idea that custom solutions are more efficient
Dear MISC readers:
John Griessen wrote:
> [JG] The argument that a little custom can go a long way fits with the
> idea of using FPGA's to start designs Any one else besides Mike Simon
> using FPGA's to make custom processors/glue logic?.
Yes. Because the designs are small they are easy to do in VLSI or
FPGA (relatively speaking). Because the designs are small they
leave room for custom I/O hardware to get high efficiency specialed
solutions. Obviously the costs are much lower in FPGA and it is
more appropriate for low to mid volume. In high volume custom
is cheaper. There is a sort of gap in between where nothing is
really cheap.
But sure. If you take the requirements and add some custom hardware
to help do the job the processing may be reduced to the level that
a 1 instruction MOVE architecture was sufficient to get high
performance. The tradeoff of having 27 instructions lets us
higher performance from the CPU than just a move instruction
only machine but we don't need 1000x as much hardware in the
CPU to do many things if a few transistors can do it more easily.
An FPG based Forth engine may be plenty fast enough with
efficient software and a little bit of hardware to do a job.
People are used to hundreds of mips not being enough to do
word processing because of the inefficiency of software. With
a custom CPU, custom hardware, and custom software you can do
a lot without very much. People don't really appreciate that
you really don't need those hundreds of mips if the effiency
of the software isn't hanging around 1% and you don't need
to pay for 99% of the hardware in general purpose designs
to solve many problems.
Mike is one of the people working with FPGA and he has been
visible in the list. iTV has also been doing processors and
custom I/O stuff in FPGA as well as VLSI. Dr. Ting has been
doing it using schematics and is doing a presentation this
Saturday on doing it in VHDL. John Rible has taught this
stuff in a course through a UCSC extension program.
> [JG] What about plastic transistors for MISC projects? I've started
> researching that
> and read references to grad students printing transistor circuits
> with ink jet printers, only the speeds are not going to be high like
> micron featured silicon circuits or the next great thing.
> Anyone have some good leads
> for finding out more about plastic transistors as low speed
> computing switches
> instead of use as LEDs?
I love the idea. Where do I get the kit to print custom
chips and circuit boards on my laser printer? Last time I
talked to Chuck we talked about this. He also was talking
about how he added circuit board design stuff to OKAD. Now
if we could just print the circuit boards out on the desktop!
I have a laser printer kit with some colored sheets that
you run through your printer with something you already
printed. It bonds colored stuff, or silver or gold, to the
black areas on the paper when it goes through the printer.
I doubt if the stuff is conductive but it sure does look
like you could print out something that really looked like
a circuit board on your laser printer.
> [From J Fox] I have thought
> of it as ten years of research and education. I wanted to be in
> a position to do something similar in a technology a million times
> faster than silicon when it becomes available. (around 2000 ;-)
>
> [JG] Also, what is this a reference to, Jeff?
> Do you want to make custom light wave calculation engines with a few
> instruction inputs togglable by "slow/multi purpose" electronic
> controllers like the F21?
No exactly. Ultra high speed optical computing technology needs to
connect to electronic technology somehow. With this kind of
technology you might have one optical fiber coming into a
router and the ability to deliver so much data that you would need
thousands or millions of PCs with gigabit network cards to deal with.
You could do the same job, more or less for 1000x less with something
like F21s instead of PCs. If you need thousands of or millions of
gigabit speed electronc devices to hang off of cheap optical devices
then they had better be cheap electronic devices or you had better
only be thinking of the high end market. There are folks who recognize
that everyone would like 10^9 mbps data streams running in and out
of their houses and wireless devices if they are cheap.
One of the P32 designs had no video but multiple gigabit
network coprocessors designed to be connected to fibers so
that one could make a high speed router for a few bucks
instead of thousands and make stringing fiber much cheaper.
There are other people interested in similar things of
course.
Of course using a MISC processor as a cheap communication
processor hanging off an optical router or fiber is only
a step toward moving the MISC chip into the optical
technolog along with memory so that it would look very
much like an F21 system with integrated memory except
that switching times would be 1,000,000 times faster
than silicon. Don't cheap 1 billion Forth Mip chips
sound nice? No that was not a misprint or too many
zeros by mistake. That is as a billion million times the
number of nodes of course. ;-)
People who only have exposure to the products that get
writeups in the popular magazines sometimes have a very
hard accepting the technologies that people have out
there that are not commercially available to everyone
at the moment. People often assume that if there was
fantastic stuff out that there that because it technically
superior it would be dominating the marketplace. But
the reality is that there are 1000 reasons whey companies
and products succeed or fail and whether the technology
is good doesn't really seem to be a big one.
One of the fun things about this project has been getting
to see some amazing stuff that I never thought I could
get clearance to see. And the people experimenting with
things like diamond semiconductor or optical computing
are not so locked into the conventional ways of seeing
computing that PC consumers seem to be. They seem to
have a much better grasp of the concept of what we
have been doing and are fun to deal with. I mean it
really isn't much fun dealing with people who are only
interested in new things that let them do everything
exactly the way they do it now. If they really must
have exactly the same software that they have now then
they are not going to find what they are looking for
in MISC or other specialized technologies.
Jeff Fox
