🚀 go-pugleaf

>>>>> On 30 Dec 1992 18:03:29 GMT, chbl@sbustd.rz.uni-sb.de (Christian Blum) said:
CB> NNTP-Posting-Host: sbustd.stud.uni-sb.de
CB> rnichols@cbnewsg.cb.att.com (robert.k.nichols) writes:
> In article <1992Dec27.105039.25690@microsoft.com> chuckst@microsoft.com (Chuck Strouss) writes:
> ...
> The length of time that a DRAM cell can hold its content without
> refresh is inversely proportional to the leakage current in the
> device.  The leakage current, in turn, is an *exponential* function of
> temperature.  What this boils down to is that a device with cells that
> must be refreshed every 2 ms at max rated temperature will hold
> information for over 20 SECONDS at a junction temperature of 25 C.
> One particularly nasty consequence of this is that it becomes very
> difficult to tell whether refresh is working correctly in your system,
> and even harder to devise a test which will reliably detect memory
> cells that might be marginal (from a refresh standpoint) at high
> temperatures.
CB> Correct, but it was a simple question and I wanted to give a simple
CB> answer. I know that DRAMs can do it without refresh for seconds at
CB> room temperature, but it's really not advisable for a greenhorn to
CB> try it...
CB> BTW, I wrote a program that switches to lower refresh rates whenever
CB> possible (some PCs fail to access diskettes if the rate has been
CB> changed, and the beep won't work correctly). If anyone is interested
CB> in it, please email.

It is possible to write a program testing DRAM refresh assuming we have
some memory which will not be destroyed when refresh discontinues to
work. I suppose video RAM can be used for this purpose.
Need: disable non-maskable interrupt and disable (CLI) interrupts,
put program (code, data, stack) in the RAM, put some pattern data in
memory, reduce refresh rate, wait some time (normal refresh in PC is
17us/address, suppose 4ms to go through all addresses, can wait e.g.
55 ms), then check the data and NMI signal (memory parity error gives
NMI). Repeat it for several patterns.
Before the test: do large memory moves on tested cells to 'heat' them.