#181 – An Interview with Dave Vandenbout

[GiskardReventlov]

Since I'm new to this place I hadn't really discovered the Amp Hour. This show was really good. Dave V. is a cool guy and I like what he's doing.  He looks too young to have done all those things he mentioned. For what it's worth, Dave V. if I had known about Xess (say like excess) I would have bought one of your boards instead of the de0-nano.

Dave J. mentioned that if you buy an ASIC it's guaranteed to work and Dave V. did make the argument that "well if my FPGA design gets turned into an ASIC design the ASIC doesn't undergo a magic transformation and become bug free" (not a direct quote but something like this).  I think he understated this. Think back to the Penitum floating point bug, etc., etc.  An ASIC has lived as software at some point in a simulator, etc. I don't know much at all about ASIC design but they aren't magically better.  And they can't be changed of course. I don't think I'm saying anything you don't already know but I think Dave V. was understated in his rebuttal of Dave J.'s assertion that ASICs are better because they're, well, ASICs. Maybe the ASIC design is more scrutinized but still it's false comfort.

My experience with FPGAs is that they are powerful and the big 2 are trying to get them into the hands of hobbiests/amateurs but it's a steep learning curve. Couple that with the fact that they drop support from their IDEs (integrated devel. environment) for older chips and it can be a big dissappointment.

Heck just getting the Altera IDE installed is ridiculous.  I think it was  11GB when it was finally installed.  I'm not worried about disk space but that's a lot of code.

I don't think I'm covering any new ground here but it's an interesting topic and I enjoyed the show.

My question to Dave V (and anyone) would be:
"With the FPGA hardware makers obsoleting their hardware, is there a chance to do an opensource FPGA chip."

This way we get something stable and we decide when to obsolete the HW. Maybe there are FPGA chip patents that have expired?

[ChrisGammell]

If you enjoyed this show about FPGAs I also recommend going back and listening to the one with Philip Friedin, who is on the EEVblog forum if I'm not mistaken. He gives more insight into the field from the perspective of a chip maker: http://www.theamphour.com/the-amp-hour-103-xenodochial-xilinx-ex-employee/

Short answer I think (because I asked something similar towards the end of the show) is: no. Open source doesn't really have much of a place in a field closely controlled by a few players.

[GiskardReventlov]

(I replied once but no idea where the post went, it's just gone)

If you enjoyed this show about FPGAs I also recommend going back and listening to the one with Philip Friedin

I will do.

Short answer I think (because I asked something similar towards the end of the show) is: no. Open source doesn't really have much of a place in a field closely controlled by a few players.

I relistened, I think there's a place for it but it just doesn't involve those few big players.  But then again maybe it's just wrong to think about an opensource FPGA. It may be a contradiction in terms.  What I mean is that FPGAs and ASICs makers live in proprietary worlds. Opensource can find similar functionality just in a different way. The more I think about FPGAs the more I want to do something else because the big players can pull the rug out and that's bad.

Maybe not a great example since it's moving so fast but...  In the bitcoin world the bitcoin mining started on CPUs, then went to GPUs, then FPGAs and now mining is mostly done on ASICs.  During that brief time (months) when mining was done on FPGAs I recall some people working on the code for the FPGAs and trying to fine tune it and then put that in a board and sell it. But by the time they got it fine tuned ASICs were available. Not to say that they didn't ever mine with FPGAs but they weren't able to move fast enough.  The writing was on the wall for FPGAs and mining. I guess my long-winded point is just that FPGA takes time and once the FPGA design is polished then it's ASIC time any way if you're serious. So FPGAs are very niche.

I'm still going to check out that myhdl that was mentioned.

[lgbeno]

Knowing a little more on the ASIC side, I'll say that your thought process is going in the right direction.

The comment about ASICs being more "proven" or containing less bugs is simply a factor of the amount of verification that goes into developing an ASIC.  Just think, if it costed $200,000 and took 3 months to print a document, you will likely proof read it (test it) more than if it cost 10cents and took an hour.  Through that rigorous process you will probably find that missing comma that if you read the doc quickly, you might look over. 

Also if this was your life (and money), you would likely invest 500k or a million dollars into EDA tools & models that allow you to verify timing, manufacturability, etc, etc.  all to improve your success rate.

In the end, there is nothing about ASICs that make them intrinsically less error prone.  In fact the is probably more that can go wrong.

Fpgas lasting and biggest problem is that there is so much resources that make them field programmable.  It's not the logic itself that makes them so expensive but more so the re-programmable routing fabric that makes the programmable.  This same fabric is what causes speed issues, power issues and cost issues. 

It's simple math to determine what makes the most sense economically for a product. If the volume high enough then the per unit costs outweigh the upfront costs.  This is a better fit for an ASIC.

Low volume may be able to bear a higher unit price that's a better fit for FPGAs.

Aside from a pure cost analysis, the bitcoin guys jumped to ASICs so quickly because the algorithm was well defined (low risk of change) and they really needed performance, ASICs will always have better performance than FPGAs because there isn't a programmable routing fabric.  Also there is a power problem with bit coins since a key metric is hashes per watt.  Lastly if ASICs have lower per unit costs, they are higher margin for the vendors and/or low cost to consumers.

Wrt to obsolescence and tool support, IMO it isn't as bad as the Dave's made it out to be.  Sure Xilinx ditched a lot of their product lines when they came out with vivado but that is a little of an anomaly Altera still supports Cyclone I devices which are probably 10 years old now.  You can always load old software if you must fix one of those products but it's not an everyday occurrence.

Developing an open source FPGA would be a massive undertaking and would yield something so far from state of the art that I doubt many people would be interested.  I don't thing the field is that closed, you can get Altera and Xilinx tools for free and kits for under $100.  That's not too bad, it's taken the big 2 many years to get to that point.

In the end though with so many cheap micros coming to market with basically every permutation on peripherals, I think Faves main point is that for any given application, you could probably make a micro or dsp do the job.  Kinda sad.

But know that for those who design micros and most ASICs, they are prototypes on an FPGA first.  Ever see this:

http://chipdesignmag.com/display.php?articleId=333

Before Zynq was taped out it was prototyped on a board with multiple top spec FPGAs

[GiskardReventlov]

lgbeno, nice post. I just had a thought.  Since FPGAs are an integral part of ASIC making is it possible that FPGAs exist because they are needed to make ASICs?  So since the FPGA makers are selling them to ASICs makers why not increase your market and sell the to other suckers, er, I mean customers that aren't ASICs makers.  I know that technology has a way of leaking out like this.  If I recall correctly, an example is the DLP chips that are used in projectors. Those chips were designed for optical network switching and then found use in projectors. I also recall that in the U.S. patents can be extended by simply finding another purpose for the device. Don't know if FPGAs match that model.  (the DLP might match that)

So my long winded question is "Were FPGAs born out of a non-ASIC-maker market? Or are they born out of an ASIC-maker market?

[hamster_nz]

I think neither. FPGAs are enabling technology for a few corner cases - mainly low run volume, high dollar value, relatively high performance, where low latency is needed, and the needs occasionally change. Things like high-end network switching, broadcast video processing, RF / Radar, avionics / space system, control for bespoke engineering systems.

If the volumes are higher (as happend with bitcoin mining) then ASICs become viable. If the unit cost is too low then you can't afford the chip itself or to recover the engineering effort. If performance / latency needs are low then you can do it with a micro-controller or DSP.

I wouldn't be surprised if the "No Such Agency" wasn't the biggest user of FPGAs.

However, I have seen smaller FPGAs in our alarm/access control system here at work, and they were on HP iLO boards at one time.

[lgbeno]

I've done a lot of work with medical imaging electronics, FPGAs play a big role there.  Part of it is the low volumes and also the mix of things that need to be accomplished in logic.  There is also different regulatory scrutiny applied to programable logic vs software.

FPGAs are big in things like high end video processing, film scanners (now a dead market).  There are quite a few applications IMO.

Back in the day people were using discrete logic gates like 74 series to do digital electronics, it took a lot of parts and huge board to implement simple stuff.  Then PALs GALs and PLDs came out, evolved into CPLDs and then finally FPGAs.  To me I think that ASIC prototyping is simply a happy side effect of FPGAs coming onto the market.

Fun fact is that one of the hottest areas that I know of for FPGA development right now is for low latency trading systems.  Definitely do not want to get into an ethics discussion on that topic though...

[GiskardReventlov]

I think neither.

So ASICs makers saw them as a tool and started using them.

If the volumes are higher (as happend with bitcoin mining) then ASICs become viable. If the unit cost is too low then you can't afford the chip itself or to recover the engineering effort. If performance / latency needs are low then you can do it with a micro-controller or DSP.

With bitcoin higher volumes were not the motive.  It was all about electric power per work.  And really I think it's become just about sheer compute power as places with geothermal are doing massive mining on ASICs. I understand what you're saying just bitcoin's not a good example in this "volume" argument.

I wouldn't be surprised if the "No Such Agency" wasn't the biggest user of FPGAs.

 
Whole other can-O-worms.  Could be, I think I heard something like that before. I wonder what they do with their old gear.

I also wonder how they create ASICs for themselves.  How would you make an ASIC secure? Seems like a lot of potential points of compromise.

However, I have seen smaller FPGAs in our alarm/access control system here at work, and they were on HP iLO boards at one time.

In the other video on FPGA mentioned by Chris, they do talk a bit about smaller, single-purpose, multiple building blocks.

[GiskardReventlov]

There is also different regulatory scrutiny applied to programable logic vs software.

That's got to be a huge motivation. I think that was mentioned recently, maybe on the older FPGA show.

FPGAs are big in things like high end video processing, film scanners (now a dead market).  There are quite a few applications IMO.

Tweakability is a good thing, er, I mean field programability.

To me I think that ASIC prototyping is simply a happy side effect of FPGAs coming onto the market.

Interesting relationship that FPGA and ASIC makers rely on each other.

Fun fact is that one of the hottest areas that I know of for FPGA development right now is for low latency trading systems.  Definitely do not want to get into an ethics discussion on that topic though...

You mean high frequency trading.  But I guess low latency is pro forma.  I'm fine with it, we just need to have transaction fees.