Vladimir Vassilevsky wrote: .....snip......>I > am wondering if anyone actually used FPGA to make a sellable product. >......snip......> Vladimir Vassilevsky > DSP and Mixed Signal Design Consultant > http://www.abvolt.comYes. Most of the products at www.rgb.com are FPGA-based. Most are video scan converters of one type or another depending on your preferred terminology (To me a scan converter is one that converts one video signal into another. To others, it has a much more specific meaning.) The common architecture uses a control processor and an FPGA for video stream processing. Using an FPGA is a lot cheaper than designing an ASIC for modest quantities. Using an FPGA is lower risk because the devices are programmable. With an FPGA one is able to fix bugs and make improvements / add features with a firmware release that's customer-downloadable. Sometimes this capability helps shorten time to market by limiting initial release features to the essential few. So, there is definitely a niche for FPGAs. Ummm....... why else would Xilinx and Altera et al still be in existence? Xilinx has sales around $1.8B .. yes, that's a big "B". Altera is similar in sales. Somebody's buying FPGAs! Fred
Online · Live + on demand
February 9-11, 2027
Price goes up Jan 1: register now to lock in today's rate
Register Now →
Anyone actually using FPGA?
Started by ●October 21, 2008
Reply by ●October 21, 20082008-10-21
Reply by ●October 22, 20082008-10-22
>Vladimir Vassilevsky wrote: >.....snip...... >>I >> am wondering if anyone actually used FPGA to make a sellable product. >> >......snip...... >> Vladimir Vassilevsky >> DSP and Mixed Signal Design Consultant >> http://www.abvolt.com > >Yes. Most of the products at www.rgb.com are FPGA-based. Most are video>scan converters of one type or another depending on your preferred >terminology (To me a scan converter is one that converts one video signal>into another. To others, it has a much more specific meaning.) Thecommon>architecture uses a control processor and an FPGA for video stream >processing. > >Using an FPGA is a lot cheaper than designing an ASIC for modestquantities.>Using an FPGA is lower risk because the devices are programmable. >With an FPGA one is able to fix bugs and make improvements / add features>with a firmware release that's customer-downloadable. Sometimes this >capability helps shorten time to market by limiting initial releasefeatures>to the essential few. > >So, there is definitely a niche for FPGAs. Ummm....... why else would >Xilinx and Altera et al still be in existence? Xilinx has sales around >$1.8B .. yes, that's a big "B". Altera is similar in sales. Somebody's>buying FPGAs!I had a quick look for a breakdown of how much of that $1.8B is from the seriously powerful devices DSP people are interested in, and how much is in the small scale devices used in many things as little more than glue logic. Is such a breakdown published somwhere? Steve
Reply by ●October 22, 20082008-10-22
Vladimir Vassilevsky <antispam_bogus@hotmail.com> writes:> The formidable FPGAs (>10k cells) come with the very formidable price > (starting at the order of $100 and growing exponentially with the > number of cells). Building any of the hyped applications of FPGA will > probably cost hundreds if not thousands. >You can do an awful lot with a less "formidable" FPGA, which are in the $10 region even from digikey.> With so much fuss about doing the signal processing in FPGAs, FPGA > development tools, FPGA student projects, zillions of articles, etc, I > am wondering if anyone actually used FPGA to make a sellable product. >We do. An automotive image-processing system, which provides feedback to the driver through the steering wheel when they are too close to the edge of their lane (as defined by the lane markings). We don't sell zillions just yet, but a reasonable number :) The processing we do was too complex for a single DSP, but fits quite handily in the FPGA we've got. You can read more about it here: http://www.trw.com/productsandtechnologies/main/1,,9_28157_28159_28162%5E4%5E28162%5E28162,FF.html and here: http://www.conekt.net/fpgas-for-ldw.html As others have noted, you pay in piece-price monetary and power consumption, and in flat-out performance against an ASIC. But the NRE is zero (over and above the time designing, validating etc, which you would have to do for both FPGA and ASIC). They are flexible, so you can do (potentially) in-the-field updates, like with micro software. As with all these things, there's a decision to be made about architecture, which has all the usual technical, commercial etc. trade-offs. FPGAs come out on top in some cases. Cheers, Martin -- martin.j.thompson@trw.com TRW Conekt - Consultancy in Engineering, Knowledge and Technology http://www.conekt.net/electronics.html
Reply by ●October 22, 20082008-10-22
Fred Marshall wrote: (snip)> Using an FPGA is a lot cheaper than designing an ASIC for modest quantities.As I understand it, even in not so modest quantities. It seems that mask prices have gone up with the newer fab processes such that mask costs are over $1 million.> Using an FPGA is lower risk because the devices are programmable. > With an FPGA one is able to fix bugs and make improvements / add features > with a firmware release that's customer-downloadable. Sometimes this > capability helps shorten time to market by limiting initial release features > to the essential few.For ASIC a bug might require a new mask set and another $1 million. You can buy a lot of FPGAs for that money, and that doesn't include the rest of the costs of actually making the ASICs. -- glen
Reply by ●October 22, 20082008-10-22
Vladimir Vassilevsky <antispam_bogus@hotmail.com> wrote in news:uvnLk.4829$yr3.3451@nlpi068.nbdc.sbc.com:> > The formidable FPGAs (>10k cells) come with the very formidable price > (starting at the order of $100 and growing exponentially with thenumber> of cells). Building any of the hyped applications of FPGA will probably > cost hundreds if not thousands. > > With so much fuss about doing the signal processing in FPGAs, FPGA > development tools, FPGA student projects, zillions of articles, etc, I > am wondering if anyone actually used FPGA to make a sellable product. > > Of course, there could be the applications like radars or cellular base > stations or prototyping CPUs or ASICs. However this is a very special > and narrow nitche. > > Then why all that commotion about FPGAs? Just because it is the cutting > edge of the technology?I use large (often >$1k) FPGAs in my projects. Some of my designs from my current job: http://www.safenet-inc.com/products/encryptors/sonet.asp (BTW, I don't work for Safenet; they just license our designs.) FPGA-based products from my previous job, using FPGAs in the $50-$1k range: http://advanced.comms.agilent.com/n2x/products/hw/xs/N2X_40G_Card.htm http://advanced.comms.agilent.com/n2x/products/hw/nextgen/Next_Gen_XFP.ht m http://advanced.comms.agilent.com/n2x/products/hw/xp/E5215A.htm (I can't claim to have designed those exact products, since they must have made several upgrades since I left. In any case, they were all team efforts with about 5 FPGA coders on each one.) The 16-port 100Mb/s tester was my baby! It used two Virtex 2 2000 parts and was designed in 2001. That 40Gb/s tester dates from around 2002-2003. I understand it is still the only tester at that rate in the market. It used Virtex 2 6000 parts, which were hellishly expensive at the time. FPGA-based products from my job before that, using FPGAs in the sub-$50 range: http://www.fujitsu.com/global/services/telecom/fsx-2000.html These were mostly converted to hardwire parts to reduce costs for later manufacturing runs. Mid to Late 1990s, Xilinx 4k and 5200 families. All of these are examples of products that would not be possible without FPGAs. I don't buy into the FPGA SOC hype though. If I wanted a microprocessor on a board, I'd be soldering on a real microprocessor, not trying to use some soft or hard core in an FPGA. BTW, I have yet to use a digital signal processor in a design. Regards, Allan
Reply by ●October 22, 20082008-10-22
On 21 Okt., 19:25, Vladimir Vassilevsky <antispam_bo...@hotmail.com> wrote:> Noway2 wrote: > > On Oct 21, 12:33 pm, Vladimir Vassilevsky <antispam_bo...@hotmail.com> > > wrote: > > >>The formidable FPGAs (>10k cells) come with the very formidable price > >>(starting at the order of $100 and growing exponentially with the number > >>of cells). Building any of the hyped applications of FPGA will probably > >>cost hundreds if not thousands. > >>Then why all that commotion about FPGAs? Just because it is the cutting > >>edge of the technology? > > > I suspect a lot of what you hear and see has to do with > > marketing propeganda more than anything. Have you seen the embedded > > systems design magaize for the last year or so? It would have you > > believe that the majority of all new designs are 'system on a chip', > > all in a customizable FPGA. > > Exactly. The amount of hype seems to be disproportionally high compared > to the amount of the real applications. > > > The majority of the designs I have worked on for about the past 8 > > years have all had a DSP and an FPGA, but the designs were partitioned > > so that the FPGA was used mostly for implementing various forms of > > glue logic or for off loading simply state-machine type tasks > > This is what is my experience, too. However the above mentioned > application usually takes only few hundreds of gates or so. Nothing like > the hyped 10k+ cell arrays with the built in multipliers and FFTs. >xc3s400 is 20$-somthing at digikey, it has 8K cells, 16 multipliers and ~35KB ram I'm use them in two projects, in both it sits next to a DSP that does all the house keeping, the low speed signal processing and runs an IP stack. The FPGA generate data and drive number of 100MHz DACs, graps data from a 100MHz ADC demodulate and decimate and send it to the DSP. I can't see anything running at those data rates but an FPGA I've only done a full SOC in an FPGA (big and expensive) for ASIC prototyping, but a few 1000$ doesn't matter when it avoids a 100K mask set -Lasse
Reply by ●October 22, 20082008-10-22
>I don't buy into the FPGA SOC hype though. If I wanted a microprocessor >on a board, I'd be soldering on a real microprocessor, not trying to use>some soft or hard core in an FPGA. > >Regards, >Allan >I think the problem with SOC is the amount of belly-rubbing and head-patting involved with programming the core. You have to hope there's a C compiler that targets the architecture in question, then you have to wrestle with integrating it into your IDE or learning a new IDE, you have to use fancy architectures like Wishbone if you want to hook things up to it, etc. I also think a lot of FPGA designers would naturally decide to create finite state machines to execute a "program" instead of putting in a soft core and writing some C. Together, I imagine that these situations severely hamper SOC adoption. Cheers, Silash
Reply by ●October 23, 20082008-10-23
Silash wrote:>>I don't buy into the FPGA SOC hype though. If I wanted a microprocessor >>on a board, I'd be soldering on a real microprocessor, not trying to use > >>some soft or hard core in an FPGA. >> >>Regards, >>Allan >> > > I think the problem with SOC is the amount of belly-rubbing and > head-patting involved with programming the core. You have to hope there's > a C compiler that targets the architecture in question, then you have to > wrestle with integrating it into your IDE or learning a new IDE, you have > to use fancy architectures like Wishbone if you want to hook things up to > it, etc. > > I also think a lot of FPGA designers would naturally decide to create > finite state machines to execute a "program" instead of putting in a soft > core and writing some C. > > Together, I imagine that these situations severely hamper SOC adoption.Did you ever really try to use one of the soft cores? I don't see your constaints. Some month ago I had to decide to use a fpga with a softcore or not. I have some previous experiance with fpga, but only as "glue-logic" to a dsp. I had a look on the main suppliers (xilinx, altera, lattice,..). Only xilinx and altera came into a preselection and I evaluated the documentation on the net app. 1 week for each. My final decision was altera, because you get more fpga for the same price (digikey) and the documentation was for me much more clearer. I started with a develpmentboard and made some tests. My experiance: The hardware generating tools (quartus2 to include nios2 softcore, altera peripherals, custom peripherals, ..) have a much more seamless interface to the (C-) software as if you have a seperate CPU - FPGA combination. For the standard peripherals the complete HAL (hardware abstraction layer) is created automatically and for you custom logic you get also a minimal interface so you do not care for exaple about adressing. You get it from your tools. On the software side there is an Eclipse-IDE with Gnu-C compiler. It is adapted to the nios2-core from altera, and I see no difference to those of a "real micros". For me this has the advantages: Only one jtag to the programmable parts of your system. Nearly all available memory-type (flash, static, sdram, ddr, ..) is supported. You include only the peripherals you really need. You can include hardware floating point if you need. Custom instructions are a very powerful tool. It is easy to optimise the speed critical aprts of the design down to the hardware. Powerful pll's help you to manage your clocks. The flexibility is very high, you can use more processors on the same chip and with the same or different memory for example. Debugging of hard- and software is in the same environment. There are also disadvantages: It is more expensive. A new architecture (for me it was only an other one). Usually there is no ADC / DAC on chip. I am not an agent from altera and all I stated above, I have tested on the development board and most will be realised on a custom board at the moment.> > Cheers, > Silash-- Freundliche Gr�sse -- Regards F. Seuhs Mailto: friedrich.seuhs a t hasos.com
Reply by ●October 23, 20082008-10-23
>Silash wrote: > >>>I don't buy into the FPGA SOC hype though. If I wanted amicroprocessor>>>on a board, I'd be soldering on a real microprocessor, not trying touse>> >>>some soft or hard core in an FPGA. >>> >>>Regards, >>>Allan >>> >> >> I think the problem with SOC is the amount of belly-rubbing and >> head-patting involved with programming the core. You have to hopethere's>> a C compiler that targets the architecture in question, then you haveto>> wrestle with integrating it into your IDE or learning a new IDE, youhave>> to use fancy architectures like Wishbone if you want to hook things upto>> it, etc. >> >> I also think a lot of FPGA designers would naturally decide to create >> finite state machines to execute a "program" instead of putting in asoft>> core and writing some C. >> >> Together, I imagine that these situations severely hamper SOCadoption.> >Did you ever really try to use one of the soft cores? I don't see your >constaints.I tried to use PicoBlaze to do some small house keeping tasks, but it turns out there isn't a decent C compiler for it and no development tools support its integration. We recently purchased Altium Designer and it appears to support simple integration of soft cores, but I've been too busy to play with it so far. I wasn't trying to knock SOC. I would love to use one for many of the reasons you mentioned. But it is somewhat difficult to implement unless you have some really awesome (and probably expensive!) tools.>Freundliche Gr�sse -- Regards >F. Seuhs >Mailto: friedrich.seuhs a t hasos.com >
Reply by ●October 23, 20082008-10-23
Hi Vladimir, Just like any technology, FPGA's have their niche (an approx $2-3B one, going on FPGA sales). And for DSP applications the FPGA niche is 1) when the product volume is low to mid levels - asic is cheaper for higher volumes. 2) where the DSP computation requirements exceed the capabilities of 1 or 2 DSP's. (how would you build a low cost 50mbit/s digital radio modem doing all the baseband stuff using a couple of DSPs? => use an FPGA ) Your pricing might be right for the "high end" Xilinx Virtex series, but you have overstated the costs for the Xilinx Spartan 3 range. For the latest Xilinx spartan 3A DSP family you can get a 33k Logic cell part for under US$30. (Don't be fooled by the high one off digikey prices - they are virtually meaningless when it comes to working out the cost for a low-mid volume (5k-20k per year) product.) In terms of straight out performance and $ per MAC the FPGA would be about 10x cheaper than a DSP chip. So there is another reason to use a FPGA - lower cost than a DSP chip. The downside is you have to work for it - the tools and algorithms are not as mature as for software, and you are often working at a lower level of abstraction so the design just takes more work. I'm currently working on a product with a digital radio receiver that has a matched filter bank requiring 26 Giga multiplies per second. I can do around 100 Gmult/sec in the $30 chip I mentioned above - tell me which DSP even comes close to that level of performance? Try asking your question on the comp.arch.fpga newsgroup and see the response you get...:) Regards Andrew On Oct 22, 5:33�am, Vladimir Vassilevsky <antispam_bo...@hotmail.com> wrote:> The formidable FPGAs (>10k cells) come with the very formidable price > (starting at the order of $100 and growing exponentially with the number > of cells). Building any of the hyped applications of FPGA will probably > cost hundreds if not thousands. > > With so much fuss about doing the signal processing in FPGAs, FPGA > development tools, FPGA student projects, zillions of articles, etc, I > am wondering if anyone actually used FPGA to make a sellable product. > > Of course, there could be the applications like radars or cellular base > stations or prototyping CPUs or ASICs. However this is a very special > and narrow nitche. > > Then why all that commotion about FPGAs? Just because it is the cutting > edge of the technology? > > Vladimir Vassilevsky > DSP and Mixed Signal Design Consultanthttp://www.abvolt.com







