I have a feeling 5nm wont go as smoothly as there timelines predict. Like what Intel did with tweaking 14nm fin heights etc I expect we will see a few years of 7nm tweaks. I could be wrong but we are really close to physical limits.

Intel is a fool on its own, if they had problems doesn't mean others will too.

TSMC has already leapfrogged Samsung...now they're going for the triple jump. YAY! TSMC is running lean and hungry. it is not complacent at all and understands its very survival entails staying ahead of all other chip fabs. the underlying story to all of this is that they worked with clients, garnering contracts with up-front money (AMD/Apple/Qualcomm) to pay for the capital of the new process(es). smart every way around. meantime Samsung... with as much money as you can throw at something, lags behind. obviously, TSMC got the better of their past collaboration.

All those fancy implanted cyberpunk gadgets we see in movies, as well as walking robots and such... will not happen with today's power hungry 95W CPU's or hundreds of Watts GPGPU's ... We need something Sci-Fi'ish (like carbon-based 1nm - or less) chips that only sip energy and can do realtime video/audio/sensors processing and analysis and make instant decisions without completely draining the robot's battery. Otherwise there's very little chance of the future in our movies will become reality. So I'm glad foundries are pushing it.. 7, 5, 3... let's get to those sub-nm transistors quick !

-Tj-:

Intel is a fool on its own, if they had problems doesn't mean others will too.

Intel has so many problems because they tried to be too ambitious. Their 10nm process is more then just a shrink, and more advanced then the competitions 7nm process - well, if it would work, that is. Hopefully they can fix it by 2019, because we all need a competitive market, which doesn't exist if Intel goes stale for a few years.

Moore's law NOT ok then!!!?????

Seems like a fast progression here! They're already talking about 3nm briefly there, and if I remember rightly it's unlikely we can ever produce silicon chips smaller than that - looks like we can easily see the end of silicon chip shrinkage here already, they will need to come up with another material or 'technology' to make further advancements, the end is nigh! By the way, I'm not an expert on this, as you can probably tell from this post!

Yep..... seems not relying on CPU but on GPUs.... Some guru experts....will "explain"...sooner or later....

wavetrex:

All those fancy implanted cyberpunk gadgets we see in movies, as well as walking robots and such... will not happen with today's power hungry 95W CPU's or hundreds of Watts GPGPU's ... We need something Sci-Fi'ish (like carbon-based 1nm - or less) chips that only sip energy and can do realtime video/audio/sensors processing and analysis and make instant decisions without completely draining the robot's battery. Otherwise there's very little chance of the future in our movies will become reality. So I'm glad foundries are pushing it.. 7, 5, 3... let's get to those sub-nm transistors quick !

Solar energy gathering tattoo. Flexible Carbon tube capacitors. Then 7nm is sufficiently energy efficient. It's not like there are 100W CPUs in cellphones, yet there were ~54nm ones. All those manufacturing processes are made to perform in certain way under certain scenario. Take your 7nm cellphone CPU, downclock it, and you have something eating 0.01W. (Actually problematic part of power consumption of cellphone is wifi and cell-network. At least in stand-by mode.)

Fox2232:

Solar energy gathering tattoo. Flexible Carbon tube capacitors. Then 7nm is sufficiently energy efficient. It's not like there are 100W CPUs in cellphones, yet there were ~54nm ones. All those manufacturing processes are made to perform in certain way under certain scenario. Take your 7nm cellphone CPU, downclock it, and you have something eating 0.01W. (Actually problematic part of power consumption of cellphone is wifi and cell-network. At least in stand-by mode.)

How did you arrive at that number?

yasamoka:

How did you arrive at that number?

By daydreaming and trying to make himself look smart.

yasamoka:

How did you arrive at that number?

Check your old inefficient phone. Battery Wh capacity, number of hours of not being used before it is drained. Divide Battery's Wh by hours of life to get average power consumption. That's going to be somewhere between 0.25~0.40W. Kill Wifi, you are somewhere around 0.12~0.18W. Kill location, sync allowance which wakes up CPU, BT, And you are at 0.07~0.11W. Enable only emergency and you are already at 0.06~0.09W. (Call and data are still on.) Battery in my SGS6e is like 11Wh, and in retard/emergency mode, phone lasts 140 hours. 11Wh / 140h = 0.079W. That's when CPU is limited to 60% (1.26GHz) of maximum clock and all non essentials are disabled. (Call and data are still on.) All components still passively drain energy. This is on 1st 14nm FinFET chip. Thing which overheats and throttles in 5 minutes of "full" load (2.1GHz). 10nm chips are not forced to run 4x fast and 4x slower, more power efficient cores. Snapdragon 835 has 8x 2.45GHz. And at same time they claim 25% lower power consumption. This chip is followed by Snapdragon 845 which has 8x 2.8GHz. Tests already shown that Those 10nm Snapdragons do not throttle. See, how much they increased clock while improving power efficiency. Put phone with such chip into limited emergency mode and and it is likely around 0.05~0.07W. Cut undesirable electronics and chip itself may idle at 0.02W already. Hard to say what part of that power consumption is just passive discharge of battery due to VRM resistance. 7nm can get there, especially if particular line was designed for maximum power efficiency at cost of lowering maximum clock. (But I did read phones will go over 3GHz.) With cellphones, we will really need some new technologically advanced antenna. Something like tiny fractal ball which will detect direction of cell tower in real time. And use less energy to transmit only in that direction.

Take your 7nm cellphone CPU, downclock it, and you have something eating 0.01W. (Actually problematic part of power consumption of cellphone is wifi and cell-network. At least in stand-by mode.)

Understood!!!!!

-Tj-:

Intel is a fool on its own, if they had problems doesn't mean others will too.

My comment had nothing to do with Intels issues moving to 10nm. When pushing 5nm unless they are going to keep the gate pitch and interconnect pitches pretty large wich likely wouldn't end up making a much more dense chip over 7nm then they will hit some real physical limits. I'm not saying they won't overcome them I just would be very surprised if they nail there dates.

-Tj-:

Intel is a fool on its own, if they had problems doesn't mean others will too.

No I can't imagine a company who has the technology and resources to do this sort of thing could possibly have the slightest clue what they're doing Also, 5nm will be many years away...not just round the corner

5nm is already showing significant problems: https://www.eetimes.com/document.asp?doc_id=1333008

Suddenly my 14nm++ cpu feels so outdated and last gen with all these single digit nm talks.. :P

Koniakki:

Suddenly my 14nm++ cpu feels so outdated and last gen with all these single digit nm talks.. 😛

And Intels next Gen process node will still be double digit nm 😉

Matt26LFC:

And Intels next Gen process node will still be double digit nm 😉

You are aware that Intels 10nm is comparable in size to TSMC 7nm, right? 🙂 The numbers are mostly marketing and everyone else opted to use smaller numbers to have a perceived advantage these days. Its the same thing as Intel just using 14nm+++ while other companies used 12nm to describe their "enhanced 14nm" process.

nevcairiel:

You are aware that Intels 10nm is comparable in size to TSMC 7nm, right? 🙂 The numbers are mostly marketing and everyone else opted to use smaller numbers to have a perceived advantage these days. Its the same thing as Intel just using 14nm+++ while other companies used 12nm to describe their "enhanced 14nm" process.

12nm LP is different than 14nm LPP that it is based on,( not hugely different, but more different than a typical node enhancement) There exists 14nm+ LPP aswell that raven ridge uses(an enhanced 14nm LPP that first gen ryzen uses), which is more inline with intel's small node improvements. ( intel 's 14nm +, 14nm ++ ect) https://www.tomshardware.com/reviews/amd-ryzen-5-2400g-zen-vega-cpu-gpu,5467-3.html Also the big difference between intel's 10 nm and tsmc's 7nm, is that tsmc can actually ship something half decent already:D tsmc is kicking a$$ and takin names right now