If memory serves correctly they achieved the best results by using a can of compressed air to freeze the ram in place before removal.
//Small edit to wording
It's a matter of being more determined than your attacker. Imagine a device that will irretrievably brick itself if tilted more than a certain angle, if left unpowered for more than a certain time, etc. and that has to be under constant guard. This seems almost incompatible with any kind of personal use. And some measures may only work in one instance, with the attacker planning for them the second time they have an operation.
Many attackers also don't have the same restrictions the police has. In the Ulbricht case the police may have been forced to use a device that copies the data with no human intervention just to preserve the chain of evidence and not have suspicions that the agent operating the laptop altered it while installing additional software. An attacker operating in the grey/dark area might just immobilize the user, snip the wrist cable, and then retrieve the necessary data either directly at the console or by siphoning it via the network. Or the police may just start video recording in great detail every step from the moment an agent touched the laptop until the data was exfiltrated to remove suspicions of tampering.
But such a tool would be of great effect against an undetermined, unsophisticated attacker committing a crime of opportunity.
Putting a regular device in a safe leaves it exposed to someone unlocking the safe and compromising the device by implanting a keylogger inside or even by putting a replacement identical device there and waiting for the user to type the boot password.
As for methods of emergency clearing sensitive data from memory while in operation, whatever method is employed will work once. The next time the attacker is ready for that particular method. For example the police might just have to completely immobilize the suspect (and their hands) and keep the laptop in the vicinity while the "dead man's switch" is bypassed.
Not exactly. You don't want someone sneaking in and misappropriating the HSM to authorize something bad. And if you set the system up for unattended recovery from a power failure, then in all likelihood someone walking off with the server the HSM is in can use those keys indefinitely. But there are options.
Some HSMs have self-destruct mechanisms that attempt to prevent physical access to the private key (ie by lapping the chip). Some vendors (nCipher, IIRC) have a smart card (a second HSM) that is required to authorize certain activities, like signing, or key recovery. In fact they had a byzantine generals solution that either had the key or a password for the key split between n cards. In the latter case you needed one of the original HSMs in order to clone the key, so a movie plot where you kidnap the entire team at a conference doesn't work. During initial setup the cert would be generated on the first HSM and copied to the others, having never seen daylight.
That system was quite difficult to explain to users, and I had to document it just so I wouldn't get confused and trigger a reset of the evaluation hardware (at which point all of our test artifacts have to be rebuilt).
It might be more complicated to start WWIII than to protect a signing certificate, but only just.
I think we're talking about exactly the same thing :). That's what I meant by "even they rely on being stored in a physically secure room and protected from theft". Despite all the hardening that is applied to the device, it must always be kept secure and supervised. As an example, this is what Safenet considers the intended installation environment should be [0].
This can't be effectively applied to a personal computer.
[0] http://cloudhsm-safenet-docs-5.3.s3-website-us-east-1.amazon...
Then your biggest problem is people thinking that stealing the cards will get them anything. Which, they're not entirely wrong, because those things are damned expensive. So you need a 'kinda' secure facility.
Again, the issue is that any hardware that will start up for you without any action on your part will likely start up for anybody else, too. Your laziness will probably win out...
"In recent years, however, it has become increasingly challenging to execute cold boot attacks or perform physical memory forensics due to the introduction of DRAM memory scramblers. Modern processors with DDR3 and DDR4 DRAM scramble data by XOR’ing it with a pseudorandom number before writing it to DRAM [5], [6]. These scramblers were initially introduced to mitigate the effects excessive current fluctuations on bus lines by ensuring bits on the memory bus transition nearly 50% of the time"
DDR4 is also yes in the lab -
https://web.eecs.umich.edu/~misiker/resources/HPCA17-coldboo...