You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
The plan for tsmc N2 HVM is 4 phases in Hsinchu and few phases in Taichung. According to the latest rumors from supplier chains, tsmc have postponed Taichung site and 3 phases in Hsinchu for N2 HVM to finished in 2026 or later and will only have one phase in Hsinchu ready in 2025.
If these are true which mean N2 will only significantly contribute to revenue till 2026. tsmc's earning call next week should address these rumors.
Per Anandtech’s coverage of TSMC’s roadmap update from last April:
"Our progress so far today for the N2 is on track," said Mr. Wei. "All I want to say is, yes, at the end of 2024, [N2] will enter the risk production. 2025, it will be in production, probably close to the second half or the – or the end of 2025. That is our schedule."
Given this statement, unless TSMC is pulling in N2, not finishing the bulk of their capacity buildout until 2026 is internally consistent with prior public statements. Doubly so if N2 is to be a 20nm situation where most customers just wait for the full N3E successor in the form of N2BSP. Either way I would still take these types of rumors with a hearty grain of salt.
Per Anandtech's coverage of TSMC's roadmap update from last April:
Given this statement, unless TSMC is pulling in N2, not finishing the bulk of their capacity buildout until 2026 is internally consistent with prior public statements. Doubly so if N2 is to be a 20nm situation where most customers just wait for the full N3E successor in the form of N2BSP. Either way I would still take these types of rumors with a hearty grain of salt.
Generally Apple is the risk production at the end of the year for mass production approximately one year later. This means the 2025 iPhones should be N2 but associated N2 revenue won't start until 2026. There is a difference between production and mass (Apple) production. Samsung can say GAA is in production but it is not Apple class of production.
TSMC 20nm was unique since double patterning was added first then FinFETs to make 16nm and yes most everybody skipped 20nm.
In my opinion most companies will stick with N3 as long as possible to see how N2 ramps up. There needs to be a good ROI for GAA because it is going to cost more.
That’s why I think there is a possibility 2026 i products might make sense. For N7 and N5 risk starts were during the first half of the year rather than the end of the year. The statement that “in production” near the end of 2025 kind of reminds me of N3 where it reached Apple ready a few months to late for an iPhone launch. However I would love to be wrong here. It would also be a huge win for TSMC. If N2 risk starts to launching products ends up being less than a year that is their fastest ramp in years on a brand new architecture. It would be a big/welcome step up from the slower ramp TSMC had on 16FF.
This means the 2025 iPhones should be N2 but associated N2 revenue won't start until 2026. There is a difference between production and mass (Apple) production. Samsung can say GAA is in production but it is not Apple class of production.
I thought TSMC gets paid when the wafer leaves their door? If Wafers aren’t leaving the door until late 2025 how could an iPhone launch in September of that year? Am I misunderstanding your statement here Dan?
TSMC 20nm was unique since double patterning was added first then FinFETs to make 16nm and yes most everybody skipped 20nm.
In my opinion most companies will stick with N3 as long as possible to see how N2 ramps up. There needs to be a good ROI for GAA because it is going to cost more.
Agreed. Given the small chip area improvement there is a good chance cost per fet might not decrease. Combine this with the meh performance improvement, I think most would want to wait for BSPD for the extra density and better routability. I am less concerned on N2 having N3 ramp issues, but maybe I’m too optimistic.
Generally Apple is the risk production at the end of the year for mass production approximately one year later. This means the 2025 iPhones should be N2 but associated N2 revenue won't start until 2026. There is a difference between production and mass (Apple) production. Samsung can say GAA is in production but it is not Apple class of production.
TSMC 20nm was unique since double patterning was added first then FinFETs to make 16nm and yes most everybody skipped 20nm.
In my opinion most companies will stick with N3 as long as possible to see how N2 ramps up. There needs to be a good ROI for GAA because it is going to cost more.
If only one phase is ready for HVM by 2025, then N2 production capability is ~20k wafers/month. It seems not enough to support 2025 iPhone. In contrast, the initial production capability for N7 and N5 is 50-60k/month.
I thought TSMC gets paid when the wafer leaves their door? If Wafers aren’t leaving the door until late 2025 how could an iPhone launch in September of that year? Am I misunderstanding your statement here Dan?
Apple historically tapes-out in Q4 for production in 2H of the following year. Revenue is based on the terms of the wafer agreement. From what I remember in the past new node revenue was reported in Q1 versus Q4 when wafers actually shipped. Apple specific revenue is not reported. I may be wrong. Look at 16nm, 10nm, etc... Maybe Apple skips the first N2. I highly doubt it but it could happen.
Apple historically tapes-out in Q4 for production in 2H of the following year. Revenue is based on the terms of the wafer agreement. From what I remember in the past new node revenue was reported in Q1 versus Q4 when wafers actually shipped. Apple specific revenue is not reported. I may be wrong. Look at 16nm, 10nm, etc... Maybe Apple skips the first N2. I highly doubt it but it could happen.
I guess we might be having different definitions of things, so let me know if my understanding is what yours is…
Taking N5 as our example:
-Risk starts began in like March 2019.
-At the end of 2019 N5 left the risk starts phase and entered production/became HVM ready
-Sometime during Q1 full production apple wafers begin their months long journey of becoming A series SOCs (however it wouldn’t surprise me if Apple was already getting some late risk start wafers shipped to them as sudo HVM product).
-Q2 Wafers get packaged and Foxconn gets the dies.
-Q3 phones get shipped to the states and go on sale at eoq3/boq4.
I guess we might be having different definitions of things, so let me know if my understanding is what yours is…
Taking N5 as our example:
-Risk starts began in like March 2019.
-At the end of 2019 N5 left the risk starts phase and entered production/became HVM ready
-Sometime during Q1 full production apple wafers begin their months long journey of becoming A series SOCs (however it wouldn’t surprise me if Apple was already getting some late risk start wafers shipped to them as sudo HVM product).
-Q2 Wafers get packaged and Foxconn gets the dies.
-Q3 phones get shipped to the states and go on sale at eoq3/boq4.
I was speaking of Apple. Apple gets a custom version of the TSMC processes which is first to silicon. TSMC does not talk about Apple business publicly. The Apple design cadence is to freeze in Q4 and start production in Q2 the following year for the iPhone. Remember, Apple is doing iterative designs for iProducts and Macs so it is a different process than a greenfield design. After Apple is in production TSMC releases an Nx version of the process for the masses. There is usually an SoC (low power) version and an HPC version followed by automotive.
The iPhone is usually first into production on a new node but this time N3 was delayed so maybe an iPad or Mac will come first. I'm pretty confident though that the iPhone 15 will be N3 based.
I was speaking of Apple. Apple gets a custom version of the TSMC processes which is first to silicon. TSMC does not talk about Apple business publicly. The Apple design cadence is to freeze in Q4 and start production in Q2 the following year for the iPhone. Remember, Apple is doing iterative designs for iProducts and Macs so it is a different process than a greenfield design. After Apple is in production TSMC releases an Nx version of the process for the masses. There is usually an SoC (low power) version and an HPC version followed by automotive.
The iPhone is usually first into production on a new node but this time N3 was delayed so maybe an iPad or Mac will come first. I'm pretty confident though that the iPhone 15 will be N3 based.
I guess that is why I am unsure if we will see an N2 iPhone (or anything for that matter) in 2025 with the current schedule. If Apple needs the process frozen in Q4 and production by Q2, then it doesn’t sound like N2 will be ready in time. If risk starts begin in “late 24” in the best case scenario that is 4Q from BOQ3 to EOQ2. Meanwhile N5 which according to TSMC is their best ramping process ever entered risk starts near the EOQ1 which is about a 5Q gap. For this reason, to me it sounds like “in production” in late 2025 is when the process is finalized (likely missing the period necessary for hitting the September 2025 iPhone launch). I could be wrong (and it would certainly be cool/impressive if I was), but unless TSMC announces a pull in for N2 I think the current schedule will miss the latest possible intercept for apple by 1-2 quarters. Although based on your statement on when the process needs to be finalized it should sync up perfectly with the 2026 iPhone launch.
If you see any flaws with this thought process, feel free to point them out Dan. I’m open minded and if anyone knows the TSMC-Apple ramp history would be you.
I guess that is why I am unsure if we will see an N2 iPhone (or anything for that matter) in 2025 with the current schedule. If Apple needs the process frozen in Q4 and production by Q2, then it doesn’t sound like N2 will be ready in time. If risk starts begin in “late 24” in the best case scenario that is 4Q from BOQ3 to EOQ2. Meanwhile N5 which according to TSMC is their best ramping process ever entered risk starts near the EOQ1 which is about a 5Q gap. For this reason, to me it sounds like “in production” in late 2025 is when the process is finalized (likely missing the period necessary for hitting the September 2025 iPhone launch). I could be wrong (and it would certainly be cool/impressive if I was), but unless TSMC announces a pull in for N2 I think the current schedule will miss the latest possible intercept for apple by 1-2 quarters. Although based on your statement on when the process needs to be finalized it should sync up perfectly with the 2026 iPhone launch.
So you know for a fact that when TSMC speaks of N2 publicly it is the Apple version of N2?
The Apple version of Nx is a lot like Intel 4 versus Intel 3. Intel 4 is for Intel designs only, Intel 3 is for everyone else. It is much easier to ramp a process for Apple because it is only for Apple, focused PDK etc...
Apple has a bit of an ego so if N2 is possible for 2025 they will do it. Personally I don't think the PPA difference between N2 and N3x will be enough for regular folks to move over quickly. We shall see.
Admittedly that is an assumption I am making. I figured it was fair, since with prior nodes when TSMC has talked about things like N7/N5 the dates for the base nodes they always seemed to be talking about the Apple version, with the more robust versions coming out after the fact. If TSMC could do it that would be a wonderful accomplishment, those engineers/scientists at fab12 deserve a nice fat bonus.
In a couple of years we shall see if my assumption was a good one or a poor one.
Admittedly that is an assumption I am making. I figured it was fair, since with prior nodes when TSMC has talked about things like N7/N5 the dates for the base nodes they always seemed to be talking about the Apple version, with the more robust versions coming out after the fact. If TSMC could do it that would be a wonderful accomplishment, those engineers/scientists at fab12 deserve a nice fat bonus.
In a couple of years we shall see if my assumption was a good one or a poor one.
From what I have heard inside the ecosystem N2 is being pushed pretty hard so I would bet it would be early versus late if I only had those two choices. I will write a quick note after the TSMC Symposium then we will go into more detail after the Taiwan and China versions. We have people attending both.
When Wei said "it will be in production, probably close to the second half or the – or the end of 2025." - everybody pessimistically took that to mean the end of 2025. But later on in the year, in the Q4 earnings call, he said "our N2 technology development is on track, actually is better than what we thought". So, if it's better than expected, now it's more likely to be 2H 2025, rather than end of 2025.
ec.ltn.com.tw
