The results we have obtained from swapping the two banks of 16K over are completely inconsistent.
With the 'I' bank we get one good page and one completely bad page (the value returned is always $2E for this page). This would imply a multiplexer or an address line.
However, the 'J' bank exhibits a completely different fault scenario.
If it was a multiplexer, or an address/data buffer, then I would expect that fault to be applicable to both banks of RAM - as it is external to the RAM. But it isn't.
I must admit that the fault does imply that the DRAM is at fault...
Now, you did check the power supply rails for both high and low frequency noise didn't you
@Desperado? If we have noise on the power rails (especially the +12V and/or -5V) then we are chasing ghosts!
The other possibility is a timing fault in the signals driving the multiplexers. We haven't checked these yet, and they should be a simple check with the oscilloscope.
I am also not happy why my little test program didn't seem to work. Did you take the signal measurements I asked for or not?
Dave