FPGAs are one of the seminal inventions of the Semiconductor Industry.    They have and continue to be a hugely productive and generally cost effective way of exploring designs in complex electronic systems.     Theory says, in fact, that they are increasingly a good choice for full production implementation of systems electronics as the cost of full ASIC design becomes prohibitive.   A whole body of pundits (1, 2)  predict the demise of custom ASICs as the capabilities of FPGA continues to progress.   In fact, there’s a very lucid presentation of why the numbers say that FPGAs will subsume ASIC development over time from the perspective of 2003 and I, generally, agree with this analysis.

OK, so now that we’re about 6 years removed from this analysis what’s happened?    I’ll put out some data.    The chart below shows three revenue sequences since 1994.    One is the SIA data for overall semiconductor industry revenue, the second is the same for fab-less from the GSA and the third is the revenue history on a yearly basis of Xilinx and Altera (summed).    The assumption is that Xilinx and Altera ARE the FPGA market, or at least an excellent proxy as they comprise some 85% or more of the market.    The graph might be a little hard to understand as I’ve scaled the three sequences differently to allow them to plot conveniently on the same axis:  “Xitera” is plotted in Billions, Fab-less is $10′s of Billions and overall Semi is $100′s of Billions.    Regardless, do you note anything unusual?

Two things stand out:   First is the fact that the fab-less industry has continued to outpace the overall industry in the last 15 years, not surprizing … but a trend that has slowed dramatically in the last few years.     Second, of course, is the anomoly of the year 2000.    It was a banner year for FPGAs (and many others in the industry).   But if you study the overall trend there’s something quite surprizing and in direct contrast with the predictions of the pundits in the last several years.     Take the same data for the overall semiductor industry and the “Xitera” sequence and normalize for 1999 (i.e. Revenue in 1999 = 1).   Plotting, the graph below results.  Conclusion:  the FPGA market has not grown faster than the overall semiconductor market for the last decade!    In other words, despite the theory that FPGA’s should be capturing more content because they are a more overall total cost effective solution when including development cost, there has been zero content capture relative to the overall industry …

I discussed this with a long time executive in this industry recently and he was quite aware.    Some of this is ‘self-imposed’.    Xilinx and Altera are among of the best financially performing companies in the industry and, uniquely in the fab-less space, pay a dividend to their shareholders.    Capturing new applications, no doubt, requires business model innovation and sometime aggressive price moves.    Arguably, they are making a choice for margin and profitability over growth and, certainly, the current investment mood favors that.    I think, however, the numbers in the analysis are all too real.    ASIC and ASSP development at advanced nodes is simply becoming too cost prohibative, FPGAs will become a better choice for overall economic cost.   

I remain convinced that FPGAs will re-assert their relative growth against the industry.    This may, however, look like another effect of the Moore’s Law exponential sequence.   It will ride underneath the measured data until the the order of magnitude of the effect makes it immediate.   Suddenly, a step function will occur.    As is often the case in these “steps” it provides an opportunity for some to “step-up” … and some to “trip”.