[net.micro.68k] Speedy 68020

gnu@l5.uucp (John Gilmore) (10/28/85)

It's certainly possible that a single 68020 in a lab can be pushed to
32MHz.  It helps a lot, as was pointed out by someone from Intel in
another newsgroup, to change the voltage and to cool the chip as much
as possible.  Getting a sample running 32MHz with freon spray on it is
different from being able to produce chips that run that way across the
full temperature and voltage range.  But if the rumor is true, it does
indicate that the chip is limited by power or temperature or
fabrication tolerances (mechanical design), rather than by switching
speed or electronic design.

One technique usable to see what parts of the chip are limiting it
is to run it at various clock rates and see what kind of failures
appear.  This may be where this 32MHz 68020 is being used.  For
example, the kind of errors that have been described on Vax 785's
(carry propagation across 30 bits) might crop up, indicating that
the adder is the next bottleneck.  Chips run this way are not really
expected to function; they're expected to point the way by how they fail.
The next time the chip mask is fixed up, they can improve the adder,
which probably improves the yield of 16.67MHz chips and 12MHz chips too.

norm@rocksanne.UUCP (11/01/85)

> full temperature and voltage range.  But if the rumor is true, it does
> indicate that the chip is limited by power or temperature or
> fabrication tolerances (mechanical design), rather than by switching
> speed or electronic design.

	I am not clear on what you are saying here, the speed of the chips is
	really limited by switching speed.  Improvements in the fab lines
	or finer geometries will help improve the switching speed by
	either reducing loads or improving drive currents.  
	
	The temperature issue is something common to all silicon MOS devices.
	In a simple sense, the speed is limited by how fast you can 
	drive rc loads.  The drain current for a MOS device has a term
	in it that decreases as temperature increases.  Some of the
	data I have seen from MOS devices that we have done, and SPICE
	simulations indicate that for a rise from 27C to 50C the
	speed of MOS devices will decrease about 10%.  Cooling the devices
	will increase the drain current and hence the switching speed.
	
	There is actually a company (a CDC or Cray spin off ?) that is
	planning to market a "supercomputer" that uses CMOS gate arrays
	to build a processor and runs the devices at 70K.  They claim a
	2x-3x speed improvement over room temp (this info is a little
	sketchy in my memory but most of the data should be close).
	Given that the 020 runs at some where between 16-20Mhz then
	32-60 Mhz chips are possible with this technique.  However, the
	implementation details of such a system interconnect etc
	are not trivial.  Also the temperature decrease that helps
	in MOS devices makes bipolar devices run slower, so the
	interface circuits, if bipolar would not want to be cooled.



--------------
Norm Zeck
Xerox Corp.
800 Phillips Rd. 128-29E
Webster N. Y., 14580
ARPA:		NZeck.Wbst@Xerox.arpa
UUCP:		{decvax, allegra, seismo, cmcl2}!rochester!rocksanne!norm
Phone:		[USA] (716) 422 6246 
--------------