The recent announcements by
Altera and
Xilinx of their respective 40-nm FPGA families was a welcome bit of good news shining amidst the grey mists that dominate today’s bleak economic landscape. If there had been any doubt until now about whether FPGAs would ever be a cost-effective alternative to ASICs, the capabilities, power consumption and value that these new programmable devices offer has now laid that to rest. Indeed, FPGAs have hit a new stage of maturity where the same market dynamics that govern other mature high-stakes, high-tech products like silicon foundries and airliners will shape their evolution from here on.
The competition between Altera and Xilinx is reminiscent of the struggle we’re seeing play out between the two big airframe makers, Boeing and Airbus, as they roll out their massively-complex next-generation aircraft. With development costs for a next-gen aircraft running into multiple billions, each aircraft maker can only afford to roll out a single new design, a design upon the future of the company depends. With development and tooling costs spiraling upward with each new generation of FPGA that emerge, Brand A and Brand X are now in a very similar situation.
Much like the super-efficient Boeing 787 and the super-large Airbus A380, the new Xilinx and Altera FPGA platforms share many common characteristics while differentiating themselves in important ways that will determine what parts of the market they manage to capture. Both companies had to take the plunge to 40-nm processes to deliver the necessary logic density and speed they need to make the inherently-larger FPGA structures competitive with merchant silicon solutions. They’ve both also equipped their value-priced logic families with high-speed SerDes transceivers that allow them to serve as system blocks for the many applications that now rely on PCI Express, XAUI and other high-speed serial interfaces.
Despite this, there are several key differences in both technology and architecture that will determine which of these giants dominates certain market segments.
One of Altera’s obvious differences is that it has inserted the SerDes-equipped Arria II series in between its high-end Stratix family and its economy-class Cyclone devices which are currently produced in 65 nm. Only time will tell whether Altera’s Arria series’ product line will continue to expand downward and eventually displace the Cyclone line or whether they will tolerate the additional overhead of the three-family structure and eventually move the Cyclone to a 40-nm or 35-nm process. In contrast, Xilinx has moved its entire low-end Spartan line to an aggressive 45-nm process and introduced a new branch of the Spartan family which boasts significant SerDes capabilities.
While both companies now offer both general-purpose products and ones that have been designed to address specific market niches they have taken very different approaches to differentiating their chips. Xilinx has aligned its Virtex family around three so-called domain-optimized FPGA platforms that have different mixes of general-purpose logic, DSP elements, memory, and serial I/O transceivers to address specific application spaces. Altera on the other hand, pretty much retains the same internal architecture but offers three different tiers of I/O capabilities, ranging from standard-grade 3.125 Gbit/s SerDes elements to a high-performance family that includes 11.2 Gbit/s transceivers which can support virtually any type of 10 Gbit/s connection with ease.
On one hand, it’s easy to see how Xilinx’ application-tuned architectures could give their devices enough of an edge to take a larger chunk of the production designs that would have previously been done on ASICs than Altera’s more general-purpose approach. On the other hand, Altera’s ability to move virtually any FPGA design to their lower-cost metal-programmable HardCopy ASIC platform may give them the edge for designs intended to ramp up to high production volumes.
It’s tough to say anything really conclusive about who (if anyone) will emerge as the victor in this latest installment of the FPGA wars but it’s a good bet that the arrival of these new platforms will really change the way many digital designs are implemented. While I think there will always be a market for the custom digital silicon that has been the mainstay of the ASIC industry, FPGAs have matured sufficiently that, for many applications, are now a viable alternative to the traditional foundry.
Comments? Questions? Other climax-stage industries you’d like to comment on?
Write me at
lhg at en-genius dot net or post your comments on our blog.
