24 April 2012
Will Intel change the fpga and foundry worlds?
Intel has been at the leading edge of semiconductor manufacturing for many years. Its 'tick tock' strategy – refining the microarchitecture one year, moving to a new process node the next – has pushed it at least one step ahead of the industry. While TSMC and GlobalFoundries wrestle with ramping their 28nm processes, Intel is manufacturing microprocessors at the 22nm node.
But one thing Intel hasn't done before is to make semiconductor devices for other companies. That changed towards the end of 2010, when fpga start up Achronix announced that it was taking advantage of Intel's technology and manufacturing expertise for its Speedster 22i family. That announcement was followed earlier in 2012 by a similar move from another fpga start up: Tabula.
While the two moves may disrupt the fpga market to some extent, Intel's arrival as a foundry has the potential to bring more general disruption to the electronics industry.
As far as the fpga market is concerned, Achronix – and Tabula, but not immediately – has the potential to break up the dominance of Altera and Xilinx. While Altera and Xilinx address a broad market, Achronix is focusing on high performance communications applications. Both Altera and Xilinx have maintained products on leading edge processes in order to provide high densities. But Achronix' chairman and founder John Lofton Holt said: "Performance is not scaling with density and static power consumption is a problem, particularly for mobile applications." His belief is that Intel's 22nm process and its trigate transistors will be a game changer. "We see a market opportunity for high end performance, but with low power and low cost," he claimed. That demand is being met – in Achronix' opinion – with the Speedster22i families. In fact, Holt said trigate technology on the 22nm process enables fpgas that are 37% faster, but which consume half as much power as a 28nm device.
Achronix is taking advantage of Intel's technology in a big way. Holt admitted the Speedster22i range is essentially a turnkey deal. "Intel is providing a complete supply chain," he said. The fpgas are set to feature a range of hard IP blocks alongside the programmable fabric. Holt has his eyes set on a share of the $11bn assp/asic market. But to play in this market, fpgas are becoming more like the products they are looking to replace.
However, Intel's arrival on the scene as a foundry has wider implications. While TSMC and GlobalFoundries struggle, apparently, with 28nm, Intel is in production at 22nm and, effectively, open for foundry business. So when companies at the leading edge contemplate their next moves – say 14nm – will they be calling Intel before they call TSMC? Intel, of course, isn't the only big company offering foundry style services; other contenders at the leading edge include Samsung and IBM.
Offering foundry services hasn't been an easy move for Intel, as Holt explained. "People think foundries are all about technology, but they aren't; they are all about services. Intel has had to learn how to turn a good fab process into a good foundry business. However, along the way, Intel has maintained its process leadership."
Meanwhile, Achronix admits that it's targeting a narrow section of the fpga market with its products. If it breaks into the fpga market in a meaningful way – remember, it hasn't shipped any 22nm parts yet – then it must surely broaden its vision to general purpose applications. The 'big two' fpga vendors would then have a choice to make. Do they stick with TSMC in the belief that its process development will keep them in touch or might they decide that Intel or Samsung could be a better bet?
There are always caveats. Many fpga companies have come and gone over the years; only four have stuck around – Altera, Xilinx, Lattice and Actel (now Microsemi). Can Achronix can join the club? And does Achronix have access to enough capacity at Intel?