can we design FPGA?

[gary36]

As an R&D idea from a tech school, it was decided to design a custom FPGA. Is it even possible? If yes, what are major challenges

 

[dpaul]

@gary36,
Theoritically possible, but bad idea to complete the project within a stipulated amount of time considering a bunch of tech school guys!

Or did you mean design for FPGA implementation?

 

[KlausST]

Is it even possible? If yes, what are major challenges

Obviously someone already did design an FPGA .. so it has to be possible.

So the question to me is: What is the context, what are the objecitves, the requirements?

Klaus

 

[gary36]

it is an R&D initiative to study the feasibility for custom FPGA solutions without depending on the third party vendor. The work includes complete tool chain design along with FPGA.

 

[KlausST]

Hi,

so you want to
* design a custom FPGA, or
* configure an existing FPGA?

Klaus

 

[barry]

An FPGA is a piece of silicon; is THAT what you want to design, or something USING an FPGA?
What do you mean by "custom solution"? An FPGA is a blank slate from which you can create a custom solution.

Creating a "custom FPGA" sounds like creating a "custom artist's canvas".

 

[danadakk]

it is an R&D initiative to study the feasibility for custom FPGA solutions without depending on the third party vendor. The work includes complete tool chain design along with FPGA.

I am guessing a team, facility, resources, say maybe $ 100M to create your own custom tools
and FPGA architecture might do the trick. Sans creating your own fab facility and manufacturing
and rel and test and.....,for that figure $ 1B .

Your goals, timeframe, to complete task ?

Regards, Dana.

 

[kaz1]

it could be a brain storming exercise to find out some new ideas about configurability of silicon platform. Once done you may use the service of ASIC people like Barry to try build a prototype. Areas to consider:

- implementing logic, most fpgas use LUTs
- soft/hard ip selection
- switching
- overall architecture, resource, clocking, io ...etc.
- power reduction problem, this is major factor against current FPGAs
- sim model, compiler
- programming method, jtag ...

Good luck

 

[dick_freebird]

The silicon is the least of it. The development
environment will make or break you.

Nonvolatile FPGAs are probably out of reach,
the NV elements are often a qual and programming
nightmare (for reasons). A SRAM based FPGA is
all standard devices. You could probably target one
of the early, smaller FPGAs from Xilinx.

There's small outfits doing bespoke FPGAs today.
QuickLogic is one I like, very aggressive in making
special ones for specific customers. They put out
a lot of material that a class might find educational.
The main man is pretty friendly and might be
convinced to lend some support.

If you're talking useful resume material, I think
more companies are interested in using FPGAs
to do interesting things, than in building them.
Employers for FPGA chip designers, are less than
10. Worldwide. Give or take.

 

[gary36]

let me start with simple GAL device. is it possible to redesign?

 

[KlausST]

Hi,

maybe you should answer a couple of our questions first ...

Klaus

 

[gary36]

I understand that the answer is a big "NO" and it may not be possible for small ventures to pursue this task. However, my only question is that can we start with simple GAL devices, where JED formats are well known and perhaps a commercial tool can be used to generate the fuse file. Also I am looking for an answer on a deeper level. Is it the storage technology(flash, EEPROM) which really matters(as it the patented), which prevents small ventures to get into this domain? Why no one has come up with open source FPGA?

 

[kaz1]

I understand that the answer is a big "NO" and it may not be possible for small ventures to pursue this task. However, my only question is that can we start with simple GAL devices, where JED formats are well known and perhaps a commercial tool can be used to generate the fuse file. Also I am looking for an answer on a deeper level. Is it the storage technology(flash, EEPROM) which really matters(as it the patented), which prevents small ventures to get into this domain? Why no one has come up with open source FPGA?

You seem unaware of history why fpgas were developed instead of ASIC fabrication.

 

[KlausST]

I give up..

Klaus

 

[dick_freebird]

Perhaps you'd be better off going bottom-up - the config root element (RAM, fuse, anti fuse), latches, CLB architecture at the lowest ("leaf cell" for array) levels with the end game being to assemble into something that fits an existing dev system and any programming hardware. Look to that last, for a field of candidates.

 

[kaz1]

Perhaps you'd be better off going bottom-up - the config root element (RAM, fuse, anti fuse), latches, CLB architecture at the lowest ("leaf cell" for array) levels with the end game being to assemble into something that fits an existing dev system and any programming hardware. Look to that last, for a field of candidates.

It is a bit new concept when reading OP statement:

" Why no one has come up with open source FPGA? "

The way I take it is that he is asking for an fpga chip design to be implemented on some other FPGA, else he is moving away to fabricate it as ASIC yet FPGA was/is meant to bypass ASIC due to cost, debugging and configurability.
So it is bizarre either way.

 

[dick_freebird]

I took it as an open source hardware project.

 

[dpaul]

Why no one has come up with open source FPGA?

I still do not get your use case!
Who would use your open source FPGA architecture?
What happens from design entry to netlist generation is already know. The generation of bitstream from netlist and mapping it on to an FPGA is the IP stuff going on. So for a standard user like me, all it matters is that my design should work as expected with the least amount of engineering obstracles, that's it.
To achieve the above why would some someone invest heavily (time and money) to create open source FPGA architectures?

Cost sensitive companies would use open source tool chains, use free-to use licenses, generate bitsream and just buy the pieces of FPGAs as needed and program them that's it!

 

[dick_freebird]

I also took it that this was about the journey, not the destination.