Re: MISC-d Digest V97 #5

[jfox@xxxxxxxx (Jeff Fox)]

Dear MISC readers:

Penio Penev <penev@venezia> asked:

>If memory serves me right, F21 is 3/4 empty and is about 15K transistors. 
>that means that another 45K are potentially sitting there unused. How many
>Ks of SRAM does that make?  If Chuck's OKAD is about 8K, I believe that
>almost anything useful can be that big and fit on chip. 
>
>Is there a chance that F21c will include experimantal on-chip RAM, albeit
>in small amounts?
>
>In other words, where is Chuck's bottleneck -- design, testing, writing
>software, developing OKAD, or just the lack of $10/fab run?

As Chuck has said before he is not good at designing ram and it is very
expensive to put it on a cpu.  This is not just true for Chuck, but in
general rams and cpu are made with different fab processes.  Most modern
rams have more process layers than cpu and some odd things like vertical
transistors.  Chuck likes to point out that the cache memory on most
cpu takes up a lot a chip real estate on very expensive chips.  If you
factor in the percent of chip devoted to ram you find that ram can add
$1000 to the price of a high end cpu chip!  As Chuck says it is expensive
to put ram on a cpu, people do it anyway but at a price.

F21 has 18 data stack registers, 17 return stack registers, 2 control
registers for video, 4 control registers for network, 1 analog control
register, 1 memory access control register, 1 parallel port control
register and a few internal control registers.  The 35 stack registers
take up about 50% of the die that is used on the chip.  And based on
that size it might be possible to add that many registers decoded
as on chip sram locations.  However at the present time these chips
produce the same timing for access to these registers as they do to
high speed sram.  So just putting them on the chip doesn't give you
anything unless you provide a faster path also to on chip memory.

Registers accessed with cpu instructions take one cycle.  (A A! PUSH POP
OVER DUP) Registers access by using and decoding an address take two
cycles at the moment, and whatever time is programmed in for high
speed sram access.

I would estimate that the design time, and the time for additional fab
runs to debug a chip with on board ram would add between six months to
a year to development time.  F21 design was paid for at a lower rate than
what Chuck would charge now, so a substantial redesign to F21 would be
expensive at this time.

Bottlenecks?  Well design time, 2 month fab delay, about 1 month typical
testing, and redesign means there is a 3 or 4 month delay in the development
process.  iTV has submitted chips each month for some time and this means
that sometimes two or three chips are made with the same bugs before one
gets through the fab and testing process to reveal a bug now in a series
of chips being fabed and tested.  A fast fab run would help, but in fact
several times those people in the fab line must wait an extra month because
of technical problems (or a bigger fish being allowed to cut into the
line I sometimes suspect).  This has led to many large companies who
are developing chips to own their own fab facilities.  OKAD and software
are not bottlenecks.  You can schedule express fab runs and get them
through a little faster, but it makes the prototype chips very expensive.

Since the beginning the biggest problem has always been that this is
an interrative experimental process.  This is true because there is just
no place for Chuck to find out the stuff he needs to know short of doing
experiments.  For the first few years Chuck spent a lot of time seeking
out information and asking questions.  Typically the people he was talking
to would say "I can tell you anything about this 2% of this chip, but I
know almost nothing about the rest of it."  Or people would say "I don't
know that, I don't need to know that, I just assume that the tools I
use ($100k - $400k cad programs) will know how that works and do all
of the calculations properly.

Perhaps when Chuck has connections to some big corporate partner he
will be able to drop his cpu into the corner of a chip with lots of
modern ram on it.  This should happen sometime.

Also Chuck has done on chip ROM and will experiment with on chip ROM
and on chip RAM.  But I would expect the first MISC chips with on chip
memory to contain 1K words or (perhaps much) less.

Eugene Leitl <Eugene.Leitl@lrz.uni-muenchen.de> stated:

>The MuP21 street price was several $10.

True.  I forget how many chips Dr. Ting had manufactured or packaged, but
even without factoring in any of the costs of several years of chip
development Dr. Ting is selling very close to his cost.

As I have said before you start out a $300 per chip and the more you put
in the cheaper they can be.  Dr. Ting morgaged his father's house to
manufacture a few K chips.  Even without factoring in development costs
he may be losing money on each chip. (but he doesn't lose as much money
as he does on the ones he doesn't sell at all.)

>An SRAM cell takes 4-6 transistors. Since unused silicon is oblivious to 
>random defect hits but SRAM cells are not, yield will go down, and 
>prices up. 

True.

Lonnie Reed <lonnie@netcom.com> asked:

>I was wondering if anyone has plans to build a java virtual machine, on 
>the mup21, or more likely the i32 processor platform. it seems that if
>it could run java applications, on a settop box (iTV), that it would have 
>a ready-made software source. and would enable people who do not own 
>computers to have access to current software, on an inexpensive, though 
>powerful system. I understand that some apps would not be suitable since 
>television resolution is limited, but these seem to be surmountable.
>maybe packing 2 pixels in 'vga 640x480' resolution to display at 320x240 or
>320x420 on a tv. this could be in addition to its native software.
>
>lonnie

Yes, it is one of the projects at the iTV Corporation.  The first settop
box using i21 will not contain Java, but a Java version is planned.

Jeff Fox
    Jeff Fox
    jfox@dnai.com              Ultra Technology Inc.
    jeff@itvcorp.com           the iTV Corporation
    http://www.dnai.com/~jfox/