[time-nuts] Cheap 5370A on eBay
yuri at ostry.ru
Fri Dec 19 20:29:08 EST 2014
Friday, December 19, 2014, 21:34:35, Poul-Henning Kamp wrote:
PH> In message <766218C3-5AC3-4189-899B-916858F1FF2C at n1k.org>, Bob Camp writes:
>>A *lot* of errors on the 5370’s (A or B) can be traced to a couple
PH> of fairly simple issues:
PH> 4) (E)PROMS loosing their memory.
Sometimes it is possible to "extract" original contents from "leaked"
EPROMs (and some parallel 28-series EEPROMS) by reading at reduced
Vcc. I done this several times with old test gear (Marconi 2955
sitting on my workbench had "leaked" calibration data in EEPROM,
for exapmle, and it was recovered successfully).
Erased cell reads as logical "one", programmed as log. "0". So, with
time, some programmed bits in "weak" cells revert to "1".
Usually I read chip at 5V a number of times, and save data in separate files,
then gradually decrease Vcc in 0.1V steps and continue to save
everything. Usually at some point there is no more new "zeros"
appearing, and comparing EPROM against readed data multiple times does
not produce any difference. If so, it is relatively good chance that
we have in buffer the same data that was originally programmed. Data
is programmed to a new chip, and old one saved for reference to return
to it if something going wrong.
If it is impossible to find "stable reading" voltage (for example for
older NMOS EPROMs that give up at 4.5V or so), it is still possible to
statistically find "unstable" bytes, processing a batch of files
readed at different voltages and from heated and cooled chip, but
result is much less reliable.
I understand, that it is always better to have a backup or to find
correct image from another unit, but sometimes EPROMs contain factory
calibration data, and you have to extract all that possible from
Yuri mailto:yuri at ostry.ru
More information about the time-nuts