[Cryptech Tech] RAM as source of entropy

[Leif Johansson]

On 2014-02-10 20:06, Warren Kumari wrote:
> On Mon, Feb 10, 2014 at 4:52 AM, Joachim Strömbergson
> <joachim at secworks.se> wrote:
> Aloha!
>
> Steven Bellovin wrote:
> >>>
> https://www.usenix.org/legacy/event/sec08/tech/full_papers/halderman/halderman_html/
> >>>
> >>>
> might have some data on that.
>
> Yes, and Warren pointed to that paper earlier in the thread. That is why
> DRAM is less interesting than SRAM. The reason for this is that for DRAM
> we need to wait a given time to allow the memory to decay "enough" - and
> that might be a long time.
>
> For SRAM it is basically the opposite, we try to trigger a fixed state
> from a power down condition. But there are remanence effects for SRAM
> too. And they are relevant to the discussions about using SRAM as a
> source of entropy as well as the storage of keys during operation of the
> Cryptrec system. Warren Kumari mentioned that HSMs move keys in memory
> around to avoid remanence effects.
>
> The paper "Data remanence in semiconductor devices" from 2001 by Peter
> Gutmann is a very good and relevant paper. He describes the remanence
> effects and specifically points to the problem of having keys in SRAM
> memories at fixed positions. This is something we probably should plan
> to handle.
> http://www.cypherpunks.to/~peter/usenix01.pdf
>
>
> Sergei P. Skorobogatov has written several papers on remanence effects
> and points to the need to detect that the ambient temperature for the
> system to not be lower than a given level. If that happens keys should
> be destroyd. This applies for all memories:
>
> "Semi-invasive attacks -- A new approach to hardware security analysis"
> from 2005 is a good paper:
> http://www.cl.cam.ac.uk/techreports/UCAM-CL-TR-630.pdf
>
> But Skorobogatov also shows that just temperature detection is enough.
> We really should do active zeroisation of memories when a relevant event
> has been reached (temperature, breach of tamper protection, unplanned
> movement of unit etc.)
>
> And external SRAM power supply connection should be automatically
> shorted to ground whenever the power is off. Floating VCC seems to
> drastically increase remanence.
>
>
> The paper "Data Remanence Effects on Memory Based Entropy Collection for
> RFID Systems" from 2011 by Nitesh Saxena and Jonathan Voris (from your
> university Steven) is probably the most relevant for the discussion of
> using SRAM as basis for entropy. The papers shows that the remanence
> effects are bad enough to make it impractical to use in RFID tags in
> which the memory is also used for other processing. The extraction of
> enough entropy simply takes too long time. Esp. when the temperature
> drops. A good paper.
> https://www.cs.columbia.edu/~jvoris/papers/ijis_11.pdf
>
>
> Finally there are at least some design work being done to implement SRAM
> cells with remanence protection. The paper "Security strategy of
> powered-off SRAM for resisting physical attack to data remanence" from
> 2009 describes modified SRAM cells that reduces remanence effects. This
> is not very applicable to Cryptech though:
> http://goo.gl/7T8Sll
>
>
> Remanence can even be used as a clock the TARDIS protocol uses decay
> memory decay as part of a secure protocol for embedded devices. Really
> cool research!
> http://goo.gl/zgeS1y
>
>
> Conclusion:
> -----------
> No, SRAM as a source of entropy is probably not a really good idea. Esp.
> as the only source of entropy. We can use a dedicated memory, ensure
> that VCC is properly grounded and write PRNG-generated patterns into the
> memory. All of this would probably make the SRAM memory a better
> entropy. But I think the writing is still clear enough to not continue
> with this.
>
> But there are still some good issues regarding SRAM (and DRAM) memory
> usage to be learned here. Issues we still need to consider, plan and
> design for in order to mitigate remanence in all other memories we will
> be using in the Cryptech system.
>
> > So, one interesting observation (at least for me) is that, if we have
> > to store keys in RAM (inside the envelope!), SRAM is probably better
> > than DRAM.
> > Yes, we still need to do things to decrease stickiness and active
> > zeroization if tamper is detected (and I like the tie power the ground
> > as a final step) as well.
>
> > This has probably been blindingly obvious to everyone else, but wasn't
> > something I'd though of...
>
>
>
> (And someday I will test my SRAM as entropy with PRNG feedback idea,
> just to put my curiosity to rest.)
>
> > Yah. That sounds really interesting / fun. I'd also like to try a dark
> > CCD vs SRAM test (both simply powering / off and PRNGing the SRAM).
> > The CCD test I can probably do fairly easily (USB webcam), the SRAM
> > one might be a bit trickier. I've got a TI MSP430 dev board in a USB
> > thumb drive form factor (http://www.ti.com/tool/ez430-f2013 ) -- this
> > with a small SRAM attached seems like it might be a good way to play,
> > because I can take it along and play with it on a plane (a while back
> > I was sitting on a plane with an Arduino with a few LEDs on it...
> > Flight attendant came over and asked me to put it away, because
> > apparently it had freaked out some other passenger....)
>

You didn't just explain it was a "TV typewriter"  :-)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cryptech.is/pipermail/tech/attachments/20140210/b7b343a2/attachment.html>