Full-scale mass production of the 3 nm node is not projected to begin until the second half of 2023, and TSMC will also begin their N3E later in 2023. ... TSMC has secured seven major customers for its upcoming 3 nm node
next gen will be on new node already, this is imo why we're seeing 450w cards,nvidia and amd have moved to one arch per node strategy,maxing out the silicon capabilities with every new series.
Usually the newest node is only for Apple. Maybe yields for the N3 node are so good, and cost of development so high, they need to spread it around as fast as possible.
Apple buys a big chunk of the production, but never all of it. There's always some companies that squeeze in. AMD is being the smartass, building the important stuff on the smallest node and I/O on an old node.
I'm not sure this is news as it is rather obvious. Its the same group that uses TSMC every major node change.
yes and no they need to spread it around as fast as possible (including political reasons), but the yield is fantastic. TSMC has increased yield so fast that if we used a time based graph from 2012 it would look like Mt. Everest...but so does the cost of R&D - where they've been just as aggressive. but just wait, ASML is running as fast as they can and TSMC is guaranteed to lose Intel's business after N3 as Intel comes online at four different fabs. so securing a wide launch for a long-term (lol) node is critical. like the previous poster said "if you build it they will come". outfitting the Intel fabs takes longer than building them as they've placed the largest order in ASML's history and have only taken (partial) delivery @ two fabs - which is taking the time used to be a "uni" for process and euv lithography. AMD is playing this well. having long ago gone to a disaggregated (chiplet) design they know how to milk the process as well as Nvidia and Intel have/had milked the monolithic so they aren't as needy for a knowledge base (poor little rich Nvidia is the one who has to create or license scalable disaggregated tech ready for production).
Seems like Intel's marketing dept is perpetually confused. Doing that would make some sense in that the average person doesn't care "how the sausage is made", but that message is useless to that group as a whole anyway. Anyone that actually cares about one fab or the next, is in enough in touch to know what Intel is doing. Hypocritical marketing is nothing new to them though. They've come at AMD for doing things they themselves have done, or plan to do. Glued together Pentium D, and Tiles/Chiplets. I do think they have another angle using TSMC, in hoping it's capacity AMD can't use. AMD is usually next to scoop up capacity after Apple is done on a node, and I recall TSMC being unwilling to give Intel as much as they originally wanted.
If i recall TSMC needs at least 18 months notice of what can be secured for production by a given customer and Apple is always the first to secure their share.
It was only rough seas until switching to EUV. Smooth sailing now that the technology has been perfected. It's pretty insane they were doing 7-10nm with light that was 193 nm !!!, via all kind of expensive and difficult tricks (which resulted in a lot of failures and low yields, clock reducing defects, etc.) EUV uses 13.5 nm light, which makes direct patterning a lot simpler and the result a lot sharper ... it's just that those EUV machines are incredibly complicated and expensive and it will take a long time to pay off. They are also slower due to the method of creating that EUV light. We'll probably see rapid shrinkages in lithography all the way into hitting the atomic limit (which is close, but not there yet). Most likely EUV will be the last system ever used for this way of making nanoelectronics. --- What comes after is anyone's guess...
