" LEDs were running at $10 for 100 lumens, or 10-cents per lumen (remember, blue and white weren't commercially available until around 2002/2003). Announcements in the last few months have shown us 2-cents per lumen (Cree), then 1.5-cents (Bridgelux), and most recently less than 1-cent for warm white (Intematix, part of today's news). It's assured that Philips, Osram, Nichia and others out there aren't standing pat at 10-cents per, they just didn't happen to specifically promote the price in the their announcements. That's a factor of 10 decrease in something like 5 years. We'll discuss what's driving that in the next installment of this commentary.

Supporting components... Suffice it to say in Part I here that there's room for improvement in both drivers (which feed and control the LEDs) and power supplies (which feed the drivers). The capable and reliable ones aren't cheap, especially when it comes to the power supplies.

Integrated lamps and luminaires... When do we get a $5 LED lightbulb? Maybe never, but not because it can't be done, but rather because it won't make sense to. At some point, a product becomes "cheap enough" that mass market adoption proceeds simply because it is a better solution than what existed before. One of my continuing favorite examples to evaluate some of what is happening, and what we think will happen in this industry, is the progression of the PC market. Introduced in the early 1980's, they started out as $2000 tools, and $1000 toys. You had to really need one for business at $2000, and most mid-sized or larger companies were doing just fine on the "cost per terminal" with their existing minicomputers. Small businesses had nothing in the way of a computer, and couldn't afford the $50,000 to $100,000 or more for their few employees who would benefit. $2000 for the PC, plus another few thousand for what was likely custom software, was way better than paying an extra accountant $30K a year (back then) to do the math on paper. As the business-level machines came closer to $1000, the 20-50 seat installations began to make sense as well, and massive adoption proceeded. Later, $500 PCs put them in most of our homes, but did you notice, they didn't keep heading on down to $300, or less (other than rare deals, so super-strippers)? The distribution channel (retailers) couldn't make the money they needed at that kind of price, and having PCs in every consumer electronics store drove far more sales than a lower price (by mail order) every would. They hit the value point at $500 and have stayed there, with features and capabilities being added, rather than prices proceeding lower.