multichip modules

[jfox@xxxxxxxxxx (Jeff Fox)]

Dear Misc-readers,

Multichip modules are great.  If I had funding I would be planning
on a multichip module package in addition to a simple ic version.

I have looked into the economic and technical issues involved in
mounting F21 in a multichip module.

The advantage is that it could be very small.  Like a microcontroller
with on chip ram and rom you don't need external memory, so you only
need a few pin to the real world.  Power, analog i/o, digital i/o,
network i/o, xtals.  The entire system can be reduced to a fraction
of a square inch.

The disadvantage is simply economic.  Like volume production of dies
there is a set up charge.  Somewhere between 10k and 100k it starts
to look economically attractive.  Below that the setup charge is
not spread accross enough units.  Also the cost is MUCH MUCH more
to have volume produced.  You can have 100k F21 manufactured for
only $100k because the die are only $.20 and the pins $.80.  But
on a multichip module you have expensive memories on each module.
If you could find the memories and get a good price it is a much
bigger problem when each unit costs $100 to manufacture.  The
up front cost is 100 times higher, and the profit margin is
much much lower.  The $50,000 up front charge to be a position
to do volume manufacture seems quite trivial if you are making
100,000 $100 units.  But we are talking several orders of magnitude
more investment dollars than are needed for making product without
memory, and a potentially much much lower profit margin.

If you spend $10 each to make CPU (total costs) and sell them for
$20 you are gambling $100k on whether you can sell between 6k and
10k chips.  If the multichip modules to get the costs down
requires a much bigger investment.  To bring the manufacture cost
down to $100 you make to invest $10,000,000 and you are gambling
much more on the return of a much smaller profit margin

Similar to the wafer scale problem, but not as large.

Unfortunately funds keep getting smaller and smaller, not larger
and larger.  I see little indication that investors would prefer
to risk 100 times as much money.

Well I have overstated the problem.  I was told that it would be
easier to do a multichip module with the 256kx4 drams not the
1mx4 drams, this would mean $20 for memory not $100.  Still
even in low volume, and small memory multichip modules we
are talking about a lot more money than anyone has offered to
invest.

It is fun to think about what we could do with funding.   But fi
you can't afford a handful of tiny die, the prospect of 
multichip modules, wafers, exotic state of the art technologies
etc is just wishful thinking.

Let me know if you know of anyone who actually wants to invest in
some of these ideas.  The problems are not technical.

Jeff Fox