During the course of debugging Pace's Marching Functions test, we came
upon a hard (i.e. repeatable) error in the hardware, in which incorrect
instruction data was fetched from memory and subsequently executed in
the instruction register, causing the program to crash.  We have seen
this type of failure before, although typically the failure happens only
once or twice before disappearing.  This failure involved k-MEM s/n 012.

Analysis with the logic analyzer revealed that the instruction fetch was
referencing the wrong physical page (0 rather than 1), and that the page
address was correct for the unused banks of DRAM.  Closer examination
revealed that the address driver for the bank 0 DRAM array (a 74AS841
at location H3 on the K Memory card) was failing to drive ALM0 (address
bit 0, A bank) high enough, although the latch-enable, output-enable and
signal inputs were all correct.  Previous instruction fetches to the
same page had completed properly, so the driver was theoretically able
to drive the address line high.

I had come across a similar problem with the same address driver, same
location, but a different bit (ALM3) in a previous problem with a different
board.  In that case, too, the address line was not driven high enough,
although all the relevant device inputs were correct.  In that case, I
initially suspected a short to another driver, or another driver on the
node being enabled at the wrong time, but a systematic search for a short
failed to find one.  I finally replaced the driver, and the problem cleared
up immediately thereafter.

Noticing that the output to the unused bank was correct, and remembering
the aforementioned previous experience with the driver in H3, I immediately
replaced the driver, and the problem immediately cleared up.  Within 24 hours
of completing the repair, I came upon another memory board (the original V2
board, without s/n) with an identical problem.  Again I investigated with the
logic analyzer, and again ALM0 was not being driven high enough by the driver.
I have not yet changed this driver.

I am beginning to suspect that we have a bad batch of chips here:

    Texas Instruments 74AS841NT (plastic package, made in Malaysia)
    date code 8714XF
    other number 6191061

Both of the devices with problems driving ALM0 belonged to this lot.