One of the interesting things about the current economic situation (crisis, downturn, reset – put your chosen word in here) is how different companies all say that while the situation is difficult for everyone else, it isn’t really troubling them because…
And at the Globalpress conference in March, there were a lot of good reasons being touted about why a particular speaker’s company was going to survive, thrive, be ready for the recovery. (And add your own chosen word in here too.).
One good reason came from Tanner EDA. The company supplies analog, mixed-signal, and MEMS circuit-design tools, competitively priced. It is privately held, so there are no investors leaning over the management’s shoulders to double guess the strategy. And design teams looking to reduce costs can start new projects using the Tanner tool chain, rather than extending the licences from the big boys.
FPGA players are also differentiating themselves. Altera and Xilinx compete on process nodes (my technology is smaller than yours), speed and capacity, on breadth of options and anything else they can find to squabble about. They do have divergent views on how to present the larger families, with Xilinx building a family of different platforms. But they both spent a lot of time explaining how FGPAs are going to reduce still further the number of ASIC/SoC starts. And that means, they say, that they won’t have so many issues with the current economic crisis. (Although both companies have since reported significant falls in revenues in the first quarter of 2009.)
The debate on ASIC vs FPGA is one of those debates where there are lies, damn lies and statistics. We will ignore both sets of lies, but the statistics are interesting. If you look at design starts, then ASICs are falling, slightly, and FPGAs are gaining, slightly. But if you look at the dollar revenues, the story is that they are both moving pretty much in sync, with the total market for FPGAs only a fraction of that for ASICs.
One area where the battle between ASIC and FPGA is significant is in small devices. (In fact most FPGA design starts are not at the headline big die/latest process nodes that the companies are using for their boasting, but one or even two generations behind these.) A panel on Low Cost FPGAs or Custom Silicon? featured three FPGA companies, Actel, Altera, and relative newcomer Silicon Blue, all of whom made the case for using FPGA in volume production, particularly where Actel and Silicon Blue can supply low-power devices. And they argued that many embedded consumer devices, such as mobile phones, with fast time-to-market and short product lives, were the natural home for these FPGAs. Altera hedged its offering by advocating their Hardcopy approach, re-targeting the FPGA design into a version of a structured ASIC, for larger designs and higher volumes. Global Unichip, the design arm of TSMC, argued instead that it was possible to be cost-effective for relatively low volumes of ASICs, if the project is correctly managed.
Another approach for mobile phones and similar applications was put forward by QuickLogic. QuickLogic is no longer an FPGA company, but instead sells Customer Specific Standard Products (CSSP), which are a mix of pre-designed blocks and an anti-fuse-based programmable fabric. Quite how these differ from, say, Xilinx platform devices, which contain hard cores and programmable fabric, is not entirely clear, but at least we have a new FLA to join ASSP and ASIC.
A step back from FPGAs is Tensilica, who provides configurable processors for SoC designers. The company has also moved from supplying just general purpose designs, which it still sells, to focused designs; for example, for audio and video, and again, they are relatively bullish their prospects.
But there were signs at Globalpress that all was not right with the world. Normally there is a full day of start-up or relatively new companies touting their wares: this year there were only two, Arctic Silicon and Netronome.
Netronome provides network flow processors. These add intelligence to the network, increasing throughput and prioritising traffic, using rules based on packet content.
Arctic Silicon, based appropriately in Trondheim, Norway, is an analog semiconductor company, concentrating on analog-to-digital conversion with ultra-low power – down to 22mW per channel.
Power was as significant a discussion point for many other speakers, from Mentor Graphics’s Wally Rhines doing his usual display of statistical fireworks to discussions from Fairchild and National Semiconductor, down to product claims by many of the other speakers.
So — is there going to be a long problem? Many people seem to think that while the economy is still dire, the embedded sector appears to be digging in, not so much for a siege, but instead, preparing to be competitive when the tide turns. And the suppliers of products to that market are hoping to be ready to take advantage of their preparedness.