🚀 go-pugleaf

In article <975@csustan.UUCP>, dave@csustan.UUCP (david j wells) writes:
> 
> Also the simpler silicon is much easier to develop.  Less time.  Less money.
> Fewer bugs.  (kind of like the diffference between coding a linked list v.
> coding a B-tree  ...)

That's for sure!  When I was working for MIPS Computer Systems on the R2000
chip, our very 1st silicon was almost 100% functional.  There were a couple
of layout problems that lead to an unexpected diode-like device in one area,
which resulted in a multiplier that wouldn't.

This was fortunate, because we didn't get packaged parts back from the fab
outfit until 12/23/85, and we had a contractual committment to deliver an
alpha prototype to an early customer.  We spent a day or so integrating the
chip into the board, and found a couple logic and timing bugs with it (the
board, not the chip).  The machine executed it's 1st instructions in the
wee hours of Christmas Day, and a new computer was born.

There was sufficient functionallity to demonstrate "significant technical
progress", as the contract required, and that customer sent us our first
revenues ... a check for $1.5 million!

Total time to R&D the R2000 was in the neighborhood of 9 months, not
counting the architectural work done at Stanford that pretty much defined
the MIPS architecture.

In case you're wondering, that 1st "program" was a series of instruction
test cases that indicated their progress by writing to a memory location
that was in reality a set of 8 leds.  Helluva lot to pay for a binary led
counter ... :-)!


> The complexity is moved into software (generally the compiler) and is
> therfore relatively trivial to modify (compared to fixing silicon).
> Can you say 386?  :-)

Since that 1st chip spin, a few more layout/fabbing problems were discovered,
and a small number of logic design bugs were found, so we ended up with a
couple more chip spins before we were satisfied with the part.  But that's
amazingly few for a ~100K gate chip!

BTW, several of the logic bugs we eventually did find were easy to get around
by having the compiler generate some "kluge" code ... no huhu, cobber!

/kim


-- 
UUCP:  kim@amdahl.amdahl.com
  or:  {sun,decwrl,hplabs,pyramid,ihnp4,uunet,oliveb,cbosgd,ames}!amdahl!kim
DDD:   408-746-8462
USPS:  Amdahl Corp.  M/S 249,  1250 E. Arques Av,  Sunnyvale, CA 94086
CIS:   76535,25