MRAM, persistance of memory

[Tim May]

On Thursday, July 10, 2003, at 08:27  AM, Eric Murray wrote:

> On Thu, Jul 10, 2003 at 04:45:58PM +0200, Thomas Shaddack wrote:
>
>> It is impossible to get access to the voltage on the DRAM cell 
>> capacitors
>> (at least if the chip is in its case and we can access only its 
>> pins). We
>> can only see if it is in the range for H or L. And after a power-down 
>> (or
>> even a sufficiently long period without a refresh of the given cell) 
>> the
>> cell capacitor loses voltage steadily, reaching the level of L (or 
>> maybe
>> H?) within at most couple seconds.
>
> I would not bet on that for sensitive data.
> See Peter Gutmans and Ross Anderson's papers on RAM memory remanance.

We were reading remnant state information in DRAMs back in the mid-70s. 
When a DRAM is powered back up after some period without power there 
are "remnants" which are not really electrons (which thermalize into 
the substrate in a matter of microseconds) but which cause 
"preferential turn-on or turn-off" in the cells, due to shifts in 
threshold voltage. (This is why irradiation of the DRAMs with gammas 
can sometimes "freeze" the stored data pattern.)

Intel was the inventor of DRAM and we led the market (along with 
Mostek) for most of the 1970s. We had some really cool tools for seeing 
the internal states of DRAMs, before, during, and after things we did 
to the devices. Powering them off and watching the states they came 
back up in was child's play.

This effect, of seeing DRAMs "wake up" in preferred states, is a very 
subtle effect. And no doubt it varies amongst vendors and even between 
design and process steppings of the same vendor's part.

I would not want to be the forensic data analyst trying to do this, but 
I expect sometimes they do. The "recover data from voice answering 
machines" gadget is no doubt much lower tech. Most answering machines 
are battery-backed (duh), so a forensics expert can keep power 
maintained and even use the battery-backed store to keep the DRAMs 
nominally refreshed.

But I thought most modern answering machines which don't use tapes are 
in fact using flash, not DRAMs. Am I wrong on this? Flash is of course 
an entirely different story.

--Tim May