RE: [oc] Modular FPGA board

[John Dalton <john.dalton@bigfoot.com>]

> you have to be careful throwing a bunch of SIMM connectors on the board
> and expecting the thing to run at 100 Mhz. Many factors play a roll as
> to what the maximum speed might be. You need to wary about trace
> capacitance and inductance, and perhaps termination (which is harder
> to do if it's not point to point). I would suggest to limit the number
> of SIMMs that are connected together to four, and figure out some way
> of bridging multiple groups of four SIMMs.

I guess the trick is to throw the SIMM connectors on the board very carefully! :)
Thanks for pointing out some of the things to watch.  Once I get something
more concrete, I will put it out for review.

> If you look at of the shelf SIMMs, you will notice that those PCBs are very
> small, and the traces very short (== low capacitance). Further when using
> SIMM memories, the max. load on each trace is known, making it easy to run
> them at 133 Mhz. Designing a universal interconnect bus, with unknown
> peripherals is much harder. Just look at the PCI spec if possible, you can
> see how many restrictions there are for both, plug in cards and
> motherboards.

At this stage, I am thinking along the lines of simply wiring up SIMMs
in parallel and working from there.  In the first cut, it can be arranged
to keep a single load on each line, arranging the system as point-to-point
links from the main FPGA to each peripheral (possibly with a short stub),
though this is not really a satisfactory long term solution.

This raises the question of defining some sort of a standard
for the 'SIMM bus'  At the physical level, simply defining which pins are
power and which pins are signal might suffice.  On top of this could
be layered some sort of logical bus structure.  Much more thought required here,
though simplicity is a high priority for me.

For starters, I'm thinking just get something designed/built, then develop it further.

> Another option would be to put a PCI bridge on the board and use a local PCI
> bus for all FPGAs and peripherals.

I'm trying to steer clear of PCI, as IMHO it is overkill for connecting a simple
ADC (or similar device) to an FPGA.  Most chips will not connect directly
to a PCI bus.


Talking about PCI and FPGAs, I should raise an idea considered
early in the design.  Take a PC, remove the motherboard and cards.
Build a new PCB to a standard motherboard from factor with a bunch of FPGAs on it.
Each peripheral is built on a PCI form factor card.  Plug the whole lot together
and power it from the standard PC power supply.  The result is a box that looks like
a PC, but is an FPGA test system.  (Put enough FPGAs in it, and you could program it
to be a PC!)   Mechanically, this would be a rather nice system, but probably more
expensive than a SIMM based system.
 
Regards
John