In the field-programmable gate array space, or FPGA, there are two major players locked in heated combat generation after generation: Xilinx (NASDAQ:XLNX) and Altera (UNKNOWN:ALTR.DL). Generally speaking, the players in this space that can get to market with a product built on the latest generation process technology from Taiwan Semiconductor (NYSE:TSM) is usually in a pretty good position to take share.
Unfortunately for Altera and fortunately for Xilinx, the latter was first to market with products built on TSMC's 28 nanometer process. With Altera's recent bold move to build its chips at Intel (NASDAQ:INTC) Custom Foundry, Xilinx has been rushing out the press releases to try to champion its leadership in this space. So, how should investors interpret Xilinx's latest press release? What are the broader implications?
20 nanometer parts are here...sort of
Xilinx issued a press release in which it claimed that it was delivering products based on TSMC's latest and greatest 20 nanometer process to customers today. At first glance, this would seem to imply that TSMC is ahead of schedule on the volume production of 20 nanometer wafers and that Xilinx has just dealt a crushing blow to its competitor Altera (which did not issue a similar PR). However, before everybody gets excited, it's important to understand just exactly what was said in the PR.
Xilinx announced that it was shipping the first 20 nanometer samples to customers with "general sampling availability" in the first quarter of 2014. This is not even close to having products in high volume manufacturing (TSMC has explicitly stated that 20 nanometers doesn't go into high volume production until Q1 2014). Indeed, the company first began sampling its Kintex™-7 FPGA in March of 2011, but it wasn't until March of 2012 that the company actually released these products to high volume production.
Assuming that Xilinx follows roughly the same schedule, this would put mass availability of these 20 nanometer products in the late 2014/early 2015 timeframe. Now, given that TSMC expects the 20 nanometer ramp to be quicker than its 28 nanometer ramp, it wouldn't be unrealistic to expect general availability of these chips (i.e. significant revenue contributions) during the second half of 2014.
Your move, Altera
Given that Xilinx has typically been able to execute more rapidly on a given TSMC node than Altera has been, it's likely that at the 20 nanometer generation Xilinx once again ends up beating Altera to market. Things get more interesting, though, at the 14 nanometer node. See, while TSMC is still figuring out how to get its planar, 20 nanometer manufacturing process in high volume production, Intel – Altera's foundry partner at 14 nanometer – has shipped hundreds of millions of 22 nanometer FinFET chips into everything from workstations to tablets. Intel's 14 nanometer process – which is a second generation FinFET process and, by Intel and Altera's claims, a more dense process than its competitors – should go into high volume production during Q1 2014.
TSMC claims that its first FinFET process (16nm) should go into high volume manufacturing during the first half of 2015. Now, given that Altera's 14 nanometer design isn't scheduled to tape out (i.e. the design is complete) until Q4 2014, it will be interesting to see if Altera will actually end up having the lead over Xilinx which could very well tape out a 16 nanometer design at TSMC by Q4 2014 as well (and both would see volume shipments perhaps in late 2015). Many questions still loom about the potential competitiveness of the TSMC 16 nanometer process against Intel's 14 nanometer, although it should become clearer in due time.
Foolish bottom line
With Intel now taking selective foundry business and charging select fabless companies a premium for a process technology edge, the competitive dynamic of Altera versus Xilinx at the high end of the FPGA market will be turned on its head. While Intel is usually first to new processes for its own products, it's not clear if this will translate into its foundry customers having a meaningful lead. If Altera had a track record of out-executing TSMC with a level playing field, this would be a no brainer – Xilinx would be in major trouble.
That being said, given Xilinx's tendency to outperform Altera at a given node, and given that Intel's foundry customers may not be able to ship products on Intel's leading edge process technology as quickly as Intel will be able to get its own products out, the space is still very uncertain and, for many, exciting.