Re: ZIPGS 2-pin Crystal

[mjmahon@aol.com (Michael J. Mahon)]

Wayne Stewart replied:

>Andrew Webber wrote:
> 
>> Thanks for that. I actually have 2 ZIPGS so I decided to physically
>> take out the other bank of 32K to use in my 2nd card (only the first
>> two slots are filled). When I take the 2nd bank the card is not stable
>> at all and will not even boot GS/OS. So putting 64K cache in and
>> setting to 32K makes the system stable. Any ideas why that is the
>> case?
>
>The main theory revolved around the second bank being a lot further 
>from the processor than the first bank and with those skinny leads.

The timing differential attributable to the longer leads is less than
0.1 ns., so unless your SRAMs are _really_ marginal (unlikely,
since narrow ones are usually _very_ fast), timing should not
be an issue.  This level of timing variation is less than the usual
differences between similarly spec'd chips.  More likely, the
additional loading of another chip degrades some critical edge.

Inadequate decoupling near the sockets could be an issue, but
I don't recall the exact configuration of the decoupling caps.

It could also be that there is another timing issue in the
custom logic which comes into play with the larger cache.

>The cheap sockets Zip used didn't help either. I originally did my 
>upgrading by soldering a half socket in the middle of the original 
>socket. Later I started wondering about the contact so I removed the 
>half sockets and replaced them with machined pin sockets. The two
>ZipGS I tried it on were previously stable at 14.5 mhz but not 15.
>After I was able to run them at 15mhz. Not spectacular but maybe 
>indicative of something.

While poor contacts with sockets is always a problem, even
inexpensive sockets usually make excellent contact with ICs
unless they or the IC leads are dirty or physically damaged.
All that is required is clean, springy contacts.

Machined pin sockets are the most expensive sockets, with
tightly controlled tolerances and better wear characteristics,
but that factor should not be decisive for normal applications.
They are definitely preferred if chips will be changed frequently.
(Of course, if they're changed frequently enough, then a ZIF
socket is a good idea--but they have poorer high frequency
characteristics than a simpler socket.)

My conclusion is that something else changed to make your
card more stable.  Like swapping two nominally identical chips
which are different in actual speed, or getting a less "noisy"
contact between a chip and a socket which was either damaged
or had some dirt in it.

When we push the speed of a digital system to the point where
it begins to fail, its behavior becomes more like the analog
system it actually is, rather than the digital abstraction we like
to think it is.  Small power supply variations, temperature effects,
induced noise from adjacent electronics, etc., will all conspire
to give Murphy a field day.

This is why "best practices" call for "schmooing" a circuit by
finding the limits of correct operation in several dimensions:
frequency, voltage(s), and temperature, then operating it
in such a way that the normal range of variations in those
parameters will never get too near the conditions under which
it becomes unreliable.

This also accounts for the usual ability to "push" a circuit
beyond its "nominal" speed, and sometimes the "schmoo"
plot is sufficiently wide that reliablilty is not compromised.
But this is not always the case--and hobbyists are rarely
willing to go to the trouble of characterizing the operation
of a modified device to determing how close they are to 
its limits.

Good advice, then, is to try pushing the speed until you find
the fastest point where it is still reliable, then back off 10%,
to allow for variations in the parameters you are not controlling.


-michael

Check out 8-bit Apple sound that will amaze you on my
Home page:  http://members.aol.com/MJMahon/