At 9:01 Uhr +0100 01.09.2010, Andrew Jury wrote:
> This is good progress, but I wonder why lifting IC6 kills the issue dead?
I don't know.
My preliminary hypothesis is as follows:
It's a capacitance issue. The stray capacitance between XTAL1 and XTAL2
(Pin 2 and 3) is different. Of course we're dealing with very small
capacitance here (C22 and C23 are 10pF and 22pF originally). But looking at
the datasheet for the 8048 it appears to me that those two caps are
specifically tailored to the stray capacitance of the board.
The replacement KLM-367A has a very slight difference in routing for the
traces to Pin 2 and 3. On the original board the traces are further apart;
they diverge somewhat. On the replacement board those traces are closer and
mostly in parallel.
Depending on the tolerances of the components one may end up with a
situation where the difference between XTAL1 and XTAL2 gets too large. On
my replacement board there was a difference of about a volt between the
amplitudes - much more than on the old boards. I don't really KNOW if
that's the problem - however with a changed value for C23 the board did run
the whole night without a single hiccup. Perhaps the whole clock circuit
becomes unstable (and hence vulnerable to other interference) if the two
signals are too different?
I will test this further, probably this evening. Once again - so far this
is only a hypothesis.