The huge flaw in Moore’s Law? It's NOT a law after all Critics have had half a century to pick apart and predict the end of Moore’s Law, which marked its Big Five Zero birthday this week. It’s unlikely that Gordon Earle Moore, the former electrical engineer who authored the eponymous law for a 1965 article, and who two-years later co-founded Intel, has any doubts over its value. …
I was also bemused by this:
... the number of transistors used in a typical CPU — the CPU transistor count — would double ....
Given that this is a techie site I should think most readers would be able to guess that the number of transistors used in a typical CPU is also known as "the CPU transistor count".
If reading the Daily Mail and nutritional advice from Experts on facebook has taught me anything, it's that chemicals are uniformly bad news. It's a little-known fact but the ingredients for table salt include both sodium (which explodes on contact with water and is also used by the nucular industry) and chlorine (used as a poison gas in world war one), baking soda (chemical food additive designation E500) also contains sodium, hydrogen (highly flammable and responsible for the Hindenburg disaster) and carbon (used, amongst other things, for extremely sharp industrial cutting tools).
I think putting stuff like that in the hands of a child and encouraging them to "play" with them isn't only a recipe for disaster but should also be classified as dangerous abuse.
Well more fool you. Scientists say that over 85% of cancer victims have been routinely exposed to dihydrogen monoxide* on a regular basis and as Wilseus points out, it's a key component of tumours. Perhaps I'm just a better parent than most but I bring my children up in a completely DHMO-free environment and so far not one of them has died from cancer.
* Even this term is making light of the issue and is a flagrant tool of hiding behind science-tiffic jargon to make this terrifying chemical seem less dangerous. The proper term is hydronium hydroxide or hydroxyl acid.
You could really put yourself in harms way by adding a bit of a highly inflammable and volatile liquid with the chemical formula C6H6O which is known to have serious psychoactive effects on human brains and make them do stupid things but it does guarantee that us ugly buggers get to have sex.
The brother of a dear friend of mine who has since passed away; blew the cornice of the building (of the college dorm he stayed in) completely off while making nitro and waiting for it to cool. Obviously this was back in the early 1940's or he would have been branded a criminal.
Sometimes the best scientists are just normal people who ignore the rules that were put in place by lesser men.
The brother of a dear friend of mine who has since passed away; blew the cornice of the building (of the college dorm he stayed in) completely off while making nitro and waiting for it to cool.
Damn hipster chemists, they had to be blowing up buildings with nitro before it was cool...
More correctly, it's catenating words. That is, the words are connected in a series. "Theregister" is a catenation in the url above. Some exponentially increasing technological trends conflate to form what's called "Moore's law".
Confused about "conflate"?
http://www.cjr.org/language_corner/language_corner_020915.php
It's concatenation, not catenation.
Catenation is a term in chemistry. A url is not language. Running words together as in the article and several comments is simple ineptitude (although I'll grant that a commentor may have had the intention of highlighting the writer's ineptitude).
You've done a good job of highlighting your own ineptitude with misuse of 'catenation' and not knowing 'concatenation'.
What kind of techie are you?
Catenation is also a term in linguistics. It's less frequently used than concatenation, but nevertheless you can catenate words.
Sadly the wikipedia article about the word doesn't include that definition, which is presumably why this isn't so widely known today.
And, for the record, I'm still waiting for you to back up that accusation of plagiarism you made against me. But then I've noticed something of a pattern in your posting recently. Accuse someone of something, declare yourself superior in some way and then bugger off when you're called on it.
re cat vs concat, I learned many years ago that the Unix 'cat' command was short for 'catenate' which is an obscure and/or archaic variant of 'concatenate'. Personally, I have no problem with 'catenate' as a synonym for 'concatenate' (and yes, either is probably what the OP meant instead of 'conflate').
/said in an isn't-it-interesting-that-the-thread-talks-about-both-catenation-and-proper-use-of-hyphens* kind of way
(*no doubt that's a proper word in German, but let's not get distracted)
I think it's a mistake rather than language drift. I could take a two-year sabbatical, and the hyphen is acceptable (and normal) usage there, but I'd be back to work two years later (no hyphen).
edit: I didn't see the later post by J.G.Harston that makes the same point, but uses grammar-type words.
"Moore’s idea not only predicted the development rate of computing power, it set an ambitious pace for all IC manufacturers to maintain"
And that's where its real relevance lies. Moore's Law has for a long time now been the expected rate of change that all chip designers and manufacturers have had to try and keep up with, and as such has become a self-fulfilling prophecy.
Imagine where we'd be now If only he'd tweaked the numbers a bit and "predicted" a tripling or quadrupling or more. Heh heh heh.
Don't you mean Crook's Law? (The crookedness of an ithing after being in one's jeans pocket will double with each iteraiton of said ithing.)
Cooks Law is a fundamental law of physics:
Sod's Law: The experimental result always turns out wrong.
Cooks Law: Raise or the denominator as required the get the expected result.
Sturgeon's Law: Remember to allow for experimental error or the result will appear fishy.
This post has been deleted by its author
In the days when he played with such chemicals it may have been illegal, with some consequences but if he were to do this in the UK today (I do not know about the US where I assume he was) as a youth that would be his entire future probably blighted, in the legal sense, beyond any hope of redemption.
Similarly the modern equivalent, cracking systems, seems to lead (from other comments I have read here in the past) to a very significant proportion of the IT security community vowing never to employ such a person regardless of any real talent* they might have.
Yet among many readers of this article there's probably a combination of admiration and yearning for the things he did in his youth. Certainly I would wish to be able to let my nipper do such things (within certain safety boundaries) but probably even raising the issue with his teachers to try to do so in a responsible way would get me reported for, well, something. Probably terrorism. And child abuse.
I feel it doesn't bode well for the ability of the human race to discover and incubate talent**. It's a shame really.
*Not script kiddies
** I concede that doing a "Sheldon" and building a fusion reactor in the garage might be going a bit far.
This post has been deleted by its author
In the 1960s the corner chemist happily sold schoolboys sulphur, potassium nitrate, strontium nitrate etc - as well as glass tubing to bend, blow, and stretch into various equipment configurations. GCE "O" Level Chemistry placed an emphasis on practical lab work.
These days it is probably impossible to buy even the ingredients for a "chemical garden" from a local source. Although I did see a kit, possibly with the Science Museum branding, in the charity shop - "unused - as new".
Yes, times have changed, and not for the better methinks. I can remember making explosives, as part of O-level chemistry practicals - working with chemials like chlorine, ammonia, (great togeather - the head always put the manufacture of chlorine and ammonia on the same bench for open days) hydrogen sulphide, phosphorus...
Chemistry WAS FUN then. A-level even more so.
MInd you, a colleage at work in the seventies did tell me that questions were asked when he was sent by his firm to the local chemist for choloform and some glass syringes and needles.
Perspex gluing for the enquiring minds out there. Chemist would not supply without a letter from the company.
I was dismayed to find that even the humble borax (laundry aid; or ingredient with PVA glue, for "slime"!) is now impossible to buy in local stores, apparently on the grounds that eating it regularly may shrink your testicles, so it must be treated as a hazard. On the other hand if the fashion for obesity propaganda causes you to doubt your "body image", you can buy dinitrophenol pills on the 'net and poison yourself without any hindrance. Bizarre.
Exponential growth rates have happened before in other industries, and have worked until some physical limit is reached. The examples I was told were:
1) passenger aircraft speeds post WW2, which increased until the sound barrier was hit, went beyond (Concorde), then dropped back to a uniform level below.
2) miles of railway track in Britain in the 1830s-40s, as companies competed to connect the country. If this exponential growth had continued for another 20 years, then the whole of Britain would have been converted to railway track by the 1870s, so the physical limit here was the space to grow.
Moore's Law (/guideline) has worked for a long time though, and beaten some physical "limits", which were worked around.
I recall a talk of his where he also showed a graph of total revenue generated by a fab line vs. wafer size, and a graph of fab cost vs. wafer size. The prediction was that a 15" wafer fab would barely break even, and an 18" fab would never recover its costs.
Moore's law wasn't that there would be exponential growth in computer chip power.
It was that the MOST COST EFFECTIVE feature size would decrease exponentially (or transistor count would increase exp)
So although a new smaller fab process would be more expensive to build and operate all the advantages, more chips/wafer, defects/chip, edge losses, kerff loses, all decrease with the square of the feature size.
With the new 14nm fab it isn't clear that it will ever be cheaper/chip than 22nm and it is even less certain for <10nm. There may be other advantages in power usage and fitting more features into a small phone - but that's not what Moore originally claimed
As pointed out, it was actually saying that the number of devices which could be integrated *at minimum cost* doubled every [x years], not the number which could be integrated. It held all the way up to 28nm and then stopped; even though in 14nm you can get twice as many devices on a chip, the chip cost is more than twice as high.
There's no technical brick wall stopping processes going to 10nm and 7nm (and maybe 5nm or below), but there seems to be an economic one. So in the original sense of Moore's law, it's already dead -- or at least, sleeping until something comes along to drastically drop the cost of advanced processes.
Actually with 28nm -> 14nm you get 4x as many devices.
And if you have a single point flaw which destroys a single device on a wafer, you now have 4x as many devices per wafer so the cost/defect drops to 25%, you then get another few % win because with smaller devices you can get closer to the edges of a circular wafer so don't waste space.
But with more devices you waste more space to allow for the saw cuts between chips (kerf loss)
Process node names like "14nm" and "28nm" are marketing labels nowadays, they have nothing to do with feature sizes or density. We're doing layouts in 14nm right now and the density is just under double (less than twice as many gates per mm2) compared to 28nm. The wafer cost is more than double due to double patterning and limited competition, so the cost per gate is higher -- very few cost-sensitive high-volume devices are big enough for yield to be a problem.
Allowing for wafer costs falling over time -- and lower-cost foundries entering the 28nm market -- the projections show that 28nm will be the cheapest per gate for at least another 3 years, maybe 5 (or more) until EUV brings down the cost of 7nm. This is unprecedented (and probably a one-off), previously processes only held the "I'm cheapest" banner for a couple of years, 28nm looks like holding it for at least 5 years, maybe 7.
Process nodes at different companies are often quite different in their capabilities, even if they share the same nominal scale. Intel's are generally acknowledged to be the best, usually by implementing new technologies not directly related to a simple "shrink" -- e.g. Hi-k metal gate and 3D tri-gate transistors. It's also important to remember that other aspects of the chip, such as interconnect technology, are also important in achieving expected density increases.
" If this exponential growth had continued for another 20 years, then the whole of Britain would have been converted to railway track by the 1870s, so the physical limit here was the space to grow."
The limit was economics. Too many competing lines between the same points. Cambridge station had three different companies' lines terminating there. The companies merged or went bankrupt until only the Big Four were left. Even without the effects of the war they were struggling with out-of-date rolling stock, poor tracks, and uneconomic routes. They were nationalised in 1948 - and Beeching axed many of the routes considered uneconomic in the 1960s.
Some of those closures and land sales are now being considered unfortunate as roads become congested - or where there is no alternative route back-up when major accidents occur.
This post has been deleted by its author
Umm, it's just a english idiom rougly meaning, in this context, something like "seriously" or maybe "professionally" and implying that this was when he got seriously stuck into that business. I think originally it was used to distinguish between practice and "for real". For example "shots fired in anger" meaning you were shooting at someone with intent as opposed to just practicing.
(Here's hoping that was a genuine language misunderstanding and I haven't missed some joke and patronisingly explained it to you anyway!)
Presumably 14nm, 10nm and so forth represent the process size rather than the "die-size". It would be quite entertaining to watch someone attempt to package a 14nm-sized die!
It's also worth noting that as one of the "traitorous eight" who left Shockley to found Fairchild, and then by leaving Fairchild to found Intel, Moore helped establish the Silicon Valley startup ethos.
The die size has been increasing steadily from millimeters the sixties to square centimeters in the teens, and the wafer size keeps increasing as well. Both contribute to the increase of the number of transistors on a die.
But if it's no law, that means I don't go to jail:-)
Bigger wafers just mean that the costs of per-wafer processing steps get amortized over more dies per wafer -- they don't directly allow for bigger dies. Intel premium CPU die sizes hit the square cm size range by the first Pentium 4 fifteen years ago (217 mm^2) and have zigzagged up and down in size since then (see http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/5 ).
Except the part that no real-life process follows exponential law for more than a transitory phase - Moore's law is probably the most enduring example, but it's not exempt of hitting a brick wall at some point either (arguably already has). You simply never get anywhere near the point where "all of land gets covered with railway tracks" and he seems to have a problem grokking that.
Well, number of transistors per unit area goes as the inverse square law of the process dimension. So even if the process dimension is only a linear function of time, the transistor count will be squared. If transistor count is linear, then obviously the process has be less than linear.
Moore's could be defined as: see Scaling Laws.
The obvious way forward is further into the third dimension. If the process can mature both thinner layers and a taller stack, you might get onto the squared growth curve.
The cheapest fix to keeping technology running smoothly would be to convince the coder drones to stop cobbling together bloated bugware. Fix that.
I, for one, avoid facebook like the plague, but they probably have a dossier on me, from morons using it mentioning me, and from my occasionally checking pages there because people don't bother updating their websites with 'what's on' information, instead doing it all through faescesbook and twatter.
Facebook has frequently been said to keep non-user profiles.
...for finally saying it. It's not a law.
Unfortunately, I keep hearing stuff from people who should know better about how Moore's Law drives technology forward. (In particular, I'm thinking of an aging "rocket scientist" and hack SF author who can't help but mention it when discussing technology.)
It's bad enough we have to explain what "theory" means whjen dealing with the evolutionarily challenged.
...for finally saying it. It's not a law.
At least half a dozen commentators feel obliged to point this out in the comments for every Reg article that includes the phrase. "Moore's Law isn't a law" is such a cliché that some people probably have it as a keyboard macro. This piece is hardly groundbreaking.
Your complaint is a nice illustration of the Gell-Mann Amnesia Effect. "Bah, those fools who are not in my area of expertise don't understand my area of expertise!" You might want to consult a biologist on the predominant ursine defecatory loci.
If that means, at Johns Hopkins, what it meant at some other places I know of, then the childhood nitroglycerin stunt may have helped him get in.
As for Chemicals Today, a teenager can still get [redacted] and [redacted], and probably even [redacted] at local shops. A friend truthfully answered his daughter's questions about such stuff (and cautioned her about the importance of small batches). With the usual "Don't tell mom", of course.
1997 P200 MMX .. Virge S 2MB graphics chip Win95 OS/R 2 32-bit .. 16MB EDO / SIMM .. 2GB IDE HDD
current .. i7 K 8 threads .. 2GB nVidia .. 1920 cores .. Win7 64-bit .. 128 GB SSD + 2 TB SATA3 HDD
what increase in crunching thru bits per second would that be ?
DRAM speed increases ? increase of clocks per frequency cycle ?
the effects of better instruction sets in CPU .. GPU ?
how much faster .. how much more work *holistically* do PCs do year by year ?
what is the rate of increase currently in ARM and SoC processors ?
Moore's Law has currently hit its economic limits, which is what was originally stated. No problem carrying on to 7nm and 5nm technically, the problem is financial. Current estimate is that to justify the design and manufacturing costs a 10nm chip needs to sell at least $500M, this will probably be $1B for 7nm -- more in both cases if it's a big complex chip.
I've always viewed it as a self fulfilling prophecy, one that in many ways has become a rod to beat Intel's own back with.
It was an observation that became a mantra for Intel and drove IC development just so they could keep from breaking Moore's "Law". That's not to say we haven't been rewarded from the forced pace of IC development, but that doesn't mean it is a Law.
Bigger wafers do reduce the cost per die (maybe about 30%?) because a significant part of processing (not all) is per-wafer and costs roughly the same for a 450mm wafer as a 300mm one. The problem is that the cost to the industry of bringing up 450mm would be enormous, and nobody wants to pick up the tab -- especially the equipment manufacturers who got royally screwed by the transition to 300mm.
So yet again, it's all about the money not the technology :-)
Here's a picture showing the real problem:
https://www.semiwiki.com/forum/attachments/content/attachments/8601d1374266204-euv4-jpg
The whole semiconductor industry -- and the electronics industry built on it -- has worked for the last fifty years on the principle that the next process would give you chips which were faster, lower power, *and* cheaper, so the next generation product gets more bang for the buck.
Now this cost model is broken, there's a very nasty wake-up call on the way -- yes you can have more bang, but it'll cost you more bucks.
When I was finishing off my Physics and Computer Science degree in 1994 and making very simple ICs at my Uni's fabrication centre, I remember a physics lecture where I was shown how, because of the expected quantum tunnelling effects - transistors on silicon would never get smaller than 24nm (and would would take 30 years to develop the techniques to get down to this size) - I can smile about it now...
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
