Open Source development at Altera

Does he say anything about opening Altera FPGA bitstreams?  From the slides, it only looks like Altera is open-sourcing CPU stuff.

He mentions that open source Altera design tools might be possible in the future.  It does sound like Open Source ideals are starting to spread through the company... so I guess that gives a little hope.

Interesting - thanks for sharing, fustini. The open sourcing of FPGA bitstreams, etc. seems to have Altera and Xilinx caught in a little (see: massive, high-stakes) game of chicken. Something's gotta eventually give...

I don't see any great pressure - the effort to write open source mapping/fitting tools for the X and A parts would be huge and experience from GCC and other such things (like GPU video support) suggests that not much will happen unless actually supported by the IP holder (X or A in this case).

Both X and A are largely software companies (ie the software to support the FPGAs is a huge chunk of their R&D spend) and they both seek competitive advantage from owning it.

What might change the picture is the emergence of a new FPGA maker without the resources to do all their own tools but so far they all seem to go and license from Synplicity -  which they do because there is no viable open source alternative. So it's all a bit chicken and eggish - if you've invested a huge load of cash into tools you don't want to give them away, and you can't establish a new chip without using tools with a huge investment in them.

I'm not really all that bothered if I could use open source tools with X and A - where they (open source tools) would help is opening up competition to X and A - which of course they are both well aware of which is why they don't open up their bit streams.

Lattice are the underdog in all this and their interest in tiny FPGAs may give them a different attitude - but I can't see the open source ICE40 effort (which Lattice don't appear to be trying to stop) making a big impact in the short term.

MK

Michael Kellett wrote:

 

Both X and A are largely software companies (ie the software to support the FPGAs is a huge chunk of their R&D spend) and they both seek competitive advantage from owning it.

I don't have insider knowledge of any FPGA companies, but my opinion is that X and A are primarily silicon companies and they don't make money unless they sell that silicon.  They have to develop and maintain FPGA software to sell that silicon, but it's pure overhead.  If they didn't have to develop and maintain that software, the silicon would be a lot more profitable.

Compare this to a typical microcontroller company.  They don't have to write their own software -- they can give away a free GNU tool-chain or let customers license software from ARM.  The microcontroller companies can concentrate on silicon and adding clever peripherals to distinguish their parts from competitors.  They don't have to re-invent the wheel like the FPGA vendors all need to do since they can't use a shared open-source tool base.

IMO FPGA companies try to spend as little resource as possible on software, and I believe this is clear from looking at that software.  Basically, what FPGA vendors want is to be an alternative to ASICs.  The software serves that purpose.  It's designed for FPGA professionals who know what they're doing, with a steep learning curve for new users.  The software is slow for complex designs, which is not a big issue if you're competing with large ASICs but which is impractical for reconfigurable computing.  Reconfigurable computing doesn't sell a lot of silicon, so it's understandable that FPGA vendors wouldn't want to spend software resources on it.  If the software were open-source, reconfigurable computing researchers could create their own software and languages that would be better suited to the task.

Here's another consideration: I don't know if this applies to any of the FPGA vendors, but one thing I've seen in medium to large companies is what I call the "business area one" phenomenon.  Basically, the part of the company that brings in the most revenue calls the shots and makes sure "business area one" gets an outsize share of resources, starving other parts of the company that could be helping the company as a whole become more successful.  For an FPGA vendor, silicon is clearly "business area one" and software is an overhead.  If there are limited resources, who is going to get "first dibs"?

JMO/YMMV

michaelkellett, to your point, existing players face a lot of pressure from their existing/established customers - see: defense & aerospace - (in addition to wanting to protect their IP) to continue to keep things closed. I definitely agree that at some point we'll see a bit of a shift as FPGAs make their way deeper into the maker/education space (assuming that ends up being a substantial enough source of revenue), perhaps starting with older, otherwise-unsupported architectures.

johnbeetem, the FPGA vendors actually expend a ton of resources on software development (and when it isn't practical to build it, they just buy it - e.g. Vivado, HLS, etc.). But it's definitely true that the revenue from development tools is a drop in the bucket. You may have event seen the recent announcement that Xilinx made their High Level Synthesis tool completely free for select devices - which is pretty amazing honestly.

Pretty exciting stuff. Hopefully FPGAs are finally ready for primetime!