Scientists develop new technology standard that could shape the future of electronics design
In a study published in the journal Scientific Reports [open, DOI: 10.1038/s41598-017-17785-1] [DX], researchers show how they have pushed the memristor – a simpler and smaller alternative to the transistor, with the capability of altering its resistance and storing multiple memory states – to a new level of performance after experimenting with its component materials.
[...] The University of Southampton team has demonstrated a new memristor technology that can store up to 128 discernible memory states per switch, almost four times more than previously reported.
In the study, they describe how they reached this level of performance by evaluating several configurations of functional oxide materials – the core component that gives the memristor its ability to alter its resistance.
[...] Professor Prodromakis and his colleagues will be showcasing the technology, and presenting seven original research papers, at ISCAS 2018, an international circuits and systems conference, in Florence, Italy, in May.
"Almost four times more"?
(Score: 3, Interesting) by FatPhil on Monday January 29 2018, @10:45AM (7 children)
Either way, the 4 times comment is a bit weird as "some recent reports of multibit capable metal-oxide memory cells, most works in literature are limited to no more than 3 bits", which would be 8 states. And 92 is a factor of 11, not 4.
Never believe press releases. I guess it's all packed there in the meat of the paper, I didn't have time to do anything but skim.
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 2) by WalksOnDirt on Monday January 29 2018, @11:05AM (2 children)
Going from three bits to over six bits is more than doubling the performance. Nice, as long as it is just as fast and stable as the three bit version was.
(Score: 4, Interesting) by FatPhil on Monday January 29 2018, @12:02PM (1 child)
[* meaning "near the time when we're coming up to the era where Moore's Law fails to be predictive", but you knew that, right?]
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 1, Touché) by Anonymous Coward on Monday January 29 2018, @03:21PM
Xkcd's Researcher Translation [xkcd.com]
(Score: 0) by Anonymous Coward on Monday January 29 2018, @12:23PM (3 children)
Yeah. Anyway, more state than USofA even if it grants statehood to Puerto Rico.
(Score: 2) by maxwell demon on Monday January 29 2018, @02:18PM (2 children)
But are the US states all distinguishable? :-)
The Tao of math: The numbers you can count are not the real numbers.
(Score: 0) by Anonymous Coward on Monday January 29 2018, @06:49PM
I can never remember which one is Mississippi and which one is backward Mississippi...
(Score: 2) by FatPhil on Tuesday January 30 2018, @08:39AM
1) The shithole one...
Your go.
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 1) by flippynet on Monday January 29 2018, @04:12PM (3 children)
If you can program a cell to one of 92 states, and there is a probability that it won't read back as the exact state but rather a different "nearby" state, then this seems like a good candidate for advanced signal processing techniques, like hard drives or radio signals, I'm talking forward error correction (FEC) codes like convolutional/turbo codes, reed-solomon codes, and the like.
To me this seems like a logical progression. As hard drives advanced to the point where individual magnetic domains were so small that their reliability dropped below 100.00000% reliability, they stopped encoding one-bit-per-domain and FEC coding techniques took over. The encoding techniques for CD/DVD/BR are rather impressive, the fact that you can take an xacto and scratch across a CD and it still plays (don't try it!).
If we're going to make transistors and memristors smaller and smaller, and cram more energy levels into them with a tradeoff of decreasing reliability, it would make sense to do the same thing to them.
(Score: 2) by Dr Spin on Monday January 29 2018, @05:50PM (2 children)
forward error correction (FEC) codes like convolutional/turbo codes, reed-solomon codes, and the like.
The problem with this approach is that the failure modes become more complex and less than easily recognised.
I worked on a comms system with FEC about 40 years ago, very similar to reed-solomon (but more bits, I think)
and was told: on no account do any error correction. Stop when you detect an error: the data includes financial
data, and if you fail to detect an error and correct it to an incorrect value - possible, but unlikely - and someone
uses it to take a decision which turns out bad - you could potentially be sued for $HUGE_AMOUNT_OF_MONEY.
Since the order was given by our customer (BT) we obeyed.
I do not want my disk reads to occasionally supply bad data without knowing. (This generally does not happen
because there are checks at both hardware and file system level, and it would be highly improbable
that low level correction would go unnoticed at the higher level - unless someone turns off checking to improve speed.
(Not sure MS check at file system level).
Seems unlikely that checking twice at RAM level would go unnoticed. ECC DRAM can usually be set not to correct, but to
flag the SIMM as bad. Recommended for servers processing financial data (eg telecomms billing).
Warning: Opening your mouth may invalidate your brain!
(Score: 2, Informative) by flippynet on Monday January 29 2018, @11:44PM (1 child)
sorry, too late for you -- hard drives already use things like cyclic codes, and other things that i don't understand well. modern hard drive magnetic moments are so small that every spot interferes with its neighbors (ISI: inter-symbol interference). you can't achieve modern disk capacities without things like cyclic codes. physical sectors also have strong CRC checks on the media so error detection is present, but CRC errors virtually never happen on modern drives because the FEC codes are so good.
cell phones (and most other RF digital transports) do it too. i work just above the PHY layer in cellular technology; UMTS uses a constraint-9 rate-1/3 convolutional coding, which can pass an error-free packet even if HALF the transport bits are corrupt! "turbo codes" are even better. in fact the error correction is so good, when measurements indicate a decent physical channel, they intentionally "puncture" some of the transport bits just to trade error-correction capability for more throughput. in that case, they're relying on the FEC to re-inject the punctured bits on reception, because their measurements tell them the physical bit error rate is within the correction capability of the FEC.
I think this is going to happen to RAM. i think it's already happened in some types of high density flash (i.e. MLC).
(Score: 0) by Anonymous Coward on Tuesday January 30 2018, @08:47AM
Oh man, this is why I come back to Soylent. Fucking fascinating. Thank you! Some amazing engineering revealing itself with just a little searching, thanks to your starting point. Wow.
(Score: 0) by Anonymous Coward on Monday January 29 2018, @06:58PM
"Almost four times more"?
zero works!