Who's Winning in FPGAs II
Software and Stuff

Last week, our heroes tried to make sense of the complex, multi-dimensional FPGA race, after slicing and dicing the market in almost as many ways as the vendors themselves do.  (Nope, we have to take that back.  We forgot to account for things like the prestigious title of "Fastest 90nm FPGA with less than 400 pins and more than 8Mb of block RAM manufactured on a Tuesday.")  After all that score-keeping, we determined that there is not easy formula for picking the best FPGA or the best FPGA company.  The market is very diverse, and each product offering has differentiating advantages for a particular class of applications.  That means you'll have to take our advice and think carefully about your design requirements instead of just blindly picking the same FPGA as the cool kids.

Picking the FPGA is just one step in the process, however.  There are lots of other choices for the well-informed FPGA user to make and lots more companies vying for supremacy in their declared niches of the market.  Behind just about every FPGA design, we find things like design tools, IP, development boards, consulting services, software, and a host of other supporting products trying to best each other to wangle their ways into our design process.

The FPGA market is special, though.  In ASIC design, it was comparatively straightforward to make a business based on offering value in some part of the design flow.  Design and verification tools, IP, services -- all offered opportunities to reap rewards with a well-differentiated product that offered compelling benefits.  In FPGA, however, the two largest vendors adopted a "provide everything" policy early on.  For any third-party company to make a profit, they had to contend with the threat of having their product marginalized by a similar offering from the FPGA companies themselves.  It's "kinda' a bummer" to jump through the hoops required to build a business on FPGA IP or a snazzy new FPGA design tool only to find that a similar capability is "included" in the vendors' design suite a year or two later.

First up, and probably most important, are design tools.  In the old days, FPGA tools were pretty simplistic knock-offs of basic ASIC tools, but today the FPGA tool chain has largely specialized and diverged its way away from the ASIC heritage.  For years, Synplicity was the unchallenged market leader in third-party FPGA synthesis.  Their Synplify line of synthesis products perennially held over 50% market share, according to Dataquest.  Of course, that was back when Dataquest tracked EDA tools and back when Synplicity was a company.  All that has changed now.  Dataquest dropped EDA like well, a cold potato, and Synplicity got swallowed up by ASIC-preoccupied Synopsys.  So far, the part of Synopsys that was once Synplicity is holding fast to its FPGA roots and continuing to serve the FPGA market with top-notch synthesis, physical synthesis, and DSP synthesis capability.  Whether the Synopsys culture can sustain interest in that challenging market remains to be seen.  Keeping market share in FPGA synthesis has always meant running fast enough that your tools are always better than the FPGA vendors' "almost free" tools, which are advancing at a rapid pace.  When you're trying to make a living on just the software tools and a large FPGA company is offering very good ones at virtually no cost - subsidized by silicon sales, you have to be good enough to always demonstrate compelling advantages in your product.  

For FPGA synthesis, those advantages were things like "Quality of Results" (mainly the max frequency and area of the resulting design), "runtime" (how long the tool required to get those results), "language coverage" (how much of the VHDL or Verilog language was synthesizable by the tool), "analysis capability" (just how easy is it to trace down the HDL boo boo that's causing your timing violations), and "ease of use".  Here, in a single design tool, we have no fewer than five categories in which a vendor could claim supremacy.  Add to this list integration with layout in order to provide "physical synthesis" capability and more deterministic timing closure. The two major third-party contenders still alive in this market are Synopsys (Synplicity) and Mentor Graphics.  Mentor's "Precision" product family has always fought hard for bragging rights against Synplicity's "Synplify" offerings, and the close competition has kept both companies' products neck and neck in most of the categories above and sufficiently distinguishable from the FPGA vendors' tools that they have managed to maintain defensible businesses selling them for a number of years.

Mentor Graphics has always ridden the "broad-line supplier" horse in the FPGA race, pairing their FPGA-specific offerings like Precision Synthesis with more general-purpose tools such as their market-dominant ModelSim family of simulation products.  ModelSim has had the driver's seat in the FPGA simulation market since engineers first began simulating FPGAs.  Their only significant competitor has been Aldec, an independent EDA company that has jousted head-on with ModelSim for about a decade and often won.  As in the synthesis market, the close competition has kept both primary contenders on their toes, rapidly evolving and improving their products.  Unlike with FPGA synthesis, however, the availability of high-quality third-party tools took away most of the motivation of the vendors to build their own offerings.  By and large, the FPGA companies just OEM a version of one of the commercial products to include in their kits.  For simulation, the battle is all about user interface, simulation speed, debugging capability, and integration with other parts of a complete verification environment.  The simulation tools in the market now are sufficiently robust that choosing one is more a matter or personal (or corporate) preference than a subject requiring lengthy evaluation and study.

The broadest of broad-line suppliers, however, is probably Altium.  Altium has re-defined the design tool suite, producing a multi-talented engineering environment that can do everything from FPGA design to board layout to embedded software development and debug.  Many of the traditional FPGA design processes are hidden behind the curtain in Altium's system, making the FPGA on their development board somewhat less intimidating - a friendly blob of squishy hardware that can do your bidding without requiring a board re-spin.  Instead of competing with the other EDA companies on a point-tool basis, Altium banks on integration to win over the masses.  Their integrated solution competes very favorably on cost as well, compared with assembling a collection of point tools from a more traditional EDA vendor.

There is a bevy of companies competing in more focused FPGA-related niches as well.  IP for FPGA design is addressed by broad suppliers like Synopsys - whose wide IP collection for ASIC design has been almost forced into the FPGA space by ASIC teams demanding to prototype their ASIC designs in FPGAs.  On the processor IP front, ARM is showing a strong interest in the FPGA game with a purpose-built FPGA version of their synthesizable Cortex processor architecture.  There are also higher level players like Impulse Accelerated Technologies whose tools design hardware accelerators for software algorithms, or like Agility who specialize in tools that allow higher levels of design abstraction such as facilitating the transition of designs from DSP tools like Matlab and Simulink to FPGA hardware. The tools-for-DSP-on-FPGA in particular has been a hotbed of activity from the days before Xilinx acquired startup AccelChip to bolster their internal offering.  With companies like the MathWorks dipping their toes in the electronic design automation market, DSP design was seen as a high opportunity application area.  In this space, success is measured by one’s ability to convince software-centric DSP and algorithm designers that FPGA design is easy enough to attempt without a PhD in HDL.  While none of these tools has yet made the jump from startup mode to mainstream industry, there are a number that are getting traction and demonstrating worthwhile capability.

Back when Dataquest did cover EDA, they managed, year after year, to rank the FPGA tool suppliers without considering the two largest suppliers of EDA tools for FPGAs - Xilinx and Altera.  This is a bit like covering the Kentucky Derby but only comparing the performance of the horses that finished out of the money.  Each of the two largest FPGA companies has probably spent more money developing design tools than the rest of the competitors combined.  The result is two very strong and diverse suites of tools that are fully capable of taking your FPGA design from soup to nuts.  Admittedly, the meat course may be overcooked now and then, the salad sometimes looks wilted, and there really should be a better dessert than brownies with ice cream every time, but, with the FPGA vendors' tools, you can skip the fine-dining tab and still not starve.  In recent years, the vendors have upped their games, keeping their synthesis technology chasing close on the heels of the third-party products, expanding into areas like DSP design and ESL, throwing in rich IP libraries, and doing a decent job of top-to-bottom integration.  The downside of these proprietary tools is that they do everything possible to lock you into the supplier's silicon.  All that sweet, free, pre-tested IP looks wonderful as long as you're willing to accept the caveat that your design will never be ported anywhere else without significant re-work.  Next year, if the other company comes out with the perfect FPGA and you want to move your design over, you're basically screwed.

Well, once again, we didn't pick a clear winner.  The field is much too complex, and, of course, that's not how we roll.  We are, however, excited by the growing ecosystem of FPGA-related products that continue to emerge.

November 18, 2008

[back to top]