Question ( Arachnotronic ) TSMC 5nm in 2020, from Apple A14.

[ksec]

This post isn't about Apple's A14 but rather TSMC finally shipped 5nm in 2020. Most would have taken it for granted. But pointing out TSMC is going to be leading edge, and shipping 5nm in 2020 was not a popular opinion 4-5 years ago on Anandtech. ( And in fact any where on the internet ) As far as I am aware some of those discussion even predate Dr Ian Cutress joining Anandtech. And if you search the TSMC Tag on Anandtech, the oldest one only goes as far as March 2016. What that means is that TSMC in the news is actually a fairly recent thing.

So I used to get a lot of sticks, insults and attacks for stating what is now considered or known as facts. But it has finally happened. TSMC shipped 5nm in 2020 with the Apple A14.

I dont know if Arachnotronic still goes on Anandtech. I hope he see this. Waited 4 years to prove a point. And we can finally settle this for good.

And Thank you Arachnotronic for backing me then.

Is TSMC a credible company? TSMC roadmap as of July 2016 (see here).
-10nm products in Q1'17 (0.52x area scaling, x'tor arch ~~ to Intel 14nm)
-7nm tape-out in H1'17, prod. in Q1'18 (1.63x area, =area of Intel 10nm, x'tor arch <= Intel 10nm, no III-V/Ge)
-5nm in 2020 [H1'19 risk production] (area < Intel 7nm, Intel 7nm in '20 according to PAO, x'tor arch <= Intel 10nm I would bet, maybe <=7nm; III-V? GAA?)
If following turns out correct, credits to TSMC, ksec & Arachnotronic. (I leave this forecast unchanged (-: to compare to what actually happened.)

 

[Hitman928]

This post isn't about Apple's A14 but rather TSMC finally shipped 5nm in 2020. Most would have taken it for granted. But pointing out TSMC is going to be leading edge, and shipping 5nm in 2020 was not a popular opinion 4-5 years ago on Anandtech. ( And in fact any where on the internet ) As far as I am aware some of those discussion even predate Dr Ian Cutress joining Anandtech. And if you search the TSMC Tag on Anandtech, the oldest one only goes as far as March 2016. What that means is that TSMC in the news is actually a fairly recent thing.

So I used to get a lot of sticks, insults and attacks for stating what is now considered or known as facts. But it has finally happened. TSMC shipped 5nm in 2020 with the Apple A14.

I dont know if Arachnotronic still goes on Anandtech. I hope he see this. Waited 4 years to prove a point. And we can finally settle this for good.

And Thank you Arachnotronic for backing me then.

You can @ people so they are notified.

@ksec

 

[moinmoin]

Times sure have changed. Before TSMC was easy to ignore since Intel monopolized all the attention for foundry advancement and TSMC wasn't directly rivaling Intel in any consumer market anyway. Now Intel has fallen behind, and how far they have fallen behind can readily be seen with directly competing x64 CPUs by AMD.

 

[SarahKerrigan]

Some of the posts on that thread have aged pretty hilariously.

 

[Dave2150]

Some of the posts on that thread have aged pretty hilariously.

Yeap, just look at this gem!

And Fudzilla today also has an article.

http://www.fudzilla.com/news/41569-tsmc-to-begin-7nm-volume-production-in-q1-2018

They probably do this for iPhone 8, they know that their Intel 14nm copycat node (called 10nm) won't be able to compete with Intel's 10nm node, so it seems they are hurrying up like crazy to try to get a share of Apple's foundry tasks away from Intel.

But remember it's still TSMC we're talking about.

Intel's 10nm didn't work out quite so well

 

[blckgrffn]

Some of the posts on that thread have aged pretty hilariously.

Thanks for sending me down that rabbit hole

I've already had one good chuckle IRL, so that's going to be a gold mine.

Keeping that tab open for later...

 

[dmens]

so it seems they are hurrying up like crazy to try to get a share of Apple's foundry tasks away from Intel.

What the heck? Even in 2016 that is a ludicrous statement. Intel custom foundry was already in its death throes by then, Apple (and every other customer) had zero interest in Intel foundry after one look at the 10nm design rules.

 

[NTMBK]

It's easy to mock @witeken now, but predicting the future is hard. Intel have ruled the roost for a long time, I don't think anybody expected them to screw up 10nm quite this hard.

 

[SarahKerrigan]

It's easy to mock @witeken now, but predicting the future is hard. Intel have ruled the roost for a long time, I don't think anybody expected them to screw up 10nm quite this hard.

Which is a good reason predictions should be made cautiously, and ideally with a degree of humility with them. When you're a jerk about it, you can expect to be called out.

 

[eek2121]

It's easy to mock @witeken now, but predicting the future is hard. Intel have ruled the roost for a long time, I don't think anybody expected them to screw up 10nm quite this hard.

I saw it coming well in advance. I knew they couldn’t keep up with the aggressive shrinks. I figured they would get stuck on 14nm, but I NEVER expected to be stuck this long. Holy crap.

Intel now has to get 7nm out to catch up, but TSMC will be ramping up 3nm by they time they do. At this point it will take a miracle for Intel to ever pull out ahead again.

Oh how the mighty have fallen.

 

[Dave2150]

I saw it coming well in advance. I knew they couldn’t keep up with the aggressive shrinks. I figured they would get stuck on 14nm, but I NEVER expected to be stuck this long. Holy crap.

Intel now has to get 7nm out to catch up, but TSMC will be ramping up 3nm by they time they do. At this point it will take a miracle for Intel to ever pull out ahead again.

Oh how the mighty have fallen.

Probably not long until "intel inside" becomes resigned to the history books, if they can't fix/rollout 10nm/7nm.

 

[Saylick]

I thought this was kinda cool to see.

TSMC Wafer Prices (Not Official, but Rigorous Estimate):

Source: @RetiredEngineer

 

[LightningZ71]

As long as TSMC continues to approach the progression of nodes like they have been in the recent past, and don't have any major natural disasters or political ones, they should at lest keep up their current cadence of new nodes. While that seems an obvious statement, I mean that they seem willing to do what we call half nodes to spread the risk out for their reearch efforts, always stepping forward every year. This compares to what Intel used to do with big node jumps that would just have rules relaxations for specific needs. If they blew their development, like they did with 10nm, it's a major problem. It seems that they have changed their pace with 10sf, so, that should help them.

 

[Doug S]

As long as TSMC continues to approach the progression of nodes like they have been in the recent past, and don't have any major natural disasters or political ones, they should at lest keep up their current cadence of new nodes. While that seems an obvious statement, I mean that they seem willing to do what we call half nodes to spread the risk out for their reearch efforts, always stepping forward every year. This compares to what Intel used to do with big node jumps that would just have rules relaxations for specific needs. If they blew their development, like they did with 10nm, it's a major problem. It seems that they have changed their pace with 10sf, so, that should help them.

The 10nm "super fin" is just the original 10nm that they tried and failed to launch. The initial 10nm they released was a deliberate step back when they were unable to make it work.

That's why the first 10nm CPUs showed no benefit at all over 14nm, now that they've finally got the real 10nm out five years late they are finally able to get the node jump's worth of performance.

 

[Gideon]

It's easy to mock @witeken now, but predicting the future is hard. Intel have ruled the roost for a long time, I don't think anybody expected them to screw up 10nm quite this hard.

I would probably agree with you if it were anybody but witeken but that dude is not sane. Some example tweets of his:

March 2020 (TSMC 7nm being worse than Intel 14+, let alone TGL which according to him is as big a jump from intel 14nm as 14nm is from 32nm)

https://twitter.com/x/status/1244904370413744129

Rocket lake predictions:

https://twitter.com/x/status/1241690622832119808

"Peak Ryzen" marketshare reached in 2018:
https://twitter.com/witeken/status/1022223615029379073

"Ice Lake crushing mobile Ryzen":
https://twitter.com/i/web/status/1141321205682757632

Claiming "Tick-Tock had hardly any mishaps vs TSMC screwing everything up" In September 2016 (when it wasn't clear how bad it was for Intel, but it was pretty clear it was bad)

And he had countless others where he belittles all the competitors as much as possible, while always choosing to believe the most outlandish claims for Intel's future products. In a tweet I can't find right now he even "predicted" Intels financials for the next 10-15 years, every category just going up by fixed percentage per year, including IoT stuff, etc ...

He gets no break from me, he's as bad as @ShintaiDK,

 

[Eug]

Well, I will admit that my original plan (back from around 2014 IIRC) was to buy a new Intel-based desktop and laptop when they went 10 nm in 2016. We all know how that turned out. I ended up buying 14 nm in 2017.

 

[TheELF]

I thought this was kinda cool to see.

TSMC Wafer Prices (Not Official, but Rigorous Estimate):

Source: @RetiredEngineer

Does this say that they spend 16,746 bucks on making them and are selling them for 16,988 or am I missing something?

 

[DrMrLordX]

The 10nm "super fin" is just the original 10nm that they tried and failed to launch.

Are you sure about that? 10nm was originally supposed to be more dense than what Intel rolled out in 2019 with IceLake, and I don't think 10SF changed that situation. Did it?

 

[name99]

It's easy to mock @witeken now, but predicting the future is hard. Intel have ruled the roost for a long time, I don't think anybody expected them to screw up 10nm quite this hard.

It depends on the mental model you are using for your predictions.
Is your model drawing lines on a graph then extrapolating them?
Is your model looking at the people involved, the management structure, the pattern of incentives, how they have dealt with previous failure?

The worse mismatch between prediction and reality tends to come from those who
(a) live in a fan bubble -- they only talk to people who already agree with them, and believe uncritically everything their fan company says AND
(b) assume that everyone else is operating on the exact same model, so they're incapable of even comprehending that someone disagreeing with them might be doing so on rational (rather than "support my team") grounds.

 

[Martimus]

This post isn't about Apple's A14 but rather TSMC finally shipped 5nm in 2020. Most would have taken it for granted. But pointing out TSMC is going to be leading edge, and shipping 5nm in 2020 was not a popular opinion 4-5 years ago on Anandtech. ( And in fact any where on the internet ) As far as I am aware some of those discussion even predate Dr Ian Cutress joining Anandtech. And if you search the TSMC Tag on Anandtech, the oldest one only goes as far as March 2016. What that means is that TSMC in the news is actually a fairly recent thing.

So I used to get a lot of sticks, insults and attacks for stating what is now considered or known as facts. But it has finally happened. TSMC shipped 5nm in 2020 with the Apple A14.

I dont know if Arachnotronic still goes on Anandtech. I hope he see this. Waited 4 years to prove a point. And we can finally settle this for good.

And Thank you Arachnotronic for backing me then.

We were talking about TSMC all the time since I joined, but in the video card forum. Before AMD went fabless, most of the talk was about how Intel had a huge process advantage over AMD and it took huge design wins in order for AMD to stand a chance. Luckily they had those in spades in the late 90s/early 2000s with the K7, integrated memory controller, x64 extensions, and high IPC and low power designs. Unfortunately they never gained a significant foothold, and Intel was able to use their process advantage to stay competitive until they developed Core-2.

Intel had the best process from the mid 90s through just a few years ago, so TSMC catching up and exceeding them is really impressive. Especially when you count how expensive it is to advance at all. That expense has seen everyone but Intel, Samsung, and TSMC give up on chasing the most advanced process nodes. I am amazed that Intel let a competitor surpass them, but I shouldn't really be surprised. It just never seemed possible 10 years ago, but 10 years is a long time and mistakes can creep up on you taking away any advantage you might have had.