Matt Shaver wrote:
> It's possible that a "universal cnc peripheral chip" could be designed
> and made this way:
> but I expect the cost benefit analysis would still favor FPGAs unless
> considerable weight were given to libre philosophical requirements.
No, much, MUCH harder than you think. IC design software is INSANELY
expensive, $250 K per seat per year in some cases. And, the educational
licenses specifically state nobody can benefit from the chips made
through the use of these tools. We have had a few chips made through
MOSIS using Mentor and Cadence tools, and since it is all for government
grant and university research, it passes muster. I even build the
boards commercially after the chips are supplied from one university to
another (my customer).
An all-digital chip is a lot easier to simulate and be sure it will
work, but the number of potential pitfalls are enormous. And, still,
even making them in small quantity, the cost is quite high. We pay
something like $200 PER CHIP for a fairly large, mostly analog signal
processing chip. (Ummm, we are in the seventh generation of one of
them, and it STILL has problems - that's my job security, taking flawed
chips and coming up with the external hacks to make them work.) The guy
who does the IC design had one chip that was missing one via between
layers, but it was connecting the main digital clock from the lead
bonding pad to the rest of the works, so the entire chip was totally DOA.
Anyway, even a quite small ASIC through MOSIS will run about $10K
installed in a package, for the first FORTY chips! If those work, you
can then get more off the same wafer for another couple grand. MOSIS
will NEVER be competitive with commercial chips unless you buy them by
the thousands, and then you can just contract directly with one of the
What I'd really like to see work is an RT_USB driver to the Cypress
FX2LP chip and an FPGA after that. I'm still not too clear on the
limitations of the USB, though, so maybe ethernet is a better choice.
But, I've been fooling around (yes, that REALLY is the right term, I
have NO IDEA what I'm doing yet) with the Cypress chip at work, on a
control project that doesn't need real time or any great speed. But, I
have a little demo on my desk that reads 39 MB/sec and writes 31 MB/sec
over USB. QUITE impressive. The FX2LP chip handles the actual
transfers entirely in hardware, so it is as fast as the USB high-speed
can go. I don't know what the latency of sending a small request out
and getting a small reply back would be, or how you might arrange to
force a small packet to be sent and prevent it from buffering them up.
But, it looks fairly promising.