home .. forth .. misc mail list archive ..

No Subject



>Hiya co-MISCers,
>
>excuse me if my basic assumptions are invalid, but
>is there a basical (that means technical, not financial)
>difficulty to integrate a MISC core, particularly a ultra-wide
>data bus one as a integral part of (slightly modified)
>standard DRAM die (e.g. a 1-4MBit one)?

>-- eugene

No technical reasons other than Chuck does not have access to
the RAM design, the extra layers and smaller geometry needed for
RAM, and the RAM designers don't have access to Chuck's designs.

Chuck is well aware of the advantages of on chip ram, and direct
access to arrays of bits.

The problem is just economic.  No one is paying for this.  Chuck
can't afford to prototype in the tiny die.  There are very few
people who have been willing/able to pay for even this low end
development.  Given the size of a tiny die, and fact that wafer
scale may be 3000 times larger, not to mention needing more layers
and smaller geometries for memory, we are talking big bucks.
If Chuck can't find anyone to pay for $3000 prototype runs it
may be a  bigger problem finding someone to pay for $10,000,000
prototype runs.

You will notice that even the big chip makers in this country have
all but given up competing in the memory market.

Sure wafer scale .35 or .2 micron could produce really sci fi devices.
If Chuck had access to this you would see a bunch of multigiga op
processors each with megabytes  of intregrated  memory  and gigabyte
per second interprocessor links on a wafer.

But although it is nice to think about what we could do if someone
provided a 1,000,000% increase in the development budget it doesn't
really help with what in reality will be a slow step by step
process mostly constrained by lack of funding.

Chip makers have acknowledged that they could incorporate one of Chuck's
processors in the corner of their parts, but they have said they no
inerest in this at all.

It would be a little like trying to convice gasoline companies to give
away 1000 mpg cars when you buy some gasoline.  They would prefer that
you need more gasoline, not less.  They don't want to support cars that
will need less gasoline.  They are certainly not going to invest large
amounts of money on things that might in the long run reduce the demand
for their main product.

It is easy to convince me that it would be nice to have 100 1 billion
instruction per second processors and several hundred megabytes of mem
on one piece of silicon.  It just seems like too big of a jump to
ever happen.  At the rate Chuck is going he will prototype a couple
one hundreths of a square inch of silicon each month at best.  Suggesting
a move to wafer scale is like suggesting that Chuck should just go
out and hire and train 1000 people to help him with vlsi layout. It
is a great idea unless you don't have a way to actually do it.

Jeff Fox