This picture illustrates the temperature sensitivity of OUM (my interpretation of it, anyway) and why I don't think multibit is feasible. Hope I'm wrong.
The graph on the right is from the Ovonyx.com slideshow and is what the other graphs are derived from.
frontiernet.net
This is the picture at military spec temperature range (-40 - 80C):
frontiernet.net
All VERY ROUGH APPROXIMATIONS, for sure.
I'm thinking there could possibly be a fairly simple, cheap and reliable way to compensate for the temperature sensitivity: Set aside a group of cells to be used as callibrators. Once per second or so, run a pulse of known amplitude across them and measure the resulting resistances right away and average the results. This will tell you the "temperature". Callibrate the programming pulses of the memory once per second according to the results of this test pulse.
To get better accuracy, divide the chip into, say, 16 sections, since the temperature may vary slightly across the chip depending on activity of the memory and on local conditions such as uneven flow of air across the chips, or greater conductive cooling near the socket. Have each of the 16 sections callibrate independently. All the callibration would be done on the writes, so the reads would be unfettered. This assumes, of course, that a programmed cell would not change its resistance with temperature, which I think is possibly true, since once the amorphousness is programmed, it is "locked in" until it is programmed the next time. So, maybe there is hope for multibit.
wily |
