Tilting at Tech Market Windmills
The debate over Dataquest numbers

In John Cooley’s column last week, he raised long overdue questions on the myth of market share perpetuated by Dataquest’s annual pilgrimage down the mountain with their stone scoreboards showing who supposedly won the year-before-last championship in the FPGA synthesis market. Perhaps we should suppress our collective yawns long enough to take a look at the relevance and reliability of the metrics provided by these market-share mavens.

Every year about this time, Dataquest sends out their analysis of various technology markets. Among those is a measure of the relative market share of EDA tools in various categories such as FPGA synthesis. The report unceremoniously declares the size of the market segment and the product revenue claimed by each competitor. Then, through the miracle of Microsoft Excel, it provides the percent of market share for each. The business model for this is curious and clever. The research firm essentially collects data from each company, then sells the same data back to them a year later after summarizing, sanitizing, and cross-checking it.

What does it tell us about the FPGA synthesis tools market? Well, it can give us an idea how much each company claims to have collected selling the specified product in the given year. Beyond that, it doesn’t tell us much. It certainly is no measure of who offers the superior solution. It is nothing at all like a measure of which products are most used in the field. It’s not even a measure of who succeeded in winning the most sales or who captured the most revenue in the segment. The conclusions actually have little resemblance to any relevant reality.

What’s wrong with the numbers? Well, first and most obvious, they’re obsolete. In the fast-paced world of design tools, particularly in a dynamic market like FPGA, getting information in 2004 on what happened in 2002 is worthy of mild, passing curiosity at most. It’s like trying to figure out which football team is best this year by looking at two seasons ago. The answers change over time – rapidly.

The second problem is that the numbers don’t reflect what tools are actively chosen by the customers that compare and buy them. In the world of big-EDA, revenues are dominated by large, complex remix deals between major EDA companies and large systems companies. While these multi-million dollar deals serve their customers well by allowing flexibility in what tools are used when, and saving resources for when they’re needed, they also have almost no specificity on what portion of the revenue is attributable to each of a broad portfolio of products included in the bundle. This means that the distribution of dollars into the various product categories is more a function of political battles between product groups within the EDA company than a realistic reflection of the products the customer actually bought. Large EDA companies cannot accurately report how much of a given product they sold because they simply do not know. The best they can do is estimate, and it is to their advantage to estimate in a way that makes them look good in the market segments they care about. The only companies that can’t take advantage of this ambiguity are the focused suppliers like Synplicity. Because their revenue comes from a comparatively small number of products, they have very little flexibility in what they report.

The third problem, in the case of FPGA synthesis, is that the numbers ignore two of the largest suppliers completely. According to our FPGA Project Survey data, Xilinx and Altera are the first and fourth largest suppliers of FPGA synthesis tools, with a combined share of almost exactly half the seats. While an accurate picture of the revenue from the vendor-supplied tools is impossible to calculate (much of the cost of the tools is obviously subsidized by silicon sales) it is a bit absurd to offer market share statistics that ignore two of the largest suppliers, and almost half of the installed seats in the market.

Fourth is the issue of where the money for FPGA synthesis tools actually comes from. For the past five years, we’d speculate that the largest consumers of third-party (EDA company-supplied) synthesis tools are the FPGA vendors themselves. During the years when the FPGA suppliers’ progress on silicon far outpaced their progress in synthesis, FPGA companies shelled out millions of dollars to commercial EDA suppliers for OEM versions of their tools to pick up the slack. They licensed these tools from EDA companies to be included in the design kits the FPGA companies provide to their customers. Many of the tools that the EDA companies now complain are being “given away for free” by the FPGA suppliers have historically, in fact, been their own products that they willingly supplied for reasonable compensation. Because of this fact, the year-to-year size of the FPGA synthesis tools market is more a function of what contracts were negotiated between FPGA vendors and EDA vendors for bundled OEM tools than any other single factor.

Fifth (do we really need five?) is the complete lack of correlation between tools purchased and tools used. In the FPGA synthesis market, the company with the best ratio here is clearly Synplicity. If someone buys their tool, they generally intend to use it. Unlike a tool from an FPGA vendor or from a major EDA supplier, it didn’t likely come along for free as part of a package or get rolled into a larger corporate purchase. In many cases, design teams own licenses to these other tools without even knowing it. Adding insult to injury in this space is the fact that Synplicity’s revenue from OEM products apparently was not included in the market share numbers, while other vendors’ OEM revenue was included.

In John Cooley’s column he points out that the Dataquest numbers simply do not make sense. We tend to agree, but we are far less than surprised, given the methods and meanings involved. John also asks for responses from readers on their experience with FPGA synthesis products. Although we are yet another second-hand source, here is ours: FPGA Journal’s 2003 FPGA Project Survey data gathered responses from readers on projects completed between August 2003 and December 2003. After filtering and weighting those responses, we have data from an estimated 310 completed FPGA projects. Of those, 35% reported using Xilinx’s XST, 32% reported using Synplicity’s Synplify or Synplify Pro, 17% reported using Mentor’s LeonardoSpectrum, 14% Altera’s QIS, and 2% Synopsys’s FPGA Compiler II. Our results should have no particular correlation to Dataquest’s since we focus on what tools were used to complete a project with no accounting for the cost of the tool or how it was acquired.

So there you go, John. You can include our answers all for free. You don’t even have to buy our report. No need to trundle down to Dataquest with the big check, either. Do these numbers lack Precision? Yes. Our guess is that Mentor’s relatively new product wasn’t yet showing up in completed projects during our survey window. It certainly wasn’t there in 2002, which is the year Dataquest’s numbers represent. Also, given Synopsys’ announcement this week of their new re-entry into the FPGA synthesis market, our results may also (even though less than two months old) be obsolete, or at least irrelevant.

Kevin Morris, FPGA and Programmable Logic Journal

March 16, 2004

[back to top]

Comments on this article? Send them to comments@fpgajournal.com