Samsung Foundry to mass produce 1.4㎚ in 2029… Focus on ‘recovery of operating rate’
Posted by SlamedCards@reddit | hardware | View on Reddit | 26 comments
KR4T0S@reddit
Its difficult to see Samsung mounting a challenge without significant demand, their foundry business must be really struggling to find the revenue to build up or out so their node supremacy dream is just a dream IMO. But if they can keep pushing forward it might not matter over the long term, eventually node shrinks wont be a thing so its a race with a finishing line. Samsung getting there several years after isnt the same as being perpetually behind a moving target.
WJMazepas@reddit
Lesser known or not as flash components were using Samsung in the recent years.
Qualcomm keeps using for a lot of their mid to low end SoCs. IBM of all places uses Samsung 7nm for their datacenters CPUs
There were Bitcoin ASICs that also used Samsung.
But yeah, all of them don't have the same demand as Apple/Nvidia/AMD and won't get all the capacity of their fabs
PastaPandaSimon@reddit
Um, the Switch 2 is using Samsung. Even though the node is older and margins are low, the volume is enormous.
Hailgod@reddit
biggest customer should be themselves with the midrange exynos chips.
Strazdas1@reddit
yeah but thats a really old node at this point. its not really benefiting new nodes that much.
PastaPandaSimon@reddit
It's indeed really old (10nm-family). But it was a highly profitable node, having been used to fab many mobile chips (including Qualcomm's), Nvidia Ampere GPUs, Tesla car chipsets, Switch 2 chips, among others. It's been producing high volumes of chips for 9 years now.
That likely gives them money and motivation to keep developing new nodes, hoping they hit a gold mine like it again.
Though indeed, it was seen as a "value" node for at least 7 out of those 9 years.
WJMazepas@reddit
Oh yeah, I totally forgot about the Switch 2
Geddagod@reddit
IBM Telum 2 is on Samsung 5nm too.
xternocleidomastoide@reddit
Samsung's research nodes percolate to their DRAM/NAND products, which is where most of their value/revenue comes from.
They do have some wins w QCOM, IBM, and NVDA.
But they are going to be behind TSCM in dynamic logic die volume for the foreseeable future. Nobody can out execute TSMC at this point
Jeep-Eep@reddit
Anyone not using government money to push their leading edge to functionality is not running their economy well tbh.
Letting GloFo fall out of leading edge was criminal incompetence for one thing...
KR4T0S@reddit
Tbh I dont think much of the computer industry would have gotten to where it is without government subsidies. The amount of money required to build out even 10% of capacity taking into consideration not only the infrastructure but also the fact that you need to educate the people working in these facilities? Government are an indispensable part of much of industry. Funny thing is recently it seems like many politicians realise this about manufacturing cars but not computers.
Jeep-Eep@reddit
Yup, and I suspect there will be some slowdowns in cooling tech after the DOGE rampage as well...
Old_Wallaby_7461@reddit
The electronics industry as a whole, not just the computer industry.
Modern solid-state components were largely a US DoD thing, starting from the Minuteman ICBM program in the late 1950s. We've forgotten that and we will pay for it.
SlamedCards@reddit (OP)
There was an IMEC roadmap out to like 2035 or 2040
So we still got lot of time to worry about shrinks stopping
Alive_Worth_2032@reddit
The true wall will be economics rather than physics.
More and faster transistors only means something as long as it can be justified by the cost.
SlamedCards@reddit (OP)
I think we are fortunate that AI has emerged as the new driver of the leading edge. Mobile might get tapped out after 14A or 10A family of nodes
KR4T0S@reddit
I believe the actual picture is more murky because there's no agreement on what say 2nm is or what it measures. Samsung for example is working on 1.4nm but that isnt 1.4nm at TSMC or Intel.
IMECs roadmap seems to prioritise packaging with things like GAAFET allowing a higher transistor density rather than shrinking transistors. The metal pitch for A7, A5 and A3 is all 16nm for example.
Dangerman1337@reddit
CFER with A7 and then A5 and after that looks great, and A10 with Forksheets looks promising as well. 2nm and 1.4nm gens especially by TSMC look very disappointing with sib psr logic density increases.
Though how it will all cost will be a bug one...
SlamedCards@reddit (OP)
Samsung delaying 1.4nm to 2029
Samsung originally targeted 2027
TSMC targets HVM of A14 in 2028
Intel targets risk production of 14A in 2027 (means HVM is likely 2028 or very late 2027)
Though Samsung 1.4nm might have been a dubious name anyway considering there 3nm and 4nm nodes have lot less PPA than TSMC or Intel
Geddagod@reddit
Idk why we sliding Intel in there, their Intel 3 node is not close to TSMC N3 either...
Strazdas1@reddit
if we consider samsung node a fair comparison then so is Intels.
Geddagod@reddit
We aren't considering samsung's node a fair comparison though, which is why I said Intel either.
But tbh, I think Samsung 3GAP is much closer to TSMC N3 than Intel 3 is.
Strazdas1@reddit
The person you responded to is comparing samsungs node to TSMC on equal terms - 1.4nm samsung vs A14 TSMC.
Geddagod@reddit
The person I was responding to was also just comparing the node namings in general, hence why he also brought up 3 and 4nm
SlamedCards@reddit (OP)
There's Intel being little off and then there's Samsung
4nm is really last node we can compare since 3nm is just a watch. And it performed worse than TSMC 4nm and had alot higher power in some galaxy phone with same soc
Inferior density to Intel 3 HD. And Intel 4 HP is only slightly worse than their UHD. Despite Intel 3 being HPC optimized and Samsung 4nm being lower power node
6950@reddit
Intel's 14A may be Q4 27/H1 28 HVM against H2 28 HVM for A14