Group Actions
Engagement
  • Author Author: spannerspencer
  • Date Created: Date Created
  • Views 1149 views
  • Likes 2 likes
  • Comments 4 comments
Related
Recommended

2016 Year in Preview: The Future of FPGA

spannerspencer

image

FPGA

During the 2015 Community Awards, we asked you to take a cursory look into the future and give us your predictions for the new platforms and technologies that are likely to dominate in 2016.


Even though it didn't make the initial nominations, Field Programmable Gate Arrays (FPGA) evidently captured your attention, as it was a subject that came up time and again in the comments, and also in the Technology of the Year polls.


It's reasonable to consider FPGA, which already has a global market value above $5 billion that's expected to land closer to $10 billion over the next few years, as the evolution of programmable ROMs due to their reconfigurable logic blocks and complex input/output functions.


But they're also so much more.


What's Next for Field Programmable Gate Arrays?


We'd considered how to look deeper into this, and other hot technologies, as part of our 2015 Year in Review. But ultimately it's you guys who have the deeper knowledge, so instead we'd like to look forward into 2016 rather than looking back at 2015.


We want your thoughts on the future of FPGA. How will it evolve, how will it reach a wider user base, and what kind of changes will it undergo over the next 12 months. What will the FPGA look like in New Year 2016, and what kind of projects will you be making with them.


Tell us all about the future of FPGA below (and what you'd like to see, as much as what we will see), and we'll reconvene this time next year to see how close we got to the mark.

  • Certainly not much has occurred this past year of 2016. Let us see if 2017 fares any better.

  • in reply to dougw

    A "killer app" that lots of people are talking about regarding Intel's Altera acquisition is making data center computations more efficient by using FPGAs to accelerate search and other operations.  I don't know what the state of the art is on this sort of thing, but I remember people talking about "database machines" decades ago.  They didn't get anywhere -- Moore's Law made general-purpose CPUs faster and cheaper than special-purpose database engines.  With Moore's Law slowing down, they might be successful this time around.

     

    I don't see Intel competing effectively with ARM chips.  They tried that before with DEC StrongARM (renamed Xscale) and ended up selling most of the technology to Marvell.  Intel likes to sell expensive high-performance CPUs, whereas ARM chips have so many competitors that the SoCs are too cheap to be profitable for Intel.  We might see an Altera with an x86 core some day.

     

    Readers who haven't heard much about Intel's Altera acquistion might be interested in last year's element14 discussion on the topic: http://www.element14.com/community/thread/43401/l/intel-buys-altera-for-54-cash-is-this-a-good-thing

    I haven't changed my mind about comment #4 image

  • Altera is already taking advantage of Intel's crucial lead in process technology to utilize 3D tri-gate transistors in a 14 nm geometry. This is a big deal - it doubles performance and reduces power by up to 70% and of course allows a lot more devices to fit on smaller dies. Altera also has libraries for a quad-core ARM Cortex A53, which would seem to bring the latest ARM chips to 14nm. This kind of leverage has to be tough on competitors, not just FPGA manufacturers, but also ARM chip manufacturers. Intel will likely lead the way to 10 nm and 5 nm as well, which will keep the pressure on, so competitors may need to find other methods to attract customers. Hopefully it will lead to lower cost programming and simulation tools, lower cost development cards and more integrated, easier-to-use tools.

    FPGAs could well follow the example of the arduino explosion which has forced all competing manufacturers to support essentially free development environments and low cost development cards. The CPU industry understands the importance of getting new users hooked on their products and tools and I'm sure the FPGA market also understands this and some of this is already occurring. However for it to go ballistic it may take a catalyzing "killer app" at the right price to get everyone hooked on FPGAs and revolutionize the industry.

    It might be interesting to see if any e14 members have a proposal for an FPGA "killer app", but lack the resources and training to pull it off. From my perspective, (I've only "dabbled" with some Altera tools) I can think of applications, but it seems like a pretty big investment in time and money to contemplate implementing any serious applications.

  • Here are my opinions.

     

    1.  In 2015, Xilinx and Altera mostly pushed their high-end FPGA families.  This is where their profits are and expanding the low-end Spartan and Cyclone lines eats away at those profits.  I expect more of the same for 2016.  I don't know if Intel's aquisition of Altera is going to change anything -- the fundamentals of what Altera does are still the same.

     

    2.  FPGA vendors don't seem that interested in expanding their user base.  I read comments that their tools keep getting harder to use and I often read of newbies getting swamped by the complexity of vendor tools and newer FPGA architectures.  I design FPGAs professionally and I've been using Xilinx tools since version 5.2 so I got used to their complexity gradually over the years (I now use 12.4).  I'd hate to be a newbie, though.  You get all the complexity thrown at you at once.  I also have the advantage of having designed FPGA tools, so I can guess at how they work and guess at what to do when they don't.

     

    The Xilinx tools do produce excellent FPGA implementations if you know what you're doing.  But they are intended for professional FPGA designers and can be daunting for newbies.

     

    Spartan-6 is now the "sweet spot" for Xilinx.  Spartan-6 is a powerful architecture, but it's really complex -- much more so than the Spartan-3 family which is a much cleaner architecture.  For example, Spartan-6 now has a separate user manual for clocking.  This used to be just a chapter.  So when you get an error or warning from the tools regarding clocking, be prepared to break out the mountain-climbing equipment to get up that learning curve.

     

    3.  On the other hand, I have good things to say about Lattice's iCE40 family, which they added to their offerings when they bought Silicon Blue.  iCE40 is a clean and simple architecture, prefect for learning about FPGAs.  It also has one of the lowest cost development boards, the US$23.50 iCEstick which plugs into a USB port for power and programming.

     

    I can't comment on FPGA tools from Lattice itself, because I've never used them.  Instead, I use the IceStorm tools which give you an open-source tool chain for the iCE40 HX1K and HX8K.  Lattice is the first FPGA that has a complete set of open-source tools, which makes it possible to create other FPGA tools and design languages that are better suited to new FPGA users.  I'm working on such tools myself (XXICC and its Flavia tools) but unfortunately I'm swamped with family obligations so I've had to put further development on hold for a while.

     

    The iCE40 also has a very nice educational "Go Board" from Nandland, funded by a Kickstarter that just ended.  It's a nice board with built-in push-buttons, discrete LEDs, 7-segment displays, and VGA output.  Like iCEstick, it plugs into a USB port for power and programming.  It's a great target for small-to-medium FPGA projects because everything is on the same board, and it supports open-source tools.

     

    4.  I would love to see the prices of FPGA development boards come down to Raspberry Pi levels.  It's silly that you can get a reasonably powerful Linux machine for US$5, but a low-end FPGA board costs at least $20 and Spartan-6 boards cost $75 or more.  I guess most of the problem is that FPGA boards aren't manufactured is large enough quantity to get the price down, and they won't be until enough people want to buy them -- which won't happen until they're a lot cheaper.  Catch-22 number 1.  Plus, large numbers of people won't get into designing FPGAs until they get a lot easier to design, but that won't happen unless there are enough (potential) users and boards out there to make easy-to-use tools worth developing.  Catch-22 number 2.

     

    Fortunately, the combination of IceStorm on iCEstick has broken this cycle, and Nandland's Go Board has the potential to help a lot, particularly if its quantity can be increased to the point that the board could be $25 instead of $50.  There's great potential here, and I'm hopeful for 2016 -- particularly when I can get back to XXICC image