Moore's Law Fails At NAND Flash Node 147
An anonymous reader writes "SanDisk sampling its 1Y-based NAND flash memory products and has revealed they are manufactured at same minimum geometry as the 1X generation: 19 nm. The author speculates that this is one of the first instances of a Moore's Law 'fail' since the self-fulfilling prophecy was made in 1965 — but that it won't be the last."
Shortage, no. (Score:5, Interesting)
There's a granularity to advancement as it is made of discrete units of advancement and invention.
Also, I wouldn't pooh pooh the use of other techniques to keep things moving. In the terms economists use to analyze advancement, this is called "substitution" [juliansimon.org], and is the source of the counter-intuitive but powerfully predictive observation that, in a free economy, people can invent ahead of the curve faster than things become problems, like shortages.
Re: (Score:2)
Didn't Moore's Law include a ten year limit anyway?
Re: (Score:3)
Didn't Moore's Law include a ten year limit anyway?
Moore's law was really just an observation made in a 1965 edition of Electronics Magazine [wikipedia.org]. The term "Moore's Law" wasn't coined until 5 years later, when, in hindsight, his prediction proved correct. Over time, many, many people have tried to predict when the trend would stop. So far they have all been wrong. But with what we know about semiconductor physics, it is hard to see the trend continuing much beyond 2020. But that doesn't mean computers will stop growing more powerful. If we can figure out h
Re: (Score:1)
Could you take about an hour from your life to watch an interesting video?
http://www.youtube.com/watch?feature=player_embedded&v=NGFhc8R_uO4 [youtube.com]
Let me know if this changes things for you.
Re: (Score:2)
Atoms are made of subatomic particles too. They're just handy storage devices for electrons.
Re: (Score:1)
Well. Atoms are made of electrons, protons, and neutrons. Which in turn are made of quarks. So who knows. Perhaps you can actually go subatomic. But it won't be easy.
Re: (Score:1)
Some are a much closer approximation than others and the closer the approximation the more it works.
Re: (Score:2)
It's a pity there are no free economises then.
I hate those free economises to pieces!
Re: (Score:3)
They never were free.
People just didn't realize how much they were under the control of the state.
This is why many immigrants are successful business people -- they haven't been here long enough to know the extent to which the state can step in and take control.
The rest of us have the sense not to make a move for fear of doing something wrong. There are so many laws that it would take a life time to comprehend them and whether or not a decision meets the state's
Re: (Score:2)
Uhh wow. When did "the state" last stop much in the way of anything? They step in for antitrust cases (very) occasionally and they'll at least investigate if you're doing something nasty enough to catch the public's attention, but for the most part if you remember to pay your taxes, "the state" leaves you the hell alone (particularly in the US with their free-market-is-a-silver-bullet zealotry no matter how ridiculous the concept is in any particular market.)
Generally speaking, if someone's going to shut
Re: (Score:2)
Re: (Score:2)
The state doesn't have to step in after the fact. The mere threat is enough to freeze action. Excessive laws create a chilling effect. [wikipedia.org]
It has not failed yet (Score:5, Informative)
Moore's Law applies to the number of transistors in a chip. Just because you have found an increase in performance that did follow Moore's Law for a while does not mean that Moore's Law is somehow about flash memory. Therefore, when the increase no longer follows Moore's Law, it does NOT mean that Moore's Law has failed. The only thing that has failed is your own prediction that things other than the number of transistors would follow that curve.
Re: (Score:1)
Re: (Score:2, Insightful)
Re: (Score:2)
Sounds to me like the author didn't have anything better to do, and this article is the journalistic equivalent of scratching his ass.
Re: (Score:3)
Moore's Law applies to the number of transistors in a chip. Just because you have found an increase in performance that did follow Moore's Law for a while does not mean that Moore's Law is somehow about flash memory.
If Moore's Law is about transistors on a chip, and NAND flash is a bunch of floating gate transistors on a chip, wouldn't logic follow that Moore's Law applied to NAND flash as well?
Re:It has not failed yet (Score:5, Interesting)
If Moore's Law is about transistors on a chip, and NAND flash is a bunch of floating gate transistors on a chip, wouldn't logic follow that Moore's Law applied to NAND flash as well?
Sort of. First, they're more like "capacistors" than transistors - their size may have some implications for them that it doesn't have for normal transistors, especially now that they're essentially using multi-valued logic for the charges in those gates. Second, most logic circuits get exercised quite a lot of the time, and heat dissipation is often the limiting factor, but this isn't the case for SRAM and Flash memories, and you could cheat Murphy by going 3D and replicating the strucure along the Z axis, which is, I believe, what a lot of companies are trying to do right now. Since Moore's law is a speculative observation, and not an induction on any specific first principles in semiconductor technology, the phrasing "Moore's law should apply to X" sort of doesn't make any sense. There's no "should" here because Moore's law doesn't shy why it should apply to any specific type of circuits.
Re: (Score:1, Redundant)
Re: (Score:2)
The capacistor (Score:2)
Re: (Score:2)
NAND flash = transistors on a chip (Score:2, Interesting)
So what do you think NAND flash is made of? Tiny spinning har
Re:NAND flash = transistors on a chip (Score:5, Informative)
It's unreasonable to claim that Moore's law applies to special cases.
Re: (Score:2)
Moore's Law says that the constraints will always be overcome.
If they can't, it's by definition a breakdown of Moore's Law. It makes no sense to say that it doesn't count as a violation of Moore's Law because of constraints--constraints are all that it,s about, so you're saying that it doesn't count as a violation of Moore's Law because it's a violation of Moore's Law.
Re: (Score:3)
Re: (Score:2)
Violations? More like statistical deviations.
Re:NAND flash = transistors on a chip (Score:4, Insightful)
* Given the persistence of the trend, and the lack of sudden leaps in technology, Moore's law may speak more to human ingenuity than integrated circuit technology.
Re: (Score:3)
It's unreasonable to claim that Moore's law applies at all, because it is not a law, was never a law, and never will be a law. Not in the legal sense, and not in the physical makeup of the universe sense*. Moore's Law is a statistical anomaly.
In other words, it would more correctly be described as "Moore's Observation"?
Re: (Score:2)
In science "laws" aren't theories that have grown up. Laws are just handy formulations that, though unproven (or even disproven), are useful.
e.g. Newtons laws of motion. the Law of gravity.
Re: (Score:2)
In science "laws" aren't theories that have grown up. Laws are just handy formulations that, though unproven (or even disproven), are useful.
e.g. Newtons laws of motion. the Law of gravity.
Granted. But if someone announced they squeezed five times as many transistors on a chip as last year, it wouldn't generate as much news as someone who announced that they have managed to hang a 20 ton boulder in mid air with no wires or supports of any kind. Gravity is a real phenomenon. Moore's law isn't.
Re: (Score:2)
Moore's Law applies to the number of transistors in a chip. Just because you have found an increase in performance that did follow Moore's Law for a while does not mean that Moore's Law is somehow about flash memory.
Uhhh, the article refers not at all to anything about performance. It refers to the fact that the chip is still using a 19nm process. i.e. the transistors are still 19nm on each side, and because of that, there's the same number of them. It's saying Moore's law has failed exactly because it's 18 months later and you would expect 13nm parts by now (which would have half the area, and hence pack twice as many in), for the same price.
Re:It has not failed yet (Score:4, Insightful)
It's saying Moore's law has failed exactly because it's 18 months later and you would expect 13nm parts by now
Or a die area twice as large.
Re: (Score:2)
Re: (Score:2)
Re: (Score:3)
Uhhh, the article refers not at all to anything about performance. It refers to the fact that the chip is still using a 19nm process. i.e. the transistors are still 19nm on each side, and because of that, there's the same number of them.
Actually, it doesn't say that. While they are still using a 19 nm process, they found a way to pack them closer together, and hence there are more of them even though they are still the same size as the previous ones. They didn't say how much closer, though. Packing the units of the same size closer together is the kind of thing you can probably only manage to get useful improvement out of once. Then they'll probably make the chips bigger once, to deliver more transistors. This sounds like the stopgap thing
Re: (Score:3)
It refers to the fact that the chip is still using a 19nm process. i.e. the transistors are still 19nm on each side
Nope. It just means they're 19nm on their short edge. The length of their long edge is unbounded. Specifically, the 1X manufacturing process was 19nm x 26nm, while the 1Y process is 19nm x 19.5nm. It's not twice the density, but it is more dense.
Re: (Score:1)
Secondly, looking closely at Moore's Law on a logarithmic scale you'll see that it doesn't EXACTLY follow the "line," som
Re:It has not failed yet (Score:5, Informative)
Moore's law which states that computing power (not necessarily transistors) will double every 18 months.
Wrong. This is what Moore actually said:
The complexity for minimum component costs has increased at a rate of roughly a factor of two per year... Certainly over the short term this rate can be expected to continue, if not to increase. Over the longer term, the rate of increase is a bit more uncertain, although there is no reason to believe it will not remain nearly constant for at least 10 years. That means by 1975, the number of components per integrated circuit for minimum cost will be 65,000. I believe that such a large circuit can be built on a single wafer.
Notice how it says nothing about "computing power".
Re: (Score:2)
Re: (Score:2)
Re: (Score:1)
Depends on the type of memory. DRAM is made of capacitors.
Re:It has not failed yet (Score:5, Informative)
And transistors (even floating gate ones - they're just transistors with an extra gate not attached to anything) has a strong correlation with capacity.
There are two kinds of ICs out there - pin-limited and area-limited. Pin limited ICs are your SoCs and CPUs and such - where the functionality of the entire chip is limited entirely by the number of I/O pads you can stuff on the die and the package while still maintaining adequate yields (the more I/O pads, the more chance of failure during bonding to the package - so while the silicon die may work fine, the attachment to the package didn't).
Area limited ICs are the opposite - these are where their functionality is limited purely by silicon area. The problem with making a die too big is the increased likelihood of failure caused by wafer imperfections, which decreases yields. As each wafer has a fixed area, a bigger die also reduces the number of ICs you can make from it. So bigger dies lead to lower yields due to imperfections and lower yields due to being able to make less per wafer (the fixed cost is actually pretty large compared to the processing costs).
Area-limited ICs include camera sensors (you want bigger sensors, but bigger sensors translate directly into lower yields as the sensor matrix has more imperfections ("dead pixels"), lower yields (bad sensors with too many bad pixels, and lower numbers of sensors per wafer), an higher costs (which is why a full-frame dSLR costs way more than one with an APS-sized sensor). Likewise, memory products are also area limited - because if you can use more die area, you can have a larger device. But too large means your high-cap dies are low yields and thus high prices. So to solve this, smaller transistors mean you can pack double the transistors in the same area (per Moore's law) and have practically twice the storage.
An area-limited IC tends to be very transistor-dense. A pin-limited IC tends to have hotspots of transistor density (embedded memories like caches) which comprise the vast majorities of transistors in a chip, but for the most part, what takes up space on pin-limited ICs is wiring. So much so that wiring tends to be the one spreading transistors out and making them less dense.
Re: (Score:2)
I thought the same exact thing, TFA is f'in stupid. What about motherboards, BIOS, DDR, harddrives & toasters? Moors law doesn't seem to apply there, so why should it here, what crappy journalism, let's apply w/e terms we want w/e we want and see who buys our BS.
DDR6 (Score:3)
What about motherboards, BIOS, DDR, harddrives & toasters? Moors law doesn't seem to apply there
For DDR, it's already up to the limits of how fast the player's feet can move. It took a few years to get from the original Dance Dance Revolution, whose hardest song "Paranoia" had bursts of six steps a second, to the 10 Hz bursts of "Max 300" in DDRMAX: Dance Dance Revolution 6th Mix. It took even longer to get to runs of over 13 Hz in "Fascination Maxx" in Dance Dance Revolution Supernova.
er... come again? (Score:1)
How does this have anything to do with Moore's law?
Moores law doesn't refer to density in any way. Especially not that of storage.
Moores law was talking about CPU's and their complexity.
But if you insist on claiming storage should be measured this way, then
Take a look at average flash disk size (and actual cost to manufacturer since you must apply some sort of baseline ) and I'd say you're easily still getting a doubling every 2 years.
Re: (Score:2)
Yes, it does. It was not, however, referring to the lithography resolution that is cited in this article. Moore's law refers to the number of transistors per area unit, which applies to both storage and CPUs (and is actually the more relevant statistic, since the point of smaller resolutions is to cram more transistors in). That fact is mentioned in the article, but not what a
Re: (Score:2)
Lithography resolution and transistors per area are inversely connected, so if the lithography doesn't shrink, the number of transistors can't increase (assuming you didn't lay them out *really* lazily the first time round).
In order to get twice as many transistors in (assuming a sane layout at first), those transistors need to have half the area, and hence 1/2 the linear size, and hence we would need 13nm lithography by now to have kept up with moore's law.
Re: (Score:2)
Re: (Score:2)
Sorry, slashdot ate the square root in there >.
Re: (Score:2)
And now it ate half my >.<
Re: (Score:2)
Re: (Score:2)
If you halve the linear size, each transistor takes 1/4 the area. For half the area, you only need
Re: (Score:2)
Yep, slashdot ate my unicode square root symbol. Sorry.
Re: (Score:2)
How does this have anything to do with Moore's law?
Moores law doesn't refer to density in any way. Especially not that of storage.
Yes it does –Moore's law says that transistor count on a chip doubles every 18 months for the same cost. Or, to put it another way, transistor size halves, or, to put it another way, process size shrinks by root two. This is an example of the process size staying exactly the same size over 18 months at no significant reduced cost. Therefore it's an example of moore's law failing.
Re: (Score:3)
No, there is no statement about size. only a statement of number of components. Flash has thus been surpassing Moore's law, feature size is irrelevant
self-fulfilling prophecy? (Score:5, Informative)
I don't think the summary writer knows what that means.
Re: (Score:1)
Moore was from Intel. Intel based their product cycle on his "law". They specifically plan, develop and release products to double transistor density every 18 months. How is this not self-fulfilling?
Re: (Score:2)
Just because one plans for it doesn't mean that they can do it. Being able to maintain this pace without some huge increase in R&D is what is so incredible.
A plan (not wish) gets you half way there (Score:2)
On the other hand, if Intel planned to increase by 20% every 18 months, that would practically guarantee they wouldn't double in that time period. I'm guessing your plan for the next 18 months, to the extent that you have one, is to leave your income exactly the same as it is today. Is th
Re: (Score:2)
They specifically plan, develop and release products to double transistor density every 18 months.
Why in the name of Babbage would they do that?
Re: (Score:1)
I thought the same at first read. However [wikipedia.org].
Re: (Score:2)
I still agree that the term "self-fulfilling prophecy" in this context is just plain wrong. Yes, if you're capable of advancing *faster* than "twice the transistors every 18 months" then it's self-fulfilling in that you can throttle-down the rate of increase. The problem comes in when you can't keep up with that 2x every 18 months rate. It's not self-fulfilling because you might not be capable of keeping up with that. Because of that seco
Re: (Score:2)
It means that they are not able to shrink smaller than 19nm for this next generation of NAND flash memory. Normally the numbering system from generation to generation included the number in the tens(and 100s) column. But since they are at that same physical size, they had to use a new numbering scheme. 1x -> 1y.
All it means is that the physical limits of shrinking the die size are being hit. And the author is simply saying that it's a sign of things to come - DUH!!! Also, to your parent post,
Journalist Wanted Moore Hits (Score:5, Informative)
From Wikipedia: Moore's law is the observation that, over the history of computing hardware, the number of transistors on integrated circuits doubles approximately every two years.
From article linked off the main article: SanDisk has now revealed that 1Y – now described as a generation rather than a node - is the company's second generation at 19-nm. What the company does claim to have achieved is a reduction in the memory cell size from 19-nm by 26-nm to 19-nm by 19.5-nm, delivering a 25 percent reduction of the memory cell area.
So, if you can fit more cells using the same size process, it doesn't go against the spirit or the letter of Moore's law. Moore's law is about computing power. If you get more computing power without reducing size to do it, that still counts.
Re: (Score:2)
It's a big stretch to call Unknown Lamer a journalist.
Re: (Score:1)
Isn't there a law about more outlandish articles getting more hits? This is pure sensational immature blather and shouldn't have been re-posted to Slashdot. The only conclusion to draw from this is that SanDisk made a product decision that didn't fit with Moore's law this one time. Wow so exciting.
Re: (Score:2)
Right, Moore's original piece was rather agnostic to the processes which would be required to increase the density.
http://download.intel.com/museum/Moores_Law/Articles-Press_Releases/Gordon_Moore_1965_Article.pdf [intel.com]
Re: (Score:1)
Moore's law is all about positive financial feedback:
- better products (capacity, performance, usability) ->
- more money ->
- process development (scaling feature size) ->
- better products.
It worked so well because there was a single variable - process feature size - that translated investment money into more attractive products. That produced 30 years of exponential growth and increased transistor density ~1e+4 times.
Tackling multiple problems (design, IO, packaging) doesn't work nearly as well - y
No shit (Score:2)
It is also common in the industry, to improve on something on an existing process. Intel does it every tingle process with their tick-tock strategy. They make a new process and release a chip with a largely existing design on it, then they make a new design on that process that is more efficient, then a new process, and so on. You see it with the GPU makers, they release updates outside of when the process shrinks and so on.
There was never anything saying that increasing transistor count was the only way to
Re: (Score:2)
> Moore's law is about computing power.
No. Moore's law is about transistors per die. More transistors may translate into extra storage, extra features, integration, lower power, higher efficiency or other things that may not necessarily relate to computing power.
With Haswell, we have even reached a point where Intel can even afford to put the VRM on-die - one of the last things I was expecting to happen until news of it leaked out... I was certain they would integrate the IO Hub / South-Bridge first but
Post-Moore Advancement (Score:2)
Moore's Law is more of a "guideline" really... (Score:2)
"And thirdly, the code is more what you'd call "guidelines" than actual rules." (Mod points for knowing where that quote comes from.. )
Given the physics of how flash actually works, I'm guessing that we will see a more step wise improvement in storage density. But Moore's law is about increasing complexity, not density. So the logical size of flash devices will continue to go up, even if the density is not improved.
I don't think we are close to a point where Moore's law is going to be proven false, not
On the other hand... (Score:1)
Re: (Score:2)
It's equally valid to say SanDisk failed (umm, violated) Moore's law.
In a way that is true, because at the same time Moore's law was discovered, it was set as an engineering goal.
Comment removed (Score:3)
Re: (Score:2)
Re: (Score:2)
Re: (Score:3)
It's not a law ... (Score:5, Insightful)
Moore's law has never been a 'law', it's a historical observation.
It has never claimed that this will be true going forward, merely that at the time it was observed that was the case, and it's largely held up since then.
The fact that it's held true this long is staggering, but the fact that it might be running out is hardly surprising. Moore never claimed this would continue forever.
Re: (Score:2)
That's what a law is in science. More precisely, it is a relation between observations, in this case device density and time. It is perfectly valid to apply the term to something purely historical and empirical.
Re: (Score:2)
Except this 'law' isn't predictive like, say, gravity, which says that the next time you drop something, it will also fall to the ground.
Moore's law never has been, and never was intended to be a 'law' in that sense.
Re: (Score:1)
Oh, it's definitely predictive, hence the predictions. What is lacks is a fundamental theoretical basis, but that is not a requirement of a scientific law. Plenty of good scientific laws got their start without any theory, and even drove the search for a theory.
Consider Kepler's laws of planetary motion which were developed without knowing that gravity was supplying the force to create the orbits. Also, there is a law of centrifugal force, despite the fact that the force is fictitious.
Re: (Score:2)
If Moore's law holds, it means NP-complete problems can be solved in polynomial time. The number of steps is exponential, but the growth of processing power is exponential, too, so one step takes less and less time when time is increased. (This observation abuses the usage of word "time" instead of "steps")
You are conflating complexity with run time. Complexity is orthogonal to actual run time. If the number of steps is exponential then the NP-complete problem is running in exponential time. No amount of handwaving about increases in processor speed will change that fundamental tenant. Now if we ever get quantum computers off of the ground then yeah, maybe you can solve an NP-complete problem in polynomial time, but that would only be because you then have built what is essentially a real-world non-determ
If I could mod an AC ... (Score:2)
Re: (Score:1)
Moore's law is about the increase in transistors existing per area. You have assumed an equal increase in transistors that are doing something per area, which is not in evidence.
Re: (Score:2)
The complexity for minimum component costs has increased at a rate of roughly a factor of two per year (see graph on next page). Certainly over the short term this rate can be expected to continue, if not to increase. Over the longer term, the rate of increase is a bit more uncertain, although there is no reason to believe it will not remain nearly constant for at least 10 years. That means by 1975, the number of components per integrated circuit for minimum cost will be 65,000.
http://download.intel.com/mus [intel.com]
Re: (Score:2)
He didn't say forever, he didn't say always. The fact that it's largely held as true since then is a bonus, but it certainly doesn't make this a 'law'.
This was a 10 year prediction, and the word 'law' doesn't appear in the entire article. Not even Moore would claim this is a law.
I'm generally a fan of The Great Moore's Law Compensator/Wirth's Law [wikipedia.org] which says that "software is getting slower more rapidly than hard
Peter Clarke is a moron (Score:1)
Every word he said is wrong, including "and" and "the".
Since Moore's Law is silent on "minimum feature size", then no observation of that metric can be contradictory to that law.
Nearing theoretical limit? (Score:3)
I am not an engineer. So, you engineers out there--are we nearing the theoretical limit on these things? I mean, 19 nm is pretty darn small. It seems to me that at some point Moore's law has to fail simply because you can't make a connection less than one atom thick. And making a connection one atom thick would be stupid, I would think, for reliability reasons. So--is Moore's law, as extended to NAND flash memory failing due to the fact that it has nearly reached its lowest theoretical size?
Re: (Score:3)
Re: (Score:2)
I am not an engineer. So, you engineers out there--are we nearing the theoretical limit on these things? I mean, 19 nm is pretty darn small. It seems to me that at some point Moore's law has to fail simply because you can't make a connection less than one atom thick. And making a connection one atom thick would be stupid, I would think, for reliability reasons. So--is Moore's law, as extended to NAND flash memory failing due to the fact that it has nearly reached its lowest theoretical size?
Quantum effects will prevent making electrical components too much smaller than they currently are. However, there's really nothing stopping companies from producing larger dies or adding more vertical layers to a die.
Re: (Score:1)
19 nm is 190 angstroms. Transistors have been demonstrated operating at ~20 angstroms feature size.
However, current trends show the market will not generate sufficient revenue to keep pushing on Moore's Law.
Economics will break before physics does.
It has to fail eventually (Score:3)
Moore's "Law" has to fail eventually, because if you keep doubling (the amount of transistors on a chip) every couple of years you would soon (in a century or two) have more transitors than there are (elementary) particles in the universe
Of course we will be up to Windows 95 by then...
Parameter limits (Score:1)
Gate oxide thickness at 19nm is about at limit.
Bubble memory (Score:2)
Remember bubble memory?
I wonder if that would have kept up with Moore's Law a lot better.
Well known issue in the industry (Score:3)
This isn't a new issue to people in the industry. Here's a more useful article from last year: "Is the cost reduction associated with IC scaling over?" [eetimes.com] "Clearly, dimensional scaling is no longer associated with lower average cost per transistor."
The cost of wafer fabs has been going up with each generation. Intel says that a cutting-edge fab now costs upwards of $10 billion, twice the previous generation. That's why higher densities no longer reduce cost. The upper limits of optical lithography are being reached because light, even "deep ultraviolet" light, is too coarse a tool. "Extreme ultraviolet" (soft X-rays, really) are being tried to get down to 10nm or so, but the processes are currently slow and barely work. Electron beam machines, which can go below 10nm, have been around since the 1980s, but they work by writing the chip with an electron beam, not with a mask, which is very slow for a production process.
This is for mostly-static memory. For active transistors, as in CPUs, heat dissipation is already limiting density. CPU clock speed maxed out between 3 and 4 GHz several years ago. (Yes, 8GHz has been achieved with an AMD CPU running in liquid helium. So?)
With the upper limits of speed and density in sight, work is now focusing on reducing cost and power consumption. Hence the push to use ARM CPUs in more applications.
The summary is barely comprehensible. (Score:1)
Just a hundred of atoms (Score:2)
Re: (Score:2)
Well, to be fair the only experiment we know of resulted in some weird temporal displacement effect, followed by a spectacular crash...