FPGA designs

[mosc]

On another topic, http://electro-music.com/forum/topic-10153.html& , Doc Justice posted some very interesting stuff about FPGA devices. I thought I'd start a new thread here in case anyone was interested.

---

[mosc]

[jksuperstar]

I guess it's a good time to give a little history/lesson about FPGAs. The first programmable devices wre based on EEPROM like technology, connecting logic gates (AND, OR, Flip-Flops) with each other through fuses. This was very limiting because you only had the gates that where in the chip, but was nice because you could burn & reburn new a "image" into the device at will.

This technology is still available, though it's gone through transformations through Flash based technologies. They are called a CPLD today, and many can be had for less than $1, and with one of these one could easily replace a whole handful of typical logic devices such as 74x04, etc. The tools needed to program a CPLD can be found for free at any one of the chip vendors taht seels these (see below).

Next came the BIG BOYS. These are known as FPGAs, and use SRAM as their technology. They are geared more for the large systems: for example, I have designs that are PCI based, with high speed serial (>100MHz) and an embedded RISC microprocessor (that was compiled from verilog!) running at 80MHz. These are no joke. By using RAM technology, logic is boiled down to look-up tables rather than actual logic gates. This allows large complex logic schemes to be built without significant timing impact due to the latency inherent in each gate. If you have 6 inputs and 2 outputs, the typical latency is ~200ps, really.

To design big things like this, there are two main "languages" used, much like one would program in C or Pascal. Whereas designing in schematics is more like writing assembly or machine code. Tools then synthesize these languages into either gates or look-up table values. This intrun is run through another tool that automatically places the tables inside the FPGA, and routes the interconnect needed. Much like an autorouter does for PCBs, but these are on much more caffeine and alot of steroids. Oh, and these tools can be found for free from the chip vendors (Xilinx, Altera, Lattice, etc). In terms of number of gates, Xilinx has many for under $10 that report 5million+ gates, but in reality this relates to something more of about 500,000. But that's ALOT of gates for that price.

The only drawback to FPGAs is the power on time: they require a couple milliseconds to "boot-up", which also requires an external programming chip to do this (like loading your program out of flash and into RAM). That is to say: the FPGAs need to be programmed each time they are turned on. But this can happen an infinite number of times without any harm to the device.

www.xilinx.com
www.altera.com
www.latticesemi.com

[DrJustice]

mosc, good of you to move the FPGA talk over here!

And jksuperstar, nice history rundown! For those who don't know, jk is a logic design professional.

From the NM forum:

[blue hell]

Hey, cool thread. Got some reading up to do now, or read upping.

[jksuperstar]

[deknow]

....i also have some reading to do...thanks!

i really would like to port my "electroflute noise filter" to a platform where it is harder for others to see what i'm doing than it is on the g2. the g2 proves it works...i've ported it to synthedit (not refined, but works well enough to know it wouldn't take to long to tweak it), but i want it in a stompbox that i can sell to flutists (haha, both of them that are interested in electric flute playing).

deknow

[DrJustice]

One thing I've been pondering is the possibility of making a really powerful additive synthesizer voice without latency. This might imply the use of an oscillator bank. It would have high resolution amplitude and phase profiles for each harmonic. So a phase vocoder++ in hardware, more or less. There would have to be a rather fancy control system for introducing performer induced dynamics in the sound. In conjunction with an analysis part there could be support for resynthesis, even in real time. It could hopefully be made using an off the shelf evaluation board, or several of them for polyphony.

Just ponderings for now though...

DJ
--

[DrJustice]

For those that would like to read more, here are some useful FPGA links:

An FPGA FAQ.

Several VHDL and Verilog tutorials (the Aldec interactive one is fun).

A couple of other Verilog tutorials can also be found here and here.

I use Verilog personally. The syntax is reminiscent of C. VHDL is more like Pascal, Ada and PL/SQL, and is often viewed as more "industrial grade". For starting out I'd strongly recommend Verilog, it is known to be quite a bit easier to get going with. The two languages can usually be mixed at the module level and you can even mix HDL and schematic entry in some tools.

If you want to get going with actual coding and simulation I recommend ModelSim. It will compile your code and simulate it with waveform display - very nice tool. You get ModelSim with the Xilinx Webpack. It is available as a separate download. The seeming Xilinx bias is because thats's what I know...

I'm sure jksuperstar will have something more to say on the subject.

Happy studying

DJ
--

[jksuperstar]

You hit all my favorite links already Mainly I refer to opencores.org alot, great set of examples that go from simple communication protocols like SPI all the way up to full on RISC processors with set-associative cache.

http://www.verilog.net

and don't forget your emacs "verilog-mode" lisp!!!
http://www.verilog.com/verilog-mode.html

Getting off-the-shelf FPGA boards is usually over $100 per board, typically closer to the $400 range. That's partly why I'd like to make a board (maybe through Olimex or someone), since it could be made for $30-$50 (I think), and have some audio specifics

[DrJustice]

I guess you mean PCI or PCI-express when you mention $400? That's probably on the mark, but there are some exceptions. Enterpoints PCI based (and stand alone) Raggedstone is a rather nice offering in this segment.

But, yes - a cheap'ish board geared towards audio would be very nice. Perhaps with RS232, MIDI, IIS and a TDM bus for hooking up several cards? What more would we want on such a board? Some kind of community project would be way cool!

DJ
--

[blue hell]

Well if ever you want to know any details about me, just ask Xilinx, they now know.

But I got their stuff, including the simulater, installed & licensed & runnable, so I'm all ready now I gues to try and make an and gate or something similar in complexity :-)

Meanwhile I can't quite keep up following the links posted here, but I'll try harder, and soon it will be weekend !

[blue hell]

[DrJustice]

[blue hell]

[Oval]

http://www.xilinx.com/xlnx/xebiz/designResources/ip_product_details.jsp?key=HW-V4-ML403-USA
IMHO the best dev board for this thing, as it has already PPC 405 implemented in one core with FPGA, and AC 97 codec.

But i prefer for full dev kit based on this as with full of development software (linux project example included) and download cable
click here

[jksuperstar]

Wow, that Ragged Stone IS cheap. $90us, cool!

Great Link!!

[DrJustice]

Oval!

Thanks for joining us!

Those boards you mention are wicked cool, and good value thinking of the Virtex 4 and all the other bits that are on them.

I'm adding a link straight to the Virtex 4. This is the big slammer, look at the FX folks.

DJ
--

[Oval]

From the NM forum:

[Oval]

[JovianPyx]

[DrJustice]

That's really cool Scott! I suppose the 'organically' varying sound in some of those samples ar due to the noise modulation you mentioned? What are you using for audio output? Any plans to make the project available?

A modular approach is indeed a good idea. I wonder if a standarized module interface could be cooked up, to allow community projects with reusable modules...

DJ
--

[JovianPyx]

[DrJustice]

[JovianPyx]