Re: FPGA

>G'day Clay (and folks),
>
>A while back I looked at putting the whole Cinematronics board onto a
>single FPGA. However two obstacles that I ran into were the RAM and the
>I/O pins. In your letter down below, Clay, you said that SRAM chewed up
>alot of the FPGA. Does that hold for dynamic RAM like the 256 words by
>12 bytes of RAM on the Cinematronics board?

(I'm no FPGA guru, so I'll give you the 50,000 feet explanation as I
understand it.)

In an SRAM, each group of gates/cells can effectively store a bit of
information (the chip die is layed out to access the bits in what's
effectively a big matrix). This gives pretty good memory density since
it's specifically designed for this memory matrix.

DRAM's use a simpler (smaller) group of gates/cells to store each bit, this
gives better bits/sq inch, but requires a periodic "refresh" to keep the
bit values.

FPGA's implement a memory cell as a latch. So an eight bit byte takes 8
latches, each of which are built up from a large number of gates. Add in
the sequencer and muxes and "SRAM" virtualized in the FPGA starts consuming
a LOT of the usable gates in the array. Essentially the technology of the
basic "gate array" is a poor choice for high-density memory...

That having been said, 256x12 isn't *that* much RAM. I was thinking like
8K or something. So you're talking 3072 latches plus muxes and addressing
stuff. That still looks like a quite a bit. :-/ I dunno how much "real"
resources that would eat in an FPGA. If I get a chance I'll look around a
little and see if there isn't some examples of RAM in FPGA's that might
give a clue to how much it really eats up...

>ps - These PIC intrigue me....how do they compare to FPGA for quick/easy
>implementation?

The PIC's just a microcontroller for all intents and purposes. (Actually
it seems a LOT like a PDP-8 ;-) They're cheap (like $2.50 for a 16C54),
pretty quick (20MHz clock = 5 MIPS), and have a very nice (free)
development environment called MP-LAB that has an assembler and simulator
and whatnot.

A company called Scenix Semiconductor has a PIC-clone, kinda like Dallas
did with the 8051. They execute one instruction per clock at 50MHz. Lots
of neato features-- flash programmability, onboard 4MHz clock, etc.

They're all pretty fast, but still not as fast as an async combinatorial
device like a 22V10 or something. They're competing for slowing
state-machine type applications pretty well though...

-Clay

Clayton N. Cowgill Engineering Manager
_______________________________________________________________________
/\ Diamond Multimedia Systems, Inc. clay@supra.com
\/ Communications Division http://www.supra.com/
Received on Mon Oct 6 16:02:38 1997