7nm EUV in 2019

[Ajay]

Can't be predicted because it comes down to what the materials guys can come up with.

Don't see how we are going get down past 1/2 nm - the materials guys will be working with xtor features only 1-2 atoms thick

 

[raghu78]

TSMC confirms 7+nm EUV process to enter risk production in June 2018. Volume production should pretty much start by mid 2019. I am pretty sure Apple Ax chips for 2019 iphones will use this process.

http://www.eetimes.com/document.asp?doc_id=1331489

TSMC / Apple juggernaut rolls on. 4+ Ghz on Cortex A72 using 7nm HPC. Intel's process lead is over. RIP Intel.

 

[Arachnotronic]

RIP Intel.

You've been saying this for years.

 

[Ajay]

TSMC confirms 7+nm EUV process to enter risk production in June 2018. Volume production should pretty much start by mid 2019. I am pretty sure Apple Ax chips for 2019 iphones will use this process.

http://www.eetimes.com/document.asp?doc_id=1331489

TSMC / Apple juggernaut rolls on. 4+ Ghz on Cortex A72 using 7nm HPC. Intel's process lead is over. RIP Intel.

Well, it will be Intel's "10nm" vs TSMC's "7nm". TSMC will probably have a density advantage, but on electrical, I think Intel will still have the lead.

 

[raghu78]

No. TSMC 7nm HPC is pretty much going to be on par with Intel 10nm even for electrical characteristics. The 7nm HPC process is optimized for 4+ Ghz CPUs. 7+nm with EUV will be better than Intel 10nm.

 

[raghu78]

Arachnotronic i am very sure that Intel's process leadership is over. Intel used its manufacturing process lead and economics of scale to beat UNIX/RISC vendors. ARM and the foundries are going to do the same thing to Intel. Intel came from the low end PC and started encroaching higher and higher into servers. Its the same with ARM. They started from mobile but slowly and surely they are going after the higher margin markets like servers/HPC. With Microsoft throwing their weight behind ARM its only a matter of time before ARM replaces x86 as the dominant ISA across the entire computing spectrum. It will not happen overnight but will happen within a decade. I think the TSMC of today is the Intel of 1990s. They have used the mobile revolution to fuel their growth for the past 10 years. For the next 10 years its going to be high performance computing. And guess who stands to lose the most - Intel.

 

[Ajay]

No. TSMC 7nm HPC is pretty much going to be on par with Intel 10nm even for electrical characteristics. The 7nm HPC process is optimized for 4+ Ghz CPUs. 7+nm with EUV will be better than Intel 10nm.

So, you have all the electricals for both processes? How about parametric yields?

 

[Arachnotronic]

No. TSMC 7nm HPC is pretty much going to be on par with Intel 10nm even for electrical characteristics. The 7nm HPC process is optimized for 4+ Ghz CPUs. 7+nm with EUV will be better than Intel 10nm.

LOL, you don't have any data on Intel 10nm so how can you say this so confidently?

p.s. Intel 14nm+ is shipping silicon -- CPUs much more complex than A72, in fact -- running at 4.5GHz stock and can OC to 5GHz today.

 

[CHADBOGA]

You've been saying this for years.

That it hasn't happened yet, is just a sign of what a far sighted visionary he is.

 

[french toast]

Considering intel had at one point a 3-4 year process leadership its fair to say by 2018 that will be far from the case, by 2019 amd will be using a more advanced process than intel for the first time in like, ever.
Icelake might be a beast that negates that, but i wouldnt be confident if i was an intel investor.

 

[Aristotelian]

I have a bit of an ignorant question but I hope you'll be a bit patient. I've been coming to Anandtech for many years (this is a new account - I lost my old one and the password ages ago) and I remember these hardware/rumour/expectation threads regarding CPUs as being something like 'mythical'. What I mean there is that, while it is true there is a massive amount of engineering expertise, research and development, foundry issues, fab competition, process competition, the question I have is:

Where does the average joe see this translate into usable products? Even with the Samsung S7 Edge (the cpu etc. inside it), I think in the Anandtech review it was stated that the increase in battery life largely resulted from an increase in battery size, not from an increase in efficiency.

And here - it seems like there's a large tension between producing 'faster' chips and producing 'more efficient' chips and it seems (from my ignorant perspective) that 'efficiency' is winning battles.

Am I way off base here? I'd love to be excited about new processor tech but I want to know what it means in terms of hardware usage too, know what I mean?

 

[sm625]

So we will see the first 7nm phone SoC in 2020?

 

[Arachnotronic]

So we will see the first 7nm phone SoC in 2020?

2018.

 

[sm625]

2018.

No way. I'll take that bet.

 

[Arachnotronic]

No way. I'll take that bet.

Apple A12 should be a TSMC 7nm SoC, and iPhones come every fall

 

[sm625]

Apple A12 should be a TSMC 7nm SoC, and iPhones come every fall

I took the history of all apple SoCs and made a log_base2 graph (shown in red). I then added a simple trendline (in black) and extended that trendline in green and added blue dots around typical iPhone release dates. I expect the 2017 and 2018 iPhones to be on 10nm. The 2019 iPhone is an odds-even tossup between 10nm and 7nm.

 

[french toast]

I took the history of all apple SoCs and made a log_base2 graph (shown in red). I then added a simple trendline (in black) and extended that trendline in green and added blue dots around typical iPhone release dates. I expect the 2017 and 2018 iPhones to be on 10nm. The 2019 iPhone is an odds-even tossup between 10nm and 7nm.

Nope, 2018 likely to be 7nm, 2019 definitely 7nm.

 

[sm625]

Nope, 2018 likely to be 7nm, 2019 definitely 7nm.

What you're saying is that the next two years are going to break a 10 year trend in process tech, for the better. In essence, outstripping Moore's Law. How likely is that, really? Of course we would need to overlay feature sizes and transistor counts onto that same graph to get a true picture, but it just seems rather unlikely, especially given how similar promises were made in the past.

 

[Lodix]

What you're saying is that the next two years are going to break a 10 year trend in process tech, for the better. In essence, outstripping Moore's Law. How likely is that, really? Of course we would need to overlay feature sizes and transistor counts onto that same graph to get a true picture, but it just seems rather unlikely, especially given how similar promises were made in the past.

You can do the graphs you want, TSMC is starting risk production next month of their 7nm and will be in mass production next year. "More's Law" is not a law and proces nodes have different marketing names.

 

[french toast]

What you're saying is that the next two years are going to break a 10 year trend in process tech, for the better. In essence, outstripping Moore's Law. How likely is that, really? Of course we would need to overlay feature sizes and transistor counts onto that same graph to get a true picture, but it just seems rather unlikely, especially given how similar promises were made in the past.

You should read the recent info on fab roadmaps, 7nm tsmc will be in full production in volume H2 2018, providing no crazy yield issues or hold ups.
Tsmc the moves to 20% denser and 10% better electrical characteristics 7nm EUV in 2019, likely in time for iphone 10.

 

[Ajay]

You should read the recent info on fab roadmaps, 7nm tsmc will be in full production in volume H2 2018, providing no crazy yield issues or hold ups.
Tsmc the moves to 20% denser and 10% better electrical characteristics 7nm EUV in 2019, likely in time for iphone 10.

Yields won't likely be good by 2H18, but it doesn't matter - Apple will pay more for it's custom ARM SOCs than anyone else will. That it what is driving TSMCs 7nm development (and what probably drove their 16FF && 12FF nodes). Without Apple, TSMC wouldn't have a customer that would buy large volumes with lower than standard yields.

 

[dark zero]

dude you are selectively quoting what you want to drive your narrative. TSMC confirmed the 7 EUV in second year of 7nm volume production. Since the 7nm HVM ramp is in H1 2018 that means TSMC 7 EUV will ramp in 2019. Its obvious since 5nm EUV is scheduled for ramp in 2020 and TSMC would want to introduce EUV on a mature 7nm node and use the experience for a smooth 5nm EUV ramp a year later.

And seems that True 7nm is the EUV one, the one which AMD will go.. so expect more refresh at 14 nm with AMD.

 

[Lodix]

Yields won't likely be good by 2H18, but it doesn't matter - Apple will pay more for it's custom ARM SOCs than anyone else will. That it what is driving TSMCs 7nm development (and what probably drove their 16FF && 12FF nodes). Without Apple, TSMC wouldn't have a customer that would buy large volumes with lower than standard yields.

I don't know why but I feel some hate/bias/personal preferences on your words

I don't see anything wrong to bring some sane competition in this market even if you are a fanatic of some monopolistic brand. TSMC said they have alredy 76% yields rates and they usually have good ramps up of their new processes so I don't see anything to proclaim what you said.

 

[Ajay]

I don't know why but I feel some hate/bias/personal preferences on your words

I don't see anything wrong to bring some sane competition in this market even if you are a fanatic of some monopolistic brand. TSMC said they have alredy 76% yields rates and they usually have good ramps up of their new processes so I don't see anything to proclaim what you said.

I have nothing against competition. I have nothing against TSMC. I have issues with allot of pure fabs, they play more games with stats, roadmaps and yields than Intel. Intel just shuts up for a while when things are going wrong.

Anywho, and more to the point, TSMC already has 76% yield rate on what product, exactly?

 

[Lodix]

I have nothing against competition. I have nothing against TSMC. I have issues with allot of pure fabs, they play more games with stats, roadmaps and yields than Intel. Intel just shuts up for a while when things are going wrong.

Anywho, and more to the point, TSMC already has 76% yield rate on what product, exactly?

Intel plays with smoke and mirrors much more since they sell finished products, TSMC offers a service and have to be more transparent to customers.

The 76% yield rate is on 256Mbit SRAM modules.