TSMC will reportedly hike the pricing of wafers processed utilizing its forefront N3 (3nm-class) course of know-how by 25% in comparison with N5 (5nm class) manufacturing node. This may instantly make advanced processors like GPUs and smartphone SoCs costlier, which can make gadgets like graphics playing cards and handsets costlier. In the meantime, prohibitively excessive prices will make multi-chiplet designs extra interesting.
One wafer processed on TSMC’s forefront N3 manufacturing know-how will value over $20,000 in keeping with DigiTimes (through @RetiredEngineer). Against this, an N5 wafer prices round $16,000, the report says.
There are a lot of the reason why making chips on N5 and N3 manufacturing nodes is dear. First up, each applied sciences use excessive ultraviolet (EUV) lithography fairly extensively for as much as 14 layers in N5 and much more with N3. Every EUV software prices $150 million, and a number of EUV scanners should be put in in a fab, which implies further prices for TSMC. Additionally, it takes a very long time to provide chips on N5 and N3, which once more means larger prices for TSMC.
TSMC’s Alleged Wafer Pricing
TSMC does not typically reveal costs for its wafers, besides to precise prospects. It is also necessary to notice that contract pricing — what bigger orders from corporations like Apple, AMD, Nvidia, and even rival Intel are possible to make use of — could also be decrease than the bottom costs. Nonetheless, here is what the report says about present costs.
| Worth per Wafer | $20,000 | $16,000 | $10,000 | $6,000 | $3,000 | $2,600 | $2,000 |
| Node | N3 | N5 | N7 | N10 | N28 | 40nm | 90nm |
| Yr | 2022 | 2020 | 2018 | 2016 | 2014 | 2008 | 2004 |
Chip builders who use TSMC’s companies are anticipated to go the prices of latest chips on to downstream prospects, which can make smartphones and graphics playing cards costlier. Even now, Apple’s iPhone 14 Professional begins at $999, whereas Nvidia’s flagship GeForce RTX 4090 is priced at $1,599. As soon as corporations like Apple and Nvidia undertake TSMC’s N3 node, we are able to count on their merchandise to get much more costly.
In fact the precise chip value can nonetheless be comparatively small in comparison with all the opposite elements that go into a contemporary smartphone or graphics card. Take Nvidia’s AD102, which measures 608mm^2. Dies per wafer calculators estimate Nvidia can get about 90 chips from an N5 wafer, or a base value of $178 per chip. Packaging, PCB prices, elements, cooling, and so on. all contribute in all probability at the very least twice that a lot, however the true value is within the R&D points of recent chip design.
| Header Cell – Column 0 | N3E vs N5 | N3 vs N5 |
|---|---|---|
| Velocity Enchancment @ Similar Energy | +18% | +10% ~ 15% |
| Energy Discount @ Similar Velocity | -34% | -25% ~ -30% |
| Logic Density | 1.7x | 1.6x |
| HVM Begin | Q2/Q3 2023 | H2 2022 |
Whereas there are rational the reason why TSMC’s costs are getting larger, it needs to be famous that that the corporate can get away with it because it at present does not have any rivals that may produce chips utilizing a forefront fabrication applied sciences with first rate yields and in excessive volumes. Whereas formally Samsung Foundry is forward of TSMC with its 3GAE course of know-how (3nm-class, gate throughout transistors), it’s believed that it’s used just for tiny cryptocurrency mining chips attributable to inadequate yields. In the meantime, Samsung Foundry’s 4nm-class course of know-how didn’t reside as much as expectations so far as efficiency is worried.
When (and if) Samsung and Intel Foundry Companies provide course of applied sciences that outperform these of TSMC, the world’s largest foundry should considerably restrict its costs, though we don’t count on chip costs to drop due to intensified competitors on the foundry market as fabs are getting costlier, chip growth prices are rising, and fabrication applied sciences are getting extra advanced.
Generally, the prices of constructing chips on forefront nodes started to rise quickly in mid-2010s when Intel, GlobalFoundries, Samsung, TSMC, and UMC adopted FinFET transistors. On the time prices rose for everybody regardless of intense competitors between contract semiconductor producers.
TSMC’s first shopper to make use of N3 is predicted to be Apple, which may afford growing an applicable SoC, produce it in excessive quantity, and nonetheless make cash on its {hardware}. Apple has not indicated what sort of processors the corporate plans to make on N3, however a follow-up to the present M2 and A16 Bionic appears logical. Different chip builders could maintain off utilizing TSMC’s N3 for now due to its prohibitively excessive prices and use chiplet-based designs as an alternative attributable to decrease growth prices, decrease dangers, and decrease manufacturing prices.
Once more, please word that TSMC doesn’t touch upon its quotes and has not commented on data that it’ll cost ~$20,000 per N3 wafer. These figures come from business insiders and will or could not mirror precise excessive quantity costs.