Is Moore's Law near its end?

The copper mean free path is ~40 nm. So if you have a copper via or trench of ~20 nm width, 40 nm height, it starts deviating from the expected bulk copper behavior. The resistivity effectively increases exponentially. The timing just happens to be right with the escalation of fabrication costs.

Nice Article.I like the way you conducted it.Great Work. Best Regards, yourlegalassist.com

Moore law will hold for few years more (say 4-8years, economic situation permitting) and then due to constraints related to atom size developers will have introduce 3D chips. However this approach also have its limit contrary to few suggestions of my predecessors. There will be increasing troubles with waste heat dissipation in 3D system so there is a limit on number of layers of semiconductor possible to incorporate in such 3D system. If we manage to move to diamond based semiconductor system (diamond is best known conductor of heat as well as it can be made to work as semiconductor), then Moore Law will get additional decade of life. So I would say that by 2014-2020 we will hit economic limit for mass production and about 10 years later we will hit physical limit related to size of atoms and heat conductivity of known materials available to man.

Nice article you got here. It would be great to read a bit more about that matter. Thnx for posting this data.

It is amazing Moore's Law has lasted so long, through so many changes. It is also amazing we happily keep calling it a "Law", yet it is not like the real laws of science. Yet the greatest importance must be that the widespread acceptance of its basic predictions are (to some extent) self-fulfilling... that is: manufacturers commit themselves to introducing new technology - smaller devices - that will loose money to start with, because they know that, if they don't, somebody will go ahead anyway and those that are shy will be left behind and go under. If the general expectation was that little improvement will be had by more R & D, they would make more dosh by sticking with technology they have and letting prices climb. Expectations are everything!

Moore's Law has always really been a logistic curve phenomena because it's really been an aggregation of multiple generations of technology adoption curves which themselves formed macro-technology adoption curves for a class of products, and so on. "Turtles all the way up" with apologies to Douglas Hofstadter. Further, the factors for scaling that allowed Moore's Law are quite delicate. I worked for Affymetrix, a genomics company that uses the same photolithography for patterning DNA/RNA hybridization combinatorial chemistry. They never have and likely never will have a "Moore's Law" with the technology simply because a handful of critical economic factors are just a bit different and prevent the market from having the "scale-free growth" of Moore's Law. There are some extension strategies for semiconductor microelectronics scaling possible: nanoelectronics and photonics. But these will be and will have some pretty radical differences from anything in microelectronics. For example, device physics equations are still quasi-classical today with just a dash of quantum mechanics stirred up which is then baked into a lumped equivalent model of familiar circuit analysis. This won't work for nanoelectronics. Schoedinger's is the replacement for the basic lumped model Ohm's Law. There is a closer resemblance to microwave distributed model circuits than to lumped model circuits. And the concept of digital logic is a "lumped model grafted onto a lumped model" so that will need to be re-interpreted somehow. That's a big leap. Photonics will require some not-currently-existent technologies to bridge back into legacy microelectronics, such as compound semiconductor optical efficiencies in either silicon or mated to silicon. For both nanoelectronics and photonics, the current microelectronics technology will become their "bond pads". So there's still going to be major role for existing silicon. What will happen though is what always happens as you pass over the peak of the technology adoption curve: you start economizing and you start to permute uses of a common platform. This will occur even more than simply having different functional designs using the same foundry platform process - it will involve aggregation of multiple technologies (like micro + nano or micro + photonics or many other possibilities).

Relying on 3D to pack even more circuits in is missing the point -- we can already get so many circuits running so fast onto a 2D chip that the resulting power consumption is unacceptable, and have been able to since the Pentium 4. To make use of higher levels of integration -- either with further geometry shrinks, or new materials, or 3D circuits, or unobtanium -- we have to find ways of designing more power-efficient chips, not faster or more complex ones. This means not only new low-power circuits and libraries (ultra-low voltage? resonant clocks? asynchronous? data-driven? bootstrapped wells?) but new design techniques and tools are needed. For example, why when most of the power is dissipated in interconnect do we (and the tools) still design and optimise circuits first and then try and connect them up -- surely we should be designing the data flow/interconnect first and then just shovelling the required circuits in underneath? There's always a need for more processing power (not just in CPU cores), especially in communications to cram more bits/Hz into each channel, which is why every new radio system is 5x-10x more complex than the preceding one, and 100Gbit systems for the Intermet backbone are similar compared to 10Gbit. But to be useful you've got to deliver this increased processing power at the same power budget as the preceding system, and this is the real challenge for the future.

Mike Bruzzone's authentic frontier gibberish is pure poetry!

First off you don't have to keep reducing process size to keep making more powerfull chips, one option already in use is 3d chips, each core could be on its own layer and we could still see, more and more processing power on a chip, as to fab costs, first off if they weren't makeing billions upon billions on these fabs they wouldn't keep building new ones, and next is a fabs usefull life is not just at the original plant, but to the other plants it is sold too, many a fab is used for many many years in other chip production, consider all the other devices that will be made latter on those very expencive peices of equipment, such as poach chips, logic, analog, and then finaly transitors and opto couplers, etc. One of the things that will slow processor development is the fact that sooner or latter they will be so fast that what we are using them for will be basicly done as fast as it needs to be, processor speed is getting so fast now that with a top end video card games which have been the major driver of the need for speed in home machines, will be as fast as a person can respond, and graphics levels will be at the point ware better graphics are undecernable by the human eye, as far as buisness use computers are much more powerfull allready than most buisness uses, I mean realy, how fast does a machine realy need to be to do a speadsheet or write a letter or even make up a flyer, sure there are still aps that will still take advantage of faster chips but they are getting to be fewer and fewer, Oh I don't see and end to processors getting faster for a while still, but sooner or latter there will be a time that chips can get faster but no one will be clamoring for them. As to microsloth, well I think thier days of adding more and more rubbish to the operating system so it wastes more and more processing power to do basicly the same job are neering an end.

Am surprised to note the 90nm down already in 2009, I was expecting a shift to 2010 or 2011, due to the financial crisis that surprisely got all people of our market. Assuming the curve of 90nm path is correct to the assunmption of the analyst, I would believe that there is a shift for at least 1 year and have the min flat of 90nm to 2011 Q2 to the volume increase between Q3-Q4 2010 of 65nm.

Marketing spins another web: Words outpacing thought.

Please stop using Moore's Law and start using Moore's View. It's just a Mr. Moore's personal view. It isn't a scientific theory or law at all! I am a EDN reader for more than twenty years and strongly hope EDN can take the leadership and start correcting it.

No one seems to have mentioned ECONOMIES OF SCALE. Moore's Law can only continue if there is a good ROI. Current 45nm technology is very expensive and only if a large volume of chips are made can the costs be justified. Currently we are in a highly 'custom' market so the economies of scale generally do not apply (unless you are Intel or the like). We have started to plateau out.

I still use a 486! Who's moore?

It sounds natural that Moore's Law will retire for semiconductor technology as Dr. Moore retired from the semiconductor industry. Thank you Dr. Moore!

Moore's Law predicted an increase in transistors per die. Transistor shrinkage was the proximate cause. At each turn of the 18 month lithography crank, you got twice as many transistors per unit area. But you also got higher speed and lower power. Shrinking the die was good in several dimensions. This drove Silicon Valley for 35 years. Moore's Law went into the ICU in 2004. The CTO of IBM wrote the epitaph, "Somewhere between 130 and 90 nanometers, we lost scalability." This means things stopped getting faster and cooler as you shrank. As you shrink further, the die gets smaller, but it also gets slower and hotter. The evidence? In ~2004, the CPUs stopped getting faster. We still do not have a 4GHz Pendium. Nor will we any time soon, barring making chips from Unobtaniun. As one person put it, at 130 nanometers, 75% of the power and 75% of the propagation delay is in the metal, not the transistors. It gets relatively worse with each shrink. And there are no scalable solutions in sight. There are lots of one-trick-pony solutions, such as copper metal, halfnium oxide gates, etc. These help prop up the speed for the current node, but they are independent of the lithography = they are not scalable. This means that the next shrink will kill their one-time advantage. Many straws are being reached for such as carbon nanotubes and room temperature superconductors, etc. to replace the metal, but do not look for change soon. So somewhere around 22 nanometers or so, we should start looking for a burial plot for Moore's Law. In the mean time, we have bottomed out at an economic trade-off point: High speed, low power, or small die: Pick any two. Silicon is on its way to becoming just another business, like electric motors or appliances. Sad, but even the Wild West finally gave way to farms.

Could it be that we are also slowly nearing the end of the demand for ever more power? The netbooks may just be the start of a new era, being satisfied with a stable OS and no improvements in processors. Why would I need ever more power for Word or Excel? Yes, semiconductors will be ever more ubiquitious, but probably not with ever more power, at least for most of mainstream applications. Probably does not apply for memory

Too bad Isaac Asimov is not here to write his typical futuristic observations. I can still recall a short story he wrote back in the 1970's about the progression of computer technology, from knife switches all the way to sub atomic decision gates.

It is not a question THAT Moore's law will eventually end. The question is WHEN it will end. I would be surprised if it came to an abrupt end in 2014. I would rather see its significance fade over one or two decades- What we currently see may be the beginning of this, or just a brief pause. This depends on how the equipment industry reacts to the cost challenge. On the test side of things, we are seeing trends towards reduced cost-of-test, for example. Then, at some point, new technology may revive Moore's law, like 50 years ago the transistor obsoleted all ideas about how and where electron tubes could go.

Last month at an event at The Tech museum in San Jose, Gordon Moore said that there's about 5 years left which pretty much matches what iSuppli said. I remember hearing then Intel CFO Andy Bryant talk about how Moore's Law was just as much about economics as technology. It's been fundamental to the business model of Intel. In Only the Paranoid Survive, Andy Grove said that Intel was built on Moore's Law, even before Intel became a processor company. I'm sure Intel has some tricks up its sleeve. But I'm surprised this story has not gotten more attention beyond your great publication. It's just as much about the viability of the business model as it is about the technology.

It could come back if the infrastructure is significantly retooled. For example, the ATE industry is still focused on cycle-based design, which went out the window over a decade ago. It's not difficult to see that a paradigm shift has to occur in that side of the business to significantly bring down the cost of design. On the other side, EDA is still trying to wring out the drop of milk in a 10-year-old cash cow. There hasn't been a significant breakthrough there either. Since it now costs upwards of $50M to get to tools you need to design a new SoC, you can't find much profit left in that business. We've passed the point that innovation in both those areas could help. We need invention.

I believe Moore?s Law will remain effective as a forecasting tool for logic designs even if some cost offset has to be applied. However the relevance of Moore?s Law to memory technologies has always been a bit fuzzy, since the complexity of memory technologies increases as a result of reducing the area in which the energy is stored. Let?s also not forget that the other driving force of Moore?s Law was consumers who saw enough value in the new products to support those investments. Is this perceived cost-related challenge to Moore?s law really due to the increasing cost of fabs, or is the scarcity of consumers willing to fund those fabs a sign of the approaching maturity of the current semiconductor technologies?

In recent years, Gordon was more concerned about economics than technology. Lithography costs have always been central to IC economics. Clearly, double patterning and EUV (IF it ever overcomes its numerous technological hurdles) are far to expensive for NAND flash, and probably DRAM. Only high margin MPUs can afford 193i and x-ray lithography. Memory will clearly migrate to imprint lithography because of dramatic CoO benefits. Imprint has always had the resolution capability, the defect and overlay problems are getting resolved, and replication solves the 1X mask problem, turning lemons into lemonade. Memory manufacturers are frantically working behind the scenes to redirect resources from EUV to imprint.

If there is actually any value to be had by computing faster, why not try something really radical such as writing more efficient code? I still recall a 4MHZ 6809 processor testing machine that would routinely be 5 to ten times faster than the(then) state of the art 486 machines, which ran a microsoft brand OS. Yes, writing in machine code was a bit harder, BUT it was more flexible and it ran faster and it was also FAR MORE reliable. Aside from all that, while geometry may shrink and take voltages down with it, ambient noise is not getting any smaller. The next step will be digital circuits triggering themselves on supply current generated noise, bringing on the era of really dismal yields. And one more thing, which is don't forget that all of this wonderful very low voltage stuff still needs to somehow connect to the(noisy) real world. Consider dealing with the 0.25 volt, 35 amp, super low voltage power traces.

If this is the problem "semiconductor manufacturing tools are too expensive to depreciate with volume production, ie, their costs will be so high, that the value of their lifetime productivity can never justify it.? what would happen if one could print the requisite images at 1/10th the cost of the EUV plan and did not require a track for each litho tool??? Would we extend the law another decade?? We need to quickly find out just how real step and flash imprint lithography might be.

You pose the question and then along comes spintronics. This new bismuth telluride based technology has superconducter properties at room temperature and is expected to revolutionize all electronics in the near future.

If transistor count can't be doubled by reducing geometries further on a 2D semiconductor die, new transistors will have to be stacked up vertically. In theory it's possible to scale up by introducing vertically stacked up dies. Given that most of the power dissipation occurs due to leakage current that increases exponentially as the transistor geometry decreases, it would even make sense to step back to a cheaper and older 90nm (or larger)process geometry to reduce leakage current and heat before attempting to stack/grow multiple dies onto a large 3D volumetric semiconductor device.

who cares? Moore's Law never drove any decision in any strategic technical or business meeting I ever was in. It seems its biggest value was as a soporific for conference keynote speakers to calm the audience.

Moores law (or at least the fundamental basis for it, namely scaling) ended a few years ago. Just check out transistor performance - it's not improving, in fact they are getting slower. It's only density that is helping, by putting more functionality on a chip. This helps system performance, particularly in big systems e.g. mainframes. So the market for bleeding edge chips is going to shrink, not expand, relative to total semiconductor sales.

It is getting way to expensive by shinking the technology.Will it be necessary?

I HAVE BEEN WORKING IN THIS SEMICONDUCTOR TECHNOLOGY FIELD SINCE 1984. IT LOOKS MOORES LAW (OBSERVATION) SATURATION IS NOT FAR AWAY. TIME WILL TELL .

Maybe the limits of semiconductors will make researchers look to a different paradigm, use light or microwaves to store and transmit information. Maybe we need to make another leap to a technology that is as unexplored as the transistor and printed circuit were in the days of vacuum tubes. We have to keep looking over the next horizon!

This really applies to Shockley's technology of 1949. We must move on to new technologies and new ideas of capatalization. Markets and overall costs will be the determining factors.

Hi I was at the Solid State Circuit Conference in Philadelphia in 1965 when Gordon Moore presented his observations. He predicted that in a few years the cost of memories for a computer would be 0.01 cent per bit. Someone in the audience commented after the talk that it would be impossible to thread cores for memories for that amount of money. Gordon's answer was: "I guess we are not going to use memory-cores in the future then" Is his comment then not valid mow? Maybe we are not going to use transistors of the type we use now in the future. They are going to be too costly to manufacture. Hans J Weedon.

I think that 32 nm is the economical limit for semiconductor technologies. After that the problem will be how to optimize the design. Also today one of the current problems is layout optimizations. So the limit of the technology will be reached in short term. Then the company's will have plenty of time to optimize their design to the limit their chips and to make their cost as small as possible. I think that the company that has the most benefit will be the one with the most patents. The goal will be then to optimize software and to use as few resources as possible. Even country's like Romania will have the possibility to make semiconductor chips without having to fear that the entire lot of chips will be inadequate.

So when do we reach the point where we can't keep up with the speed of our computers? Sure, there are specific uses that require a lot of computing speed, but won't we cluster and accomplish our desired output at least for the nonce? I'm pushing for bio because of the inherent speed. The human mind is astonishing in more ways than one. Unfortunately.

Semi's continue their slump for a good reason. Leading edge process nodes accelerated ahead by a few now exist on the trailing edge of what was the CMOS average cost curve. Even before .90 nanometers CMOS derivatives at submicron lithographies began their transit back from a commercial fabrication art, toward applied science and now applied research. Financially aiming for upside today's 45 nanometer & finer geometry designs, fabrication and test 'input costs' are higher. Output volumes are staggering and price per die often less. Demand is not making up for necessity of making an accounting profit greater than supplier's economic costs. An inflationary spiral in input costs is now opposed to voluminous deflation in sales values. In this sort of innovation phase at the trailing edge of the process average cost curve, for components, die size economics has become a fallacy. For an industry constantly productizing, to limit and even fly over fabrication equipment commercialization phases, causes commercialization to be pushed out infinitely toward the end of the industries process life; where all geometries of process can now saturate. And this is always how a handful of supply regimes have played the game. Semiconductors is a supply side business. Demand is primarily an after effect of the supply equation. To say supply & demand in the same breath for components is a misconception because it places emphasis on demand, when in fact logical demand drivers are all within the realm of supply. Fail on any supply driver and the demand drivers suffer. So yes we are nearing a semi bottom as the .32 nanometer process node passes. This frees all prior nodes for commodity component fabrication from legacy equipment; if you can find the parts. Component?s design fabrication is typified by high cost and margin at the leading edge of a process generation's cost curve, known costs and known margins at mid phase, high cost and low margins at trailing edge. Once over this current innovation rise in cost and past the innovation hump to molecular electronics, this process cost and margin wave form will begin again as it has always to benefit entire industries and economies. So for now we can put behind us all those majestic deductive systems of philosophy proposed to draw scientific laws out of a few axioms and principles. There is no magic hat in science. Every thing taken from the hat in works must first be put into it by observation or experiment. And not by mere casual observation, nor by simple enumeration of data, but by sought after experiment. Mike Bruzzzone Camp Marketing

Failure to follow Moorse curve does not mean slowdown in electronic industry. The industry will seek solution through other means to meet the demands from markets. The limiting factor is the physical size of atoms. Soon or later, you will hit that wall. New approach is needed.

Several comments suggest we're witnessing the structural slow-down of the electronics industry generally. I'm not so sure. Not only are the massive middle classes of China, India, Indonesia and Mexico about to become major consumers of electronic products, electronic components are found in more and more products all the time. Cars today hold 3x the value of components of cars 6 years ago. This doubles the size of the electronics-buying world and have them purchase more per capita! Constant innovation and churn in the developed economies will drive sustained component growth (not just replacement rates). Seen together we're about to see a major growth spurt and sustained volatility as the industry explores new markets and develops new products for entirely different customer types. Buckle up everyone!

The article said something very true. For example - DRAM industry had the biggest market ever in 1995, around $45 billion. At that time, a typical fab costed $500 million. Today, DRAM market is less than $20 billion and yet it cost $3 billion to build a fab. A wet scanner alone can cost $55 m. So, the math does not add up. It is true that DRAM is an extreme case, not the typical case. But even consider the overall semiconductor market, $150 b vs $250 b now, the growth rate of the market is no where near the growth rate of the fab cost. I remember 20 years ago, there is another analysis which showed the growth of electronics industry at a rate of over 10% a year while the growth of overall economy is only 4% a year. If the trend continues, by 2020, the electronics industry will be bigger than the overall economy. This cannot be true logically. So, somewhere something has to give. It may not be the technology shrink, as predicted by Moore law. It most likely be the economics. I believe that the technology progress will not only not slowing down, but it will speed up. But it will take a different direction, not by shrinking devices by brutal force but by building devices from molecular level and up using self assembling process.

If it truly was a "law" then it could not be limited by time and inovation. OHMS law Kirschoff voltage LAW, and several others can never go away...this is all just some hyped nonsense by a salesman. Even the incredible shrinking man came to an end. So did teh Roman empire, BUT I will ALWAYS = E/R. No matter what year it is, so Moore's ain't no law.

While physical sizes may be limited, going horizontal will make package sizes still appear to follow Moore's Law for a considerable time yet, particularly after shrinkage of size limits until we move to optical processors or other developed technology out about 2025 to give us a new paradigm to follow. Limiting/controlling heat dissipation seems to be a major factor in how much component slices of such 3D processors can be sandwiched together, multiplexing of "goesintos" and "comesoutas" may give a more effective "pin out" of these super processors.

This is an argument that has been denied time and time again by corporate executives. Of course Moore's Law is running out of turns of the crank. Design for manufacturing, (DFM) engineering teams have validated this point for several years. Engineers have been innovating using more and more of the periodic table to counter the effects of scaling and introducing Hi-K dielectrics to counter the creeping leakage issues. There is nothing left in the bag of tricks now! Yields keep deteriorating as dimensions are reduced. Scaling as a strategy is dead and has been dead for several years. It just didn't know enough to lay down!

I made the same prediction in 1994. I predicted that by 2015 feature size will be hard if not impossible to decrease. Silicon based digital ICs will start to level at a feature size ~ 01 microns (10 nm) and performance improvements will be hard to increase. Architecture and clever design rather than feature size improvements will take over with modest performance improvements in the following 5-10 years! Electronics business will levele off because people will find little reason to buy a new PC or a new PDA. We have to get ready to to live with much lower growth rates in the Electronics Business & Computers? New Energy adventures would become more attractive than electronics ventures?

I believe that MoE is correct. I've been expecting the biological breakthrough for years, but the semi companies seem to think they can keep shrinking and not realizing that there are increasing noise factors with shrinking geometries. It's either going to be optical or biological or both.

Moore's law was a self-fulfilling prophecy. The law was proclaimed based on an observation, then the whole industry worked towards it becasue the feeling was that everybody was going to have to obey or get lost. One could even argue that it held back the technology becasue it was the benchmark to meet, ie make a 30% reduction in pitch, you get a doubling of the circuits per area and no reason to be more aggressive, since that's what everybody else will do. Of course it will run out of steam. It may have already, or should have a long time ago if the investment hadn't been so conservative. Atoms are only so small and you can't have features smaller than an atom. Future technologies may be based on future non-semiconductors and a new Moores law will be established.

It was an observation, not a "law". So someday it will fade away. With rapid development of organic semiconductor technologies and others, the life cycle could be even shortened. You could walk into a print shop next to a grocery store to make your customized phone, at the same time print your pictures. Of course, this will take awhile, maybe 20 years? Most of us will still be around then.

Innovative design and better software will yield the performance improvements that consumers (and manufacturers) have come to anticipate from technology node advancements.

Moore's law may be in a swamp state based on today's technology, yet advances in other candidate technologies (i.e., all-optical, biological), driven strongly by market demand, will set the stage for a new or continuation of the Moore's law in electronic devices as we know it.

Will Moore's law continue? The most reliable answer will come with time. Let the market forces and the bean counters decide.

No exponential will last forever. You will reach the atomic dimensions and scaling will stop abruptly. Even before you reach that scale you will be experiencing difficulties that will be mounting exponentially. Moore's law will end is a forgone conclusion.