Brit brainiacs say they've cracked non-volatile RAM that uses 100 times less power A new type of memory that could make computers and smartphones far more energy-efficient, if it ever reaches production, has been developed at the UK's University of Lancaster. Most computers, phones and data centre boxes today rely on a combination of two main types of memories: DRAM and flash. As Reg readers know, DRAM is …
The times or multiplication operator when used with a number greater than one produces a result which is larger rather than smaller than the original. Or perhaps "times less" is the operator, whatever that means. The article itself uses the correct "1 percent" terminology.
"So... uh, Schrodinger's Dogshit...?"
Yecch! I just trod in it.
No I didn't.
Or both.
My office floor stinks and doesn't stink at the same time, I have been fired and not fired, my brain is exploding, no it isn't, that man Schrodinger has a lot to answer for, no he doesn't, Aaugh! I am quite calm about it all...
This post has been deleted by its author
> "Using the new memory in phones and PCs could allow them to be low-cost dumb terminals with minimal local energy consumption,"
How does that in any way follow? For neither phones nor PCs is memory the most significant power sink, so how is reducing the power consumption of just memory going to be a game changer?
> how is reducing the power consumption of just memory going to be a game changer?
It won't, of course. It's just an opportunity to advertise the business model where everything you do on your phone (assuming you live in a spot with enough bandwidth) happens on Google/Facebook servers, and you've only got a dumb terminal to feed them your information.
(The difference with today is that nowadays some things do (or at least can) happen on your phone, and Big Brother only gets notifications of them happening. With a thin client things will happen on Big Brother's servers, and you'll only get a notification of them happening - and the bill.)
" "Using the new memory in phones and PCs could allow them to be low-cost dumb terminals with minimal local energy consumption,"
How does that in any way follow?"
It seems to me that,"It'll all be fat-server-thin-client any day now," was the archetype for,"This will be the year of Linux on the desktop."
The distinction being that when you ask,"Why is that a good thing?" the latter crowd will throw more reasons at you than you would ever want to hear, whereas the former will, at best, respaff a bunch of meaningless marketing-speak.
10/100 Ethernet was going to get us there. 32-bit processors were going to get us there. Java was going to get us there. 64-bit processors were going to get us there. Optical mice were going to do it (I'm sure somebody said it).
Not saying it hasn't got some traction with the whole,"cloud," thing, just saying it's mostly bollocks.
"How does that in any way follow?"
I wondered about that but then the account ent off into data-centre power consumption. I think what they were getting at would be that it would enable data centres to be bigger and (for some value) better, enabling the thin client thing. The trouble is that, as others have posted, "better" is for the likes of Google & FaceBook. So no thanks, the better option would be more capability in the device and less need to go online which latter, in any case, seems likely to be a bigger drain on power than DRAM.
Hmm. 'Simulations'. Let's see some wafers first and measurements thereon.
There are far too many "Look! We've just made this amazing breakthrough in speed, power consumption, energy density, carbon reduction, etc." articles which never get anywhere. Perhaps you should produce a working model & characterise it before publishing?
DRAM is fast but volatile, meaning it loses information when its power source is cut
<pedant>
Well, that is true, but it also loses information in < 1/10 second if not refreshed (i.e. recharge the memory cells because they are a transistor used as a very small capacitor). The typical amount of time for a complete refresh (all memory cells get a recharge) is 64 milliseconds for current devices from DDR1 to the latest versions.
This is an overhead in all DRAM implementations (although it is transparent to userland but not to those writing the initialisation code which is quite interesting if you are using ICs and not SODIMMs).
The implementation of refresh circuitry is now well understood, but it still requires a lot of extra circuitry within the memory controller as well as the memory device and some power to operate as does the transaction based burst nature of all DRAMs (which makes them unsuitable for many applications).
SRAM (static RAM) also loses it's information if power is cut but does not need refresh; the penalty is more power required but they are very popular in some applications including some server applications.
Both the above are volatile types of memory.
NVRAM that runs at the same speed as standard SDRAM (DDR3/4 and so on) is available but they are a bit expensive but there are advantages such as very fast boot times as the OS (if used) does not need to be loaded from one memory (typically flash) to main DRAM.
</pedant>
There are a few things out there, although not at the highest of speeds of current DDR4; for many applications, though, that speed is not really necessary (especially in the embedded space).
Everspin has a couple of devices with reasonable capacities but they are quite expensive compared to standard products; the advantage is that they survive power dropouts (very common in many applications such as avionics) and the application can reside within the devices.
They are also touting the parts for data centres.
Does this mean that operating systems running with this ULTRARAM will need to wipe memory before it powers down ? If not then crooks/Gov't-spys could move the RAM chips into a reader and get passwords, keys, etc.
There was a story a couple of years back about being able to read RAM after a reboot, but this needed to be done in a few seconds. With this it could be done at leisure.
Where's the RAM encryption key held? Can't that be read from some location in the chipset with an electron microscope of sufficient sensitivity?
Although, I guess, for RAM, the key could be randomly generated on each boot, and only 'stored' for the duration of the current power-cycle, because it's not like you need to maintain the key across reboots with non-volatile RAM therefore a key doesn't need to be stored in NVRAM. As long as there is a mechanism in place to zap the key on a shutdown then I guess it wouldn't matter. (although, personally, I've always thought DRAM modules should have an in-built chip-level hardware zero-out capability, so on shutdown the entire stick can zero itself out rather than relying on natural discharge rates.)
Maybe restart could be changed to be clear RAM and reload OS. Although the clear part may only be needed for security.
But there would probably also need to be some sort of method in the BIOS and motherboard to do a forced clear of RAM and reload OS for those times when the OS goes TITSUP. The present holding the power button down would probably work with a few BIOS tweaks.
The apparent answer is a combination of both: encrypt the user storage memory but allow programs to operate in a sandbox. You can't get access to the user data until you decrypt, and you can quickly collapse both the user data and program memory heap by simply terminating the sandbox all at once if necessary. The base OS kernel can, with VERY careful writing, be left latent in memory for quick restore.
You will find it filed under "Racetrack: memory (IBM?). It's all the rage, like MRAM which is, IIRC, Thin Film magnetic memory (as used in the early 70s), but "on a chip", or FRAM, again IIRC, Ferroeletric RAM (demonstrated at Bell Labs in the Mid 1960s) "on a chip". This whole "on a chip" thing is like patenting some 500 year old concept, but "on a computer".
Maybe I should look at SAW devices with integrated electronics to implement the EDSAC acoustic delay line memory "on a chip"
Queue the Pink Floyd:
ULTRARAM could change how computers work from head to toe, according to Hayne.
Sounds like memristor all over again. See https://www.theregister.co.uk/2008/05/01/hp_labs_unveils_memsistor/ which a dozen years later we're all still waiting for. Which HP said would lead to a totally new approach to building systems. This led to "The Machine" (see https://www.theregister.co.uk/2016/11/24/hpes_machinations_to_rewrite_server_design_laws/) and the inevitable Pink Floyd reference. Given their promotional video and it's discussed usage case they seemed not to have listened to the rest of song, which seemed to have guessed what the new technology would be used for 30years earlier.
Maybe @Dazed and Confused did mean "Queue the Pink Floyd" in the same way this memory technology is being queued. That is, stick it in the queue of existing proposed but unrealised memory technologies, where one day it may come out of that queue, but probably won't.
Why is it that every new tech is automatically used as another excuse for flinging my data about to points I do not control and want nothing to do with ?
How is it that a slightly better memory tech means my local phone will suddenly stop processing anything ?
Stop it. So you might have a better version of RAM, fine. It's still RAM, and it will not change a thing as far as data processing is concerned. It will most certainly not create a world where everything millions of users do on their phones will be sent to a central computer (can we say mainframe again ?) for processing and the result sent back.
I'm sure IBM would like that, but no, not gonna happen.
> Why is it that every new tech is automatically used as another excuse for flinging my data about to points I do not control and want nothing to do with ?
Well, to be honest it's an old trend I've already observed in personal computing, long before smart phones. Apparently everyone dreams to go back to the ancient "big iron - dumb terminal" configuration. Nostalgia? Greed? Hunger for ultimate power over the "lusers"? Never managed to know. I realize it allows some interesting things though:
- Blackmail them at will ("Nice set of data you have there, it would be a pity if you lost access to it because you didn't accept our latest 250% fee increase.")
- Sell them additional services, like remedies against the bugs and shortcomings of your platform ("Premium Priority Pass - For when you really need your device to reliably work at least once every week").
- Delete their files and watch them squirm, the ugly, nasty worms.
- Sell their data to anybody who might give you any money for it. Over and over again.
And so on. So yes, there are reasons for it, just no good reasons.
I saw this and wondered immediately how switching from moving data around circuits in one place to chucking it across vast distances, which incurs lag even with speed-of-light transmission, could possibly make something more efficient. Perhaps they mean efficiency in terms of energy usage, rather than in terms of time taken to process something, but still one wonders about the energy costs of moving information about.
There's an entropic limit here in terms of how much change you have to make in the communications systems involved, and there is a physical minimum energy cost associated with the change in entropy (in chemical systems this is known as enthalpy). There might be efficiency gains from economies of scale, but I bet you they're not going to be measurable ones in the bigger scheme of things.
Funded by 5G investors? Nevermind the all power needed to get mmWave to pass through anything more dense than a vacuum.
The whole point of desktop computers was escaping the constraints of centralized computing. History says that local computing was a very good idea.
I had a non-volatile ram once. I called him Yoda, mostly due to his voice. Instead of butting me out of his way, he'd just stand there patiently staring me down, like he was trying to move me with The Force. I suspect he may have been the result of a bizarre military experiment.
The big problem with Flash isn't so much that the cells wear out but this process is random and essentially unpredictable. So a cell which should be rewritable another 10,000 times might break down on the next write.
Unclear if this will solve the problem but resonant tunneling is interesting.
If the memory cell states can be changed using far less power than now, isn't there a risk that memory will get corrupted more easily by the effects of background radiation? This has to be accounted for in space electronics, due to cosmic rays. That is seriously heavy stuff, which is not normally a problem at normal altitudes, where the earth's atmosphere provides shielding. But there is always a bit of weaker radiation about, that could cause trouble for this technology. Perhaps you cannot use this technology in Cornwall. I would not fancy it anywhere near a cancer treatment machine, either.
Well, you'll have to take into account that every time there is a thunderstorm outside, your device wipes itself. Same will happen if you wear wool or synthetic fibers on dry winter days. Or brush your hair, or sit on a chair with synthetic upholstery. Never come close to any transformer, high tension line, talkie-walkie, old electrical device (from power drill to elevator). Minor annoyances, all.
Well, the main advantage is that data centers don't have their power sources sealed such that after 3 years of use you just have to build an entire data center. For maximally rosy statements about power efficiency, please to ignore the embedded energy of building a Facebook or Google scale data center.
Yes, I believe the real meaning is "maximally efficient at transferring power over data from users to corporations". See Also: Chrome, iCloud, Office365...
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
