Moore’s law "really" is dead this time

[the.fire.bird]

The chip industry is no longer going to treat Gordon Moore's law as the target to aim for.

 

Moore's law is the observation that the number of transistors in a dense integrated circuit doubles approximately every two years. The observation is named after Gordon E. Moore, the co-founder of Intel and Fairchild Semiconductor, whose 1965 paper described a doubling every year in the number of components per integrated circuit,[2] and projected this rate of growth would continue for at least another decade.In 1975 looking forward to the next decade, he revised the forecast to doubling every two years. ^Wikipedia

 

Quote

ARS TECHNICA:

Moore's law has died at the age of 51 after an extended illness.

Problems with the original formulation of Moore's law became apparent at an early date. In 1975, with more empirical data available, Gordon Moore himself updated the law to have a doubling time of 24 months rather than the initial 12. Still, for three decades, simple geometric scaling—just making everything on a chip smaller—enabled steady shrinks and conformed with Moore's prediction.

In the 2000s, it was clear that this geometric scaling was at an end, but various technical measures were devised to keep pace of the Moore's law curves. At 90nm, strained silicon was introduced; at 45nm, new materials to increase the capacitance of each transistor layered on the silicon were introduced. At 22nm, tri-gate transistors maintained the scaling.

The photo-lithography process used to transfer the chip patterns to the silicon wafer has been under considerable pressure: currently, light with a 193nanometre wavelength is used to create chips with features just 14nanometres.

Technology may provide ways to further increase the number of transistors packed into a chip, but the manufacturing facilities to build these chips may beprohibitively expensive-a situation compounded by the growing use of smaller,cheaper processors.

What we see instead are multiple processor cores within a single chip.

Rather than focus on the technology used in the chips, the new roadmap will take an approach it describes as "More than Moore." The growth of smart phones and Internet of Things, for example, means that a diverse array of sensors and low power processors are now of great importance to chip companies.

The highly integrated chips used in these devices mean that it's desirable tobuild processors that aren't just logic and cache, but which also include RAM,power regulation, analog components for GPS, cellular, and Wi-Fi radios, oreven microelectromechanical components such as gyroscopes andaccelerometers.

While a lesser priority, scaling is not off the roadmap entirely. Beyond tri-gate transistors, perhaps around 2020, are "gate all around" transistors and nano wires. The mid-2020s could bring monolithic 3D chips, where a single piece of silicon has multiple layers of components that are built up on a single die.

As for the future, massive scaling isn't off the cards completely. The use of alternative materials, different quantum effects, or even more exotic techniques such as superconducting may provide a way to bring back the easy scaling that was enjoyed for decades, or even the more complex scaling of the last fifteen years. A big enough boost could even reinvigorate the demand for processors that are just plain faster, rather than smaller or lower power.

But for now, lawbreaking is going to be the new normal. Moore's law's time as a guide of what will come next, and as a rule to be followed, is at an end.

 

 

SOURCE: ARS TECHNICA^UK

[rockking1379]

All good things must come to an end some time

[2FA]

I'm just going to point out that the scale on that graph is absolutely atrocious.

[Watermelon Guy]

Just now, DeadEyePsycho said:

I'm just going to point out that the scale on that graph is absolutely atrocious.

Kind of, its not meant to be exact, only for demonstration purposes. 

[2FA]

Just now, Watermelon Guy said:

Kind of, its not meant to be exact, only for demonstration purposes. 

It starts at 2300, then to 10000 where it starts to by exponential increases until 1 billion where the next is 2.6 billion. Anyone who has taken a Stats course will know that scales should always increase by the same amount.

[Watermelon Guy]

1 hour ago, DeadEyePsycho said:

It starts at 2300, then to 10000 where it starts to by exponential increases until 1 billion where the next is 2.6 billion. Anyone who has taken a Stats course will know that scales should always increase by the same amount.

Unless its not practical. .  But you cant have a real scale being accurate at that size where the data ranges from 2,300 to 2600 Million 

[christianled59]

2 hours ago, DeadEyePsycho said:

It starts at 2300, then to 10000 where it starts to by exponential increases until 1 billion where the next is 2.6 billion. Anyone who has taken a Stats course will know that scales should always increase by the same amount.

If you look at other charts, the line is exponential. This was to make it a linear line but keep the same data.

[Doobeedoo]

Really can't wait to see the new type of processors in future. With new materials and techniques being made off it will be exciting to see it once available.

So in time we may come to point where processor is so small and dense just being operating with like per atom or layer and photonic operations.

[Now Thats Damp]

Yes, this is unfortunate, but what is the real cost that Moore's law doesn't hold true anymore?

It almost feels analogous to the end of the GHz race, only now we're coming to the end of the transistor race. Meaning, are we now going to find that transistors per cm²  isn't necessarily the most important thing regarding performance, just as AMD made the world realize that frequency wasn't necessarily the only way to increase processor performance?

 

Take the 780ti vs 980ti, for example. The 780ti has 64 more Cuda cores, is on the same manufacturing process, same memory technology, same fundamental characteristics. Yet the Maxwell manages, even with the same manufacturing process and fewer cores, to achieve ~ 40% greater performance using the same amount of power. They didn't need to smash more transistors in to achieve such a jump in performance, they simply improved the circuits themselves mathematically, made the circuits more efficient, and got the processing power out that way.

 

My point is that maybe Moore's law is coming to an end; just as we hit the practical limit of core speed in the early 2000's and had to search for new methods of improving performance. However, just because the transistors aren't being herded like cattle anymore doesn't mean we can't still observe the doubling in performance every 18 months we have gotten so used to. I trust the industry, and they did because at the end of the GHz race, to find new avenues of improvement. Perhaps 10 years from now we will call these days the end of the transistor race.

[BuckGup]

What happens when we hit 1nm?

[VinsinityKT]

It started to slow down because of the physical limitations of the material silicon.

[vanished]

9 hours ago, Watermelon Guy said:

Kind of, its not meant to be exact, only for demonstration purposes. 

Just like this right?

[Now Thats Damp]

Lolz

[Now Thats Damp]

9 minutes ago, SamStrecker said:

What happens when we hit 1nm?

Perhaps we start forming transistors out of subatomic particles? 

Or perhaps quantum computing will have taken over at that point?

Or some new route will be found that we haven't thought of yet?

Who knows? I certainly don't. And I don't think anyone does. That's what makes technology so fun; you never know what the next improvement will be.

[Griffscavern]

18 minutes ago, SamStrecker said:

What happens when we hit 1nm?

 

8 minutes ago, Now Thats Damp said:

Perhaps we start forming transistors out of subatomic particles? 

Or perhaps quantum computing will have taken over at that point?

Or some new route will be found that we haven't thought of yet?

Who knows? I certainly don't. And I don't think anyone does. That's what makes technology so fun; you never know what the next improvement will be.

 

 

This may be what's next.

 

 

http://www.pcworld.com/article/3018382/hardware/revolutionary-light-based-photonic-processor-could-lead-to-ultra-fast-data-transfers.html

[Briggsy]

It's now in the hands of software engineers to better utilize more smaller cores, like it should have been to begin with. The race to have the biggest core with the best IPC is basically dead, ironically at the same time AMD decides to start making strong cores again...

[Now Thats Damp]

1 minute ago, Griffscavern said:

 

 

 

This may be what's next.

 

 

http://www.pcworld.com/article/3018382/hardware/revolutionary-light-based-photonic-processor-could-lead-to-ultra-fast-data-transfers.html

Hm, so it seems that it, basically, replaces the electrons with photons.

 

Very interesting 

[vanished]

25 minutes ago, SamStrecker said:

What happens when we hit 1nm?

Well we won't be hitting it on silicon, but perhaps some other technology/material can.  And if it can, it will have it's own limit (just like silicon has ~7 nm), so 1 nm may mean nothing.

[the.fire.bird]

Should I change the graph? Got it from Wikipeada.

[vanished]

Just now, the.fire.bird said:

Should I change the graph? Got it from Wikipeada.

I think it's fine.  It's a little odd that the first and last numbers of both axis were filled in to the exact value, but from what I can tell the scale is actually fine, it just looks weird at first glance because of that.

[LukeTim]

14 hours ago, DeadEyePsycho said:

It starts at 2300, then to 10000 where it starts to by exponential increases until 1 billion where the next is 2.6 billion. Anyone who has taken a Stats course will know that scales should always increase by the same amount.

 

Or, in the case of a logarithmic scale like this one, the same order of magnitude.

[Lautaro]

14 hours ago, Doobeedoo said:

Really can't wait to see the new type of processors in future. With new materials and techniques being made off it will be exciting to see it once available.

So in time we may come to point where processor is so small and dense just being operating with like per atom or layer and photonic operations.

Quantum computing

[Coaxialgamer]

21 hours ago, the.fire.bird said:

The chip industry is no longer going to treat Gordon Moore's law as the target to aim for.

 

Moore's law is the observation that the number of transistors in a dense integrated circuit doubles approximately every two years. The observation is named after Gordon E. Moore, the co-founder of Intel and Fairchild Semiconductor, whose 1965 paper described a doubling every year in the number of components per integrated circuit,[2] and projected this rate of growth would continue for at least another decade.In 1975 looking forward to the next decade, he revised the forecast to doubling every two years. ^Wikipedia

 

 

SOURCE: ARS TECHNICA^UK

Transistors size improvements are slowing down,  bit once we get to the limit,  we can just atack them like 3D nand! 

[Coaxialgamer]

2 hours ago, Lautaro said:

Quantum computing

I would rather see ternary computing,  because quantum computers can't really help normal use.

[Coaxialgamer]

9 hours ago, SamStrecker said:

What happens when we hit 1nm?

We probably never will.  The limit of standard silicon is around 7nn, and you can maybe go down to 5nm with EUV lithography,  but unless we don't use entirely different materials,  we wont get there.