Path: utzoo!attcan!uunet!husc6!bloom-beacon!mit-eddie!uw-beaver!uw-june!uw-entropy!dataio!pilchuck!ssc!markz
From: markz@ssc.UUCP (Mark Zenier)
Newsgroups: comp.misc
Subject: Re: Mythical microprocessors
Summary: Not Good Enough
Message-ID: <1404@ssc.UUCP>
Date: 8 Aug 88 18:01:39 GMT
References: <677@buengc.BU.EDU> <79700005@p.cs.uiuc.edu> <11419@oberon.USC.EDU>
Organization: SSC, Inc., Seattle, WA
Lines: 43

In article <11419@oberon.USC.EDU>, mlinar@eve.usc.edu (Mitch Mlinar) writes:
> In article <79700005@p.cs.uiuc.edu> gillies@p.cs.uiuc.edu writes:
> >
> >I heard that the (oriental) designer of the Z80 and Z8000 did most of
> >the design in his head.  That is, he understood almost the entire CPU
> >and its layout in VLSI.  I think he left from Zilog before the Z80,000
> >was complete.  Had he stayed (to influence the Z80,000 design), the
> >chip might have been more successful.
> >
| 
| The name of the designer escapes me right now, but I believe he was from
| India.  Also, he laid out the Z80 and Z8000 primarily using RANDOM LOGIC.  He
| was *very* good at it, but also realized that regular structures like PLAs,

Not good enough,  I wasted a week chasing down the fact that some early
revisions of the z8002 would confuse the user stack pointer register with
the system stack pointer register if the timing clock was within a 1 nanosecond
window, when the moon was in the right phase, and the temperature was just 
wrong.  Just what you need in an embedded controller.

| etc.  would be required for the next generation.  The Z8000 took a *lot* of
| man-months in comparison to other CPUs of that time due to the random logic,
| but was faster because of it.
| 
| To this day, I know of no synthesis tools for CPUs which can handle timing
| generators (the method used in the Z80 and Z8000).  Current chips are all
| micro/nanocontrollers and/or PLA table driven.  Timing generators ARE faster
| for a comparable size design, but the complexity versus speed gain is just
| not worth it.  Then, there is the issue of verification.  Ugh!
| 
| -Mitch


They (in '84) only got the z8000s up to 12 Mhz, while the 68000 got up to
16.  Even though the z8000s only needed 3 cycles per memory reference compared
to the 68000's 4 cycles, their slow silicon didn't cut it.

Another problem with the z8000s was it took two years from first announcement
until it was real.  The 68000 was announced a year later, but they both were
available at the same time.

Mark Zenier	uunet!pilchuck!ssc!markz		
LOGGERMIST CROTEHAVEN