my opinion on some MISC mail list topics for the record

[jfox@xxxxxxxxxxxxxxxxxxx (Jeff Fox)]

------------------
Dear MISC readers:

There have been a few posts about MISC being or not being suited to
specific embedded applications or even a more general purpose MISC PC.  
I have wanted to put the discussion in the context of the way that I see it.

Chuck's approach was to first get his cad system working and then
produce lots of custom designs really fast since he is incredibly
productive with his tools.  Of course he discovered that the people
who use the expensive tools don't know how they work.  The people
who wrote them only know about a little bit of them.  They also
don't know anything about what the fabs actually do.  There was no one
who could answer Chuck's questions so he just had to use trial and
error and cleverness.  It did take a lot longer than any of us expected.
It has even been a couple of years since I insisted that he needed thermal
modeling in OKAD.

He has said that converting F21 to F32 is about a one day job, he has
done it before.  He doesn't lie about this stuff.   He wants to have 
a large menu of chip components and say that you can pick them from the
list and he can tinker it together in a few days.  Most of the time he
would drive me crazy by telling me that he was starting the layout changes
to the next itteration of the F21 design the day before submitting it to
fab, but that is the truth of the how fast he works.  He told me that
he found a fatal flaw in F21a a couple of hours after we submitted it
for the fab run.

The real factors are things like that adding one more pin to the F21 design
would have cost me $40,000 in prototype fab costs alone.  I have had
hundreds of pieces of mail telling me that I should just have added
a hundred pins or bought up the so and so fab line myself so that I
would not have to wait in line for prototype fabs.

Meanwhile we would submit a design, wait three months to get it back,
then begin testing.  Sometimes we would be forced to not submit a new
design for testing for six months or a year because no one would pay 
for it.  When chips were submitted every month the bugs would be in
the next three sets of fabs before you would even know it existed.
No funding almost stopped us a few times and certainly slowed things
down.

There were years when the only test equipment was an osciloscope so
infering exactly what is really going on inside of the chips and at
the foundry where they fab them is another one of those tasks that
most people could not comprehend.  But we never let things like no
money or no equipment or no support stop us.

If you have money (or own a fab) you can get the prototypes back much
faster.  But the point is that Chuck's idea years ago was that if
someone would fund his work he could get to where you look at the
menu, talk to him, and a few days later you get a copy of OKAD with
your chip all laid out ready for testing in the simulator and 
production if you are satisfied.  He has done that now although the
lack of funding did delay the tuning the CAD tools.

If no one ever pays for any chips none will ever be made.  None of
us own our own fabs or can make chips without some funding.  I feel
very very good about proving all of the stuff that we claimed for
years while so many people were calling us liars and snake oil
salesmen or cultists but I never said that we definately would find a person
or company with both interest and money to actualy produce product.
I got to do it and it was fun.  It would be fun to do production too
but just to prove that we were right is plenty for me.

It is great to think what you could do with a tiny bit of money.  I
mean I got F21 with almost none compared to any other chip or computer
designs out there.  If one didn't have to pay for the ten years of
research and could only pay for the development as Chuck intended
it would be quite nice.

>>Where they really shine is being able to integrate peripherals
>>on the chip.
>
>Problem is that pin-count and surface size determine real mass market
>costs. And that battery life-time is very important.

That is not the problem that is the whole idea of MISC.  Look at P8!
It has six pins and $.01 worth of silicon.  It was designed to be given
away and run forever on a hearing aid battery.  It had a custom I/O
coprocessor designed exactly for the job.  The die was so small that
you could hardly see it.  Too bad that no one wanted to pay for it.

I would like to add some RAM and ROM to F21.  I would like to keep the
silicon cost down at $1 and just have the I/O and power pins like P8.
I would like it to be the size of a printed character and be manufactured
for $1  If it doesn't happen it will be for the same reason that I have
fasted for half the time the last few months.  It is not a technical issue!

If I had a tiny bit of money I could do that, then I would have 200mip $1
nodes that could be packed really close together.  Wearable, or in the 
ear supercompters would not be just blue sky ideas if we had 1% of the
funding that people normally throw at problems.  I think we have done 
a lot with minimal and spotty funding but hundreds of people have told
me that we will fail because I haven't done enough.

Still I am very excited about having 25 F21, even if they do require
external memory.  They may be far away from what we could do but they
are also far away from most everything else too.

It would be nice to make one that had more pins.  I mean yes they cost
more than silicon, about a penny each, but they are good for some things.
If you don't put memory on chip then more pins is more memory bandwidth
and we have always acknowledged that we were trying to cleverly minimize
the pin count on the first prototype chips.  I am glad that I didn't have
to pay twice as much for another pin but yes 500 mips looks nicer than a
200 mips throughput.

Then general rule is twice as fast for 20% smaller transistors.
I would like to switch to .35 or .15 or something where stuff is so
small that you could hardly find our chips and no one would believe how
fast they would be.  They would cost nothing to make.  

We have been tantilized from time to time when we did get to do 
things like go to Allied Semiconductor and look at the
chips while they were running with a scanning electron microscope or
make plans to port to diamond instead of silicon.  It is fun to get
that close to funded projects if not a little frustating.  I also
hadn't been out in weather that cold for years.  It was really a fun
trip because I got to spend a lot of time talking to Chuck about
things other than Forth and chips.

I was first interested in Forth hardware when I compared Novix to 386
and thought about such things as single electron transistors and
quantum tunneling transistors and optical transistors and mechanical
nano-switches and the problems of developing computers for those
technologies.  I wouldn't want to try to port Pentium to one of those
technologies, but I always thought of MISC as moving in that direction.
In fact Chuck has pushed ordinary silicon transistors way beyond what
anyone expected.  He methods and approach is also perfectly suited
to experimenting in those sorts of new technologies.

>Why not integrate the RAM and ROM? RAM is only 128 to 256 bytes
>for most applications. ROM size is much more essential. I once had to do
>a project in the 4Kbyte of a 6811. Was really a shuffle of which functions
>to include in the test-versions, because during testing a lot of extra code is
>often added temporarily.
>These things make software engineers cost a lot of money.
>
>>>PS. Why aren't we discussing this on the list?

When we take advantage of Forth and the basic operation of the I/O
coprocessors we do things like throw in a GUI that takes a few K
instead of a few megabytes.  By making it 1000 times smaller and simpler
it does a lot of things 1000 times faster.  If we use the real power
in the coprocessors we sometimes see another 1000x speedup.  That is
also much of philosphy behind Chuck's low fat computing approach.  It
is more than just using a minimal instruction set CPU.

I have learned to not hit the mouse button to often on my Pentium.
I have learned that with Windows 95 and a 120mhz Pentium the system
will just appear dead a lot of the time when it is doing stuff
in the Wintel way.  I have learned that task switches happen and
can take as long as twenty seconds on my machine.  If you wait long
enough you can start typing or using the mouse again. I prefer systems 
with 20ns task switching myself.

If you can add a windows accelerator with $.001 worth of silicon and
speed things up 1000x why not?  If you can get application specific
custom I/O coprocessors that can do things like gigabit I/O with no
system overhead on $.001 worth of silicon and $.02 worth of pins why not? 

Things like video and Radar and optical fibers look fast to most computers but
they look slow on the inside of our chips, even in .8u technology!  If
we had a .15 micron, or diamond instead of silicon it really gets scarry.

>Seems like the default is 'answer to the poster' and not 'to the list'.
>It requires special attention in most email readers to get the thing right.
>Can't this be changed mailing list manager?

Along with 30 or 40 other jobs I used to do it.  I don't know who is doing
that job at the moment.  The list was down for a while but it has been
pretty active lately.

>Groeten/Greetings,
>Jaap
>
>-- Chip Directory
>-- http://www.chipdir.com/chipdir/
>-- And about 30 mirror sites world-wide.

I think chipdir was the first site to consider MISC to be postable 
information.  For many years I argued with people in email that P21
was real.  The argument against us was basically that it would not be
considered a computer unless we provide them with a copy of Unix
and Specmarks.  I wondered why they are willing to post information
about other chips that were not designed to run Unix but wanted
us to prove that P21 was designed for Unix workstations before sites
would post any information about us on the web.  I was very pleased
to see that we were noticed at chipdir.

Jeff Fox
Ultra Technology
----------------