New non-volatile memory promises 'instant-on' computing American boffins say they have made a significant step towards new chip technologies which could lead to faster, more durable non-volatile memory in place of RAM - and so to "instant-on" computing without boot-ups or hard disks. The new research comes in the area of ferroelectrics, materials used today in so-called FeRAM or …
One thing I wonder about with non-volatile memory replacing RAM is encrypted data. Some data is required to be encrypted when it is not in memory ("at rest"). If the data becomes unencrypted and then placed in FeRAM, it is then unencrypted at rest. This poses obvious security questions. Perhaps those who need encryption will have to stick with RAM and a disk.
On the flip side it will be great for devices that don't need to encrypt things that heavily!
In response to the encryption issues I would say people writing encryption software will just ensure that encrypted data remains encrypted until it is executed by the CPU and so even when loaded into this new type of RAM it would be in encrypted state and then as the CPU needs to use a piece of data the encryption software would have to decrypt the data and feed it to the CPU and then manage the output ensuring that no unencrypted data processed from the CPU remains in RAM after (possibly by encrypting the output of the CPU).
The encryption software would essentially become a memory management program sitting low level in the OS and kicked into action when ever an encrypted piece of data is used.
There are no blocking issues to encryption introduced by this new technology, just new problems to solve.
Those of us who remember what the 1980s were really like (as opposed to the romanticised versions portrayed in the media) may remember "magnetic bubble memory", which was always supposed to be the Next Big Thing. Till it vanished from the face of the earth.
Surprised nobody, around the time cores were taking over from drums, built a box using both a drum and a solid-state cache big enough to hold an entire repeating loop of a decent length. With careful programming, it'd've blistered along .....
Those who need encryption will need a bit of specialized hardware, as well as an appropriate driver, and security software that makes use of it.
But this isn't as unreasonable as it sounds - merging the concepts of HDD and RAM into a generic "memory" would require a major re-engineering of, well, pretty much everything anyway: motherboards, OS, and applications. No more RAM and SATA controllers, no more paging, no more "save" command. Figuring out how to stick a bit of volatile memory on the mobo to store crypto keys would take an irrelevant design time compared to *that*.
As a bonus it could actually be designed with security in mind (so as to somehow avoid those annoying freeze-the-RAM attacks).
Trademarks are held for particular "classes". Different companies can hold trademarks for the same name in sufficiently dissimilar classes, where brand confusion is unlikely to occur.
Filters != computing != football.
Also, Trademarks are country- or EU-wide. Ramtron might not hold the trademark in Finland. And I can't be arsed to check :-)
Hasn't this this been discussed to death here?
The are sound architectural reasons for separating data storage and working memory (security concerns as mentioned by the AC above being just one of them).
In any case, if non-volatility were so important, every motherboard manufacturer would just stick a bloody great capacitor next to the memory slots and be done with it. As it is we have suspend to ram for when the machine is only going to be idle a short time and suspend to disk for longer periods of inactivity.
Still, I see at least someone at El Reg gets it, given the article's URL.
As for instant-on technology, this goes back to way before the C64. I'm old enough to have actually worked (in high school, I'm note _that_ old) on a machine with ferrite core memory: The old kind with rings of ferrite with crossing wires through the hole. Even if you yank out the power, it just continues from where it left when you put the power back on (match that, C64!).
As for encryption, it should be possible to encrypt the RAM: Each block of 4 to 8 machine words is encrypted and decoded when loaded into cache. This makes cache misses more costly, but if these are rare enough, it will be bearable. And the cache can be made of volatile memory, so it is cleared at power-down.
If you add dedicated encryption/decryption hardware, it won't be too bad. You had better not lose your password, though. :-)
I worked on testing an experimental ferrite ring with wires through hole RAM module in the seventies. It had 16KBytes then - and was a fair bit larger and more expensive than the similar capacity silicon memory modules then, but the latter were not thought reliable enough for the processors that controlled phone exchange switches.
Nonvolatile RAM does not necessarily mean any of these things:
inability to do a hard reset,
loss of separation between storage and working memory,
no more paging,
no more save command.
However, it may result in the following:
much less time to do a hard reset,
architecturally flexible separation between storage and working memory (see Windows CE (!) for an already available example of that),
paging/caching only when necessary,
instantaneous saves.
Consider:
If every block of memory can be flagged as storage or working memory, a hard reset would be simply wiping the working memory and continuing on.
Then the amount of working memory is only limited by how much total memory there is and how much is used for storage.
Flash or an HDD can be used to provide less-expensive storage for paging, but may need to be cached.
Saving would consist of 1.) flagging the memory for the current object as storage, 2.) invalidating (if desired) the previous version, and 3.) copying the current version to working memory (if you plan to continue to work on it.) All of this would happen in memory and so would be quite fast. The option for (2) would also allow you to keep as many versions of something you want.
The real problem is cost. Regular RAM is nowhere near as cost-effective even as flash (which is still not quite as cost-effective as an HDD). I don't see this type of tech becoming cost-effective within my lifetime.
Which is probably why they went to mostly Flash storage for PalmOS 5 models like my T|X. A little more expensive, but at least it keeps data even when the battery runs down or loses charge. But flash as an all-in-one is impractical for anything beyond PDAs and embedded systems at this point--beyond a certain capacity, the costs get too exorbitant. Plus there's the issue of some apps needing fast memory (flash is notoriously slow as a memory medium).
Can we please stop propagating the myth of "instant-on"? Even if there was no time required to read data into memory, there are still many things a system must do when it boots, and these things require processing time, not disk read time. Examples are time synchronization, network detection and activation, enabling wireless devices and detecting access points and nearby devices, etc. Would your time be considerably less than a typical cold boot from magnetic media? Sure. Would it be "instant"? Not even close.
And to all the people still going on about Windows a reboots -- quite it, would you? I'm certainly no Windows fanboi, but I've never seen always-on being a problem with XP. My systems stay on 24/7, and the only time they reboot is when I install updates or install/uninstall software, and I don't have any problems with stability; neither do the friends or family members running XP. That's not to say that nobody has those problems, but I find it hard to believe that they're as prevalent as people make them out to be (unlike, say, Win98 which did have severe stability issues when kept on 24/7).
FRAMs, trademarked or not, have always promised more than they were able to deliver. Ramtron spent years telling me that production and density problems were about to be solved, and they never were. As the nineties wore on, FRAM dwindled ever further into Flash and DRAM's rear view mirror. Now a new material, which looks like it's from the same family as PZT, which had awful deposition problems. Arse.
Expect this technology to have no impact whatsoever.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
