Xilinx sends lawyers after online educators.

[wraper]

There's that old saw that goes, "If the young hobbyist learns Brand [$BRAND]'s devices and software, when s/he gets on the job s/he can push the organization towards that brand's parts." And that's bullshit. Any company that already does FPGAs already has a preferred vendor which won't change unless there's a Damn Good Reason to do so.

And all of those companies are of course many decades old and FPGA choice was made by godly existence and not real people... Did you hear about such thing as startup? Many of those made well established players bite the dust.

[Berni]

A bit of the problem is also that FPGAs tend to be useful only in pretty advanced projects.

Typical maker projects are just gluing together existing modules and copy pasting some arduino code together. The usual arduino user wouldn't even bother to learn the C language properly, just enough to copy paste and randomly figure out how to get something together. At the same time microcontrolers and SBCs like the raspberry keep getting more powerful, letting them do some pretty impressive stuff.

The only sort of projects where you do still need a FPGA is places where you move around very large amounts of data. So things like live video processing or running anything with a very fast ADC or DAC in it such as a fancy SDR or test equipment. All of these tend to be projects that involve >1 month of just programming to get basic functionality running. The extra complexity of FPGA design and the low level nature of it makes it a lot more work to learn than just C, so makers cant be expected to learn it if they can't even get themselves to learn C.

Tho in my opinion getting an electronics engineering degree should involve at some point designing and building a custom CPU from scratch on a FPGA.

[janoc]

There's that old saw that goes, "If the young hobbyist learns Brand [$BRAND]'s devices and software, when s/he gets on the job s/he can push the organization towards that brand's parts." And that's bullshit. Any company that already does FPGAs already has a preferred vendor which won't change unless there's a Damn Good Reason to do so.

And all of those companies are of course many decades old and FPGA choice was made by godly existence and not real people... Did you hear about such thing as startup? Many of those made well established players bite the dust.

And how many startups actually use FPGAs and manage to survive long enough for the product to see the market? FPGA is a very expensive solution for niche needs, you won't find it in cheap consumer gear, for example.

I think the only recent exceptions that I am aware of are the OSVR HDK 1.0 head mounted display kit (in HDK 2.0 they have removed the FPGA as part of cost cutting because nothing used it even in the 1.0 version) and the project is dead (AFAIK) today. Its former CEO is now head of marketing for that laser phone charger thingamajig that Dave made a video on. One other consumer product that uses an FPGA (Lattice Ice40 I believe) is the HTC Vive and its controllers - as part of the Lighthouse tracking system where the thing is used to measure pulse time from the 30-something photodiodes. But that's gear that costs $500+.

[wraper]

And how many startups actually use FPGAs and manage to survive long enough for the product to see the market?

Not that many. But some of them survive and become big companies. And by the time they become big companies they already may have FPGA preference inherited from early days. To name a few widely known, say Tesla, Spacex, Siglent. As already said many of applications are quite specialized therefore those companies are not that widely known. https://angel.co/fpga-1/jobs

[wraper]

Also it's not like it would be necessarily hobbyist who will use free tools and free lessons. It may be experienced engineer who has never done anything FPGA related at day job. Then learns on it's own and the uses it in practice, probably at startup.

[NiHaoMike]

A bit of the problem is also that FPGAs tend to be useful only in pretty advanced projects.

Typical maker projects are just gluing together existing modules and copy pasting some arduino code together. The usual arduino user wouldn't even bother to learn the C language properly, just enough to copy paste and randomly figure out how to get something together. At the same time microcontrolers and SBCs like the raspberry keep getting more powerful, letting them do some pretty impressive stuff.

What would be nice is some sort of maker-oriented device or subsystem that contains a FPGA but is still easy to use - e.g. a drone flight controller or 3D printer/CNC motherboard. The average maker isn't going to do anything with the low levels of the FPGA but that's not the point.

[Berni]

Well a CNC controller does have some use for FPGAs. It can be used to interpolate between the movement nodes very accurately in real time and generate very timing precise step signals. There are a few that use it, but most others are directly run from a microcontroller because the accuracies of the hobby printers, router, engravers are just not high enough for the sloppy MCU control to make much of a difference. And if you know what you are doing in terms of programing you can get rather tight timings out of MCUs

In terms of what is actually using FPGAs. Well all of these high speed camera startups such as Chronos and similar have to use a FPGA to move data around. Sometimes game console mod chips for older consoles might include a FPGA to fiddle with some communication bus in real time. There are also some retro gaming video upscalers that turn console video into HDMI, one in particular even did deconvolution in real time to fix blurrynes in N64 graphics. Ben heck also had a good use for a FPGA where it was used to turn the LCD signals in a GameBoy into VGA video so that it could be shown on a big screen.

Also i have spotted a fair bit of programmable logic in test equipment. Some of HP/Agilent PSUs and electronic loads have Xilinx FPGAs in them and here and there you sometimes find a CPLD for doing glue logic. Sometimes they modernize an old instrument design and minimize all the separate peripherals of a 68000 CPU down into a single FPGA chip. This makes the board much smaller while still running the same firmware. A lot of high end networking equipment also has FPGAs in there because its the only way to get so much data around.

When you strike a deal directly with a FPGA manufacturer you can get a big FPGA with transcievers and everythyng for 30 USD. But they will of course want you to buy a pretty large quantity. While Lattice chips can be so cheap that it even makes sense to use them for fairly dumb tasks that are simply annoying in nature for a MCU.

[daqq]

A lot of high end networking equipment also has FPGAs in there because its the only way to get so much data around.

Also low end but specialized networking equipment - I've seen a funny use of an FPGA where they needed a lot (30+) of serial lines transmitted via a single Ethernet connection. Basically they just used a Zynq - the ARM bit did the networking and linuxing, the FPGA managed all of the serial lines.

There's a lot of places where there are FPGAs used. It's true that most of the time it's a product of a larger team/company, but that is not a rule - and should not be.

[Berni]

Yes i should have mentioned that too.

They are very useful when you have to do a weird thing that nobody made a purpose built chip for because its such an obscure case. Or you need to interface to some weird chip or bus that doesn't speak any protocol a MCU or MPU might have. For example almost all ADCs and DACs >200MSPS have a proprietary LVDS bus on them that your only hope is to connect to a FPGA.

[EEVblog]

WOW, this is called nominative fair use, the Xilinx lawyers had better look that up.
https://en.wikipedia.org/wiki/Nominative_use

[EEVblog]

If @IntelFPGA or @latticesemi are smart they will put out an official response condemning this unprecedented attack on creators

[Nerull]

Do people not actually *read* books now?  I chose this example for the simple reason I am reading it at the moment. 

Read the text at the bottom of the page.  That's *all* you have to do to keep everyone happy.  I could easily choose any other device specific book in my library.  It would have a similar section on  trademarks somewhere.  Publishing houses know they need this.  YouTubers don't, so they get scary letters from lawyers.

I am not a lawyer.  I planned becoming one when I started university, but by the time I got ready for law school it was clear that the legal profession was not what it was in my grandfather's day.  So I became a scientist instead.

Maybe people just read better sources than you - companies are not *forced* to send C&Ds about any use of their trademark to keep their trademark. That's just trivially wrong.

If another tissue company tries to sell boxes of tissues named Kleenexes, than yes, Kleenex needs to do something about it. They do not, however, need to send C&D notices to the random guy in a store who tells someone to go pick up a box of Kleenexes. They are not required to send C&Ds to anyone who talks about Kleenexes. They are not required to send C&Ds to anyone who writes a how-to guide on using Kleenexes. They are not required to send C&Ds to anyone who reviews Kleenexes. They are not required to send C&Ds to me for writing this post, because I am not attempting to misrepresent my association with the Kleenex brand.

Youtube is filled with educational videos on trademarked products, and somehow all these other companies manage not to lose their trademarks without sending C&Ds everywhere.

[Daixiwen]

If @IntelFPGA or @latticesemi are smart they will put out an official response condemning this unprecedented attack on creators

I don't think they care, at least for Intel. IntelFPGA's free support (a.k.a. "community forums") is a complete mess. After more than 6 months working on it, it's barely usable and still not on par with Altera's old forum, which was already terrible. Nowadays the only contact between Intel and the real world is through FAEs (thank god for them ).
Altera was kind of interested from time to time with hobbyists and creators, giving free kits, but so far I haven't seen that behaviour from Intel.

[Berni]

Intel was doing the Edison, Joule, Galileo... etc boards that are targeted at makers. But it never really caught on all that much for some reason. I think the pricing and ease of use was not there.

[rhb]

Do people not actually *read* books now?  I chose this example for the simple reason I am reading it at the moment. 

Read the text at the bottom of the page.  That's *all* you have to do to keep everyone happy.  I could easily choose any other device specific book in my library.  It would have a similar section on  trademarks somewhere.  Publishing houses know they need this.  YouTubers don't, so they get scary letters from lawyers.

I am not a lawyer.  I planned becoming one when I started university, but by the time I got ready for law school it was clear that the legal profession was not what it was in my grandfather's day.  So I became a scientist instead.

Maybe people just read better sources than you - companies are not *forced* to send C&Ds about any use of their trademark to keep their trademark. That's just trivially wrong.

If another tissue company tries to sell boxes of tissues named Kleenexes, than yes, Kleenex needs to do something about it. They do not, however, need to send C&D notices to the random guy in a store who tells someone to go pick up a box of Kleenexes. They are not required to send C&Ds to anyone who talks about Kleenexes. They are not required to send C&Ds to anyone who writes a how-to guide on using Kleenexes. They are not required to send C&Ds to anyone who reviews Kleenexes. They are not required to send C&Ds to me for writing this post, because I am not attempting to misrepresent my association with the Kleenex brand.

Youtube is filled with educational videos on trademarked products, and somehow all these other companies manage not to lose their trademarks without sending C&Ds everywhere.

I spent most of my career as a solo consultant.   So I *had* to know the basics of IP law and contracts.  I have a number of books on intellectual property law intended for consultants and entrepeneurs including  "West's Business Law".   I suspect the people you refer to have not read *anything* on the subject.

The fact that someone does not get a C&D does not mean they won't.   And they may well get one and not freak out the way the guy in the start of the thread did.  It's not as if they have to stop what they are doing.  They just need to insert an acknowledgement that the trademark belongs to someone else.   You've probably seen them, but didn't understand enough IP law to know why they were there.

The loss of a trademark only happens when someone starts using it, the owner sues and the infringer shows that there was a long history of the company failing to enforce the trademark.   And that is "game over" for the trademark owner.  There is no "do over" allowed.  The trademark is now in the public domain and *anyone* can use it for anything they want. Sending out those C&D letters shows active attempts at enforcement.  In a case like this merely keeping a record of sending the letter is probably all that they would need to do.

In short, what you see on YouTube only proves that it is on YouTube.  It proves *nothing* else.

As I pointed out previously the cure is trivial.  It's an explicit acknowledgement of trademark ownership. That can be a single line of text at the bottom of a title screen.

[coppercone2]

Does anyone else think that FPGA is like one of the biggest dogs to work with in the electronics world?

Massively complicated, closed software (and jesus I thought C compilers were bad), hard to use, expensive, difficult to solder, windows can probably destroy your design, etc.

I think I would rather work with any other component then a FPGA if I could avoid it. I think they need all the help they can get to get rid of all the engineering questions associated with them. If they had advanced open source software I might try it. Having to do all the complications in a closed ecosystem?? NO THANKS

I think when business people decide to make something, and they hear 'we need to get a FPGA guy for this' they think 'lets make something else'.

I don't even WANT to learn it because what I am going to get stuck in some kind of closed ecosystem. Even assembly is kind of crappy because you can learn like ATMEL or PIC but it can kinda transfer over. C is better. For all I know learning one of those FPGA systems is like learning freaking FORTRAN!!! What is going to happen eventually with that kind of system is someone is gonna go out of business then you can get a fake ass 'wizard' job maintaining some bloated legacy system some where. I don't want that kind of career path. A open standard for the FPGA industry would alleviate this quite a bit. Until they start making 'build your own ASIC online flash workshop'.

I swear in like 20 years you will have some company that tells you "we have the last two remaining workstations ON EARTH that can handle this version of windows with these updates with this FPGA software from a defunct vendor". They might also have a dedicated acquisitions person that goes out to abandoned siberian oil mining operations to look for replacement hardware and negotiate with the locals for firewood.

 I already have a pretty much dedicated computer that seems to work well with a particular PIC programmer and PICs without the need for voodoo rituals. I can't imagine what can happen with a complex FPGA compared to a simple 8 bit MCU. You can't even get software to work on different RASPI versions sometimes.

[janoc]

Well a CNC controller does have some use for FPGAs. It can be used to interpolate between the movement nodes very accurately in real time and generate very timing precise step signals. There are a few that use it, but most others are directly run from a microcontroller because the accuracies of the hobby printers, router, engravers are just not high enough for the sloppy MCU control to make much of a difference. And if you know what you are doing in terms of programing you can get rather tight timings out of MCUs

Given that the motor control is almost always done by dedicated chips (Allegro or Trinamic drivers are popular) the only thing that the MCU has to be doing is interpret the G-code and calculate the steps. FPGA won't help you any there, the MCU is perfectly fine to send the necessary pulses and the stepper motors don't really care that much if there is a bit of a jitter in the signal here and there.

And even if you are using closed loop control, with encoders and what not, then you are doing it using the hardware timers - both to read the encoders and to generate the motor signals. Typically you would have a controller per motor/encoder pair talking to the main controller board.

I don't see what difference it would make to implement a timer on an FPGA or use a hardware timer in an MCU (assuming it is a decent timer designed for encoders and/or motor control so you don't need to rely e.g. on interrupts to reload it). I don't see where you would get any "sloppy timing" there. Even the expensive industrial machines don't use FPGAs for this - there is no need for it there.

In terms of what is actually using FPGAs. Well all of these high speed camera startups such as Chronos and similar have to use a FPGA to move data around. Sometimes game console mod chips for older consoles might include a FPGA to fiddle with some communication bus in real time. There are also some retro gaming video upscalers that turn console video into HDMI, one in particular even did deconvolution in real time to fix blurrynes in N64 graphics. Ben heck also had a good use for a FPGA where it was used to turn the LCD signals in a GameBoy into VGA video so that it could be shown on a big screen.

Sure but this are hardly projects that would make someone like Xilinx or Intel go suddenly bonkers in the fear of missing out on a market. Which was the point of the discussion, not whether or not FPGAs have their uses (of course they do).
 

[coppercone2]

what if you use LVDT or something for sensing ?

[daqq]

A open standard for the FPGA industry would alleviate this quite a bit

Both Verilog and VHDL are well defined standard languages. What kind of standard would you like?

I think when business people decide to make something, and they hear 'we need to get a FPGA guy for this' they think 'lets make something else'.

Good luck doing something useful with anything above, say, 50 megasamples per second with anything else but an FPGA, or a heavy duty DSP processor (which are generally just as proprietary). There are things that you simply need an FPGA - or an ASIC, which is much, MUCH worse.

I don't even WANT to learn it because what I am going to get stuck in some kind of closed ecosystem.

Not really. The concepts, understanding and skills are transferable. Yes, there are proprietary bits, but there are similar, or equivalent systems regardless of the vendor. I'm currently using Xilinx stuff. True, a switchover to say, Altera would be pain in the ass, but it would be similar to switching from AVRs to PICs. Sure, some things would have to be redone, but the concepts stay the same.

[Berni]

Yes you do have the Trinamic chips to do actual motor driving, but you have to feed them with accurate step pulses. Sure you can generate pretty accurate pulses with a timer peripheral on a MCU but quickly calculating things and reloading the timer and doing it for all axis at the same time while parsing G code at the same time can get pretty hard. Its fine for a little plastic squirting 3D printer. Its not fine anymore when you get to a big profesional CNC machine.

Those big proper CNC machines have feedback controlled servo motors that have control loops running in the digital domain, they move at speeds and resolutions that would require a stepper to drive step pulses in its input measured in MHz. The motion controller for it has to calculate arc motions in 3D space with very high accuracy or the error will show in the surface finish of the metal part you are milling. All the motion also has to always be running at the same speed, its not acceptable to slow down in a high detail area where you might get thousands of G code commands coming in per second, this can again affect surface finish. High speed tapping operations involve synchronizing the spindle going at thousands of RPM to all the other movement. All of this starts to get into some serious electronics with plenty of processing grunt for it.

The speed and responsiveness of FPGAs is simply unparalleled by any other solution. But for the vast majority of cases a microprocessor is more than good enugh.

That being said FPGAs are indeed a pain in the ass to work with. So when you do end up using a FPGA its likely because it couldn't easily be done by any other solution. Eather due to lack of processing power or being too big and power hungry to get the needed processing power, or just simply not having the right kind of port to plug the weird bus into.

Learning FPGA coding is actually not that hard. The HDL languages are indeed a piece of shit but the language itself is not that difficult. More of the difficulty comes out of figuring out how to implement your idea in digital logic since flipflops are a lot dumber than a CPU. Also there is a ton of documentation to dig trough to find out how any of this works. In the end its more about learning digital circuit design theory than actually learning the language.

Also the things you learn about one vendors FPGAs can be applied to other vendors pretty easily. They all use the same language, the FPGAs mostly work identically inside, the IDE workflows are the same, just the buttons look different and are in different spots. Tho porting code from one vendor to another is a major pain but it can be done.

[jmelson]

Does anyone else think that FPGA is like one of the biggest dogs to work with in the electronics world?

Massively complicated, closed software (and jesus I thought C compilers were bad), hard to use, expensive, difficult to solder, windows can probably destroy your design, etc.

I think I would rather work with any other component then a FPGA if I could avoid it.

HUH?  Have you ever done an FPGA or CPLD design?  What's the alternative?  A whole board stuffed with 74xx chips?  Revisions with wire-wrap wires?
I will admit there's a learning curve to anybody's tools and the HDL language you choose.  But, I have to say I am now kicking myself for being so slow to adopt the technology. 

Have you used a logic simulator to check that a VHDL file does what you want it to do before even looking at real hardware?

I actually PREFER an FPGA to a micro.  With the simulator, and the methodology of HDL design, you can be much better assured of predictable FPGA functionality, with no really crazy side effects of one process/subroutine interfering with some other one due to mixed up variables or whatever.

I have several dozen CPLD and FPGA designs I'm currently manufacturing/supporting.

As for soldering, there are MANY smaller FPGAs that are in reasonable quad packs, starting at about 100 pins.  There are some CPLDs in 44 pin packages.

As for Windows, well i do all my CPLD/FPGA work in Linux.  Yes, there are some slight installation quirks, but I still use some very old Xilinx ise design packages, so I can support some older FPGA families that are the cheapest.  $10 each without internal config PROM, or $13 with internal FLASH PROM.

Jon

[rhb]

The alternative to FPGAs is discrete  ECL logic.  Physical size makes that a lot slower even if it could switch as fast.  This is why hard cores on the chip with the PL fabric are such a big advantage.  The signal paths are much shorter and far better behaved.

So far I've only been reading documentation, but I don't see anything particularly intimidating about FPGAs.  It's a great mass of glue logic on a crossbar switch matrix.  The routing problem is NP-hard, so they use a bunch of heuristics to lay things out.  I'm sure the algorithms are quite interesting involving both convex optimization, Karnaugh mapping  and computational geometry.  The only real problem is the jargon is very vendor specific.  But I bought a DE-10 Nano to accompany my Zybo Z7-20 so I can test my Verilog code on both when I write it.  I'm sure the tool chains will have me pulling my hair out, but that's true of most software these days.

The Zynq supports partial reconfiguration.  So you can load a complex digital filter and get an output every cycle.  Need a different filter?  No worries, mate.  Just load that filter instead.  The only other way to do that is an ASIC.

FPGAs are a compromise between the flexibility of a CPU and the performance of an ASIC. TANSTAFL

[langwadt]

Well a CNC controller does have some use for FPGAs. It can be used to interpolate between the movement nodes very accurately in real time and generate very timing precise step signals. There are a few that use it, but most others are directly run from a microcontroller because the accuracies of the hobby printers, router, engravers are just not high enough for the sloppy MCU control to make much of a difference. And if you know what you are doing in terms of programing you can get rather tight timings out of MCUs

Given that the motor control is almost always done by dedicated chips (Allegro or Trinamic drivers are popular) the only thing that the MCU has to be doing is interpret the G-code and calculate the steps. FPGA won't help you any there, the MCU is perfectly fine to send the necessary pulses and the stepper motors don't really care that much if there is a bit of a jitter in the signal here and there.

And even if you are using closed loop control, with encoders and what not, then you are doing it using the hardware timers - both to read the encoders and to generate the motor signals. Typically you would have a controller per motor/encoder pair talking to the main controller board.

I don't see what difference it would make to implement a timer on an FPGA or use a hardware timer in an MCU (assuming it is a decent timer designed for encoders and/or motor control so you don't need to rely e.g. on interrupts to reload it). I don't see where you would get any "sloppy timing" there. Even the expensive industrial machines don't use FPGAs for this - there is no need for it there.

I think the most common config of linuxcnc/mach3/pathpilot IO is a Mesa card which uses a Xilinx FPGA

it is the most flexible solution, need another timer - just add one, need more quadrature decoders - just add them

and some would argue that control loops implemented in "HW" adds a bit of safety and reliability  compared to software

[mickmake]

Massively complicated, closed software (and jesus I thought C compilers were bad), hard to use, expensive, difficult to solder, windows can probably destroy your design, etc.

I think I would rather work with any other component then a FPGA if I could avoid it. I think they need all the help they can get to get rid of all the engineering questions associated with them. If they had advanced open source software I might try it. Having to do all the complications in a closed ecosystem?? NO THANKS

A lot of work is being done on OSS toolchains for FPGAs. It won't be long before we have grandma noobies cutting and pasting FPGA blocks from websites.

Don't normally plug my own videos, but here's an example of how far the OSS tools have come.
https://youtu.be/CeH5t3GS4Uc?t=656

[EEVblog]

Does anyone else think that FPGA is like one of the biggest dogs to work with in the electronics world?

Yes, famously so, and that is one of the reasons why they are not popular for general purpose stuff.