TSMC will reportedly raise the price of wafers processed using its cutting-edge N3 (3nm class) process technology by 25% compared to its N5 (5nm class) production node. This quickly makes complex processors like GPUs and smartphone SoCs expensive, and devices like graphics cards and handsets expensive. On the other hand, the prohibitive cost makes multi-chiplet designs more attractive.
A single wafer processed with TSMC’s state-of-the-art N3 manufacturing technology can cost over $20,000. Digi Times (via @RetiredEngineer). By contrast, N5 wafers cost around $16,000, the report said.
There are many reasons why it’s expensive to build chips on N5 and N3 production nodes. First, both technologies make fairly extensive use of extreme ultraviolet (EUV) lithography, with up to 14 layers for N5 and even more for N3. Each EUV tool costs $150 million and requires multiple of his EUV scanners to be installed in the fab, creating additional costs for TSMC. Also, it takes a long time to manufacture chips on N5 and N3. This means higher costs for TSMC.
TSMC’s Alleged Wafer Prices
TSMC generally does not disclose wafer pricing except to actual customers. Also note that the contract price (likely used in large orders from companies like Apple, AMD, Nvidia, and even rival Intel) could be lower than the base price. It’s also important to do. Still, here’s what the report says about current prices:
|price per wafer||$20,000||$16,000||$10,000||$6,000||$3,000||$2,600||$2,000|
|node||N3||N5||N7||N10||N28||40 nm||90 nm|
Chip developers using TSMC’s services are expected to pass on the cost of new chips to downstream customers, making smartphones and graphics cards more expensive. Even now, Apple’s iPhone 14 Pro starts at $999, while Nvidia’s flagship GeForce RTX 4090 costs his $1,599. Once companies like Apple and his Nvidia adopt his TSMC’s N3 node, expect the products to become even more expensive.
Of course, the actual chip cost may still be relatively small compared to all the other components used in modern smartphones and graphics cards. His AD102 from Nvidia measures 608mm^2. A die-per-wafer calculator estimates that Nvidia can get around 90 chips from his N5 wafers, or a base cost of $178 per chip. Packaging, PCB cost, components, cooling, etc. all probably have at least double that impact, but the real cost lies in the R&D side of modern chip designs.
|Header Cell – Column 0||N3E vs N5||N3 vs N5|
|More Speed @ Same Power||+18%||+10% ~ 15%|
|Power Reduction @ Same Speed||-34%||-25% ~ -30%|
|logic density||1.7 times||1.6 times|
|HVM start||Q2/Q3 2023||Second half of 2022|
TSMC’s prices are so high for good reason, but currently there are no rivals that can produce chips at a reasonable yield and high productivity using state-of-the-art manufacturing technology, so TSMC’s can avoid it. volume. Officially, Samsung Foundry is ahead of TSMC in 3GAE process technology (3nm class, gate all-around transistor), but it is believed that it is only used for small cryptocurrency mining chips due to insufficient yields. . Meanwhile, Samsung Foundry’s 4nm-class process technology has fallen short of expectations when it comes to performance.
If Samsung and Intel Foundry Services offer process technology that surpasses TSMC’s (if any), the world’s largest foundries will have to cap their prices somewhat, but competition in the foundry market as a fab will increase. Therefore, we do not expect the price of chips to fall. are becoming more expensive, chip development costs are rising, and manufacturing techniques are becoming more complex.
Generally, in the mid-2010s, when Intel, GlobalFoundries, Samsung, TSMC, and UMC adopted FinFET transistors, the cost of manufacturing chips at leading-edge nodes began to rise rapidly. At the time, despite fierce competition among contract semiconductor manufacturers, costs rose for everyone.
TSMC’s first client with N3 is expected to be Apple, which can afford to develop a suitable SoC, mass-produce it, and make money on the hardware. Apple hasn’t indicated what kind of processor the company plans to make with the N3, but the current M2 and A16 Bionic follow-ups seem to make sense. will likely stay away from TSMC’s N3 for the time being due to its prohibitive cost, and instead use a chiplet-based design due to lower development, risk and manufacturing costs.
Again, it should be noted that TSMC has not commented on the estimate, nor has it commented on the information that it will charge around $20,000 per N3 wafer. These numbers are from industry insiders and may or may not reflect actual mass production prices.