Compared to it’s N5 node, N3 promises to improve performance by 10-15% at the same power levels, or reduce power by 25-30% at the same transistor speeds. Furthermore, TSMC promises a logic area density improvement of 1.7x, meaning that we’ll see a 0.58x scaling factor between N5 and N3 logic. This aggressive shrink doesn’t directly translate to all structures, as SRAM density is disclosed at only getting a 20% improvement which would mean a 0.8x scaling factor, and analog structures scaling even worse at 1.1x the density.
The figures are nonsense, and are made up by an analyst paper using dubious investment costs and formulas: https://cset.georgetown.edu/wp-content/uploads/AI-Chips—What-They-Are-and-Why-They-Matter.pdf
We know for a fact the costs (from actual cost data) are far below those figures.
I don't have permission / we'll have to clear it first.Can you share the wafer cost? I don't believe cost is common knowledge.
That table also has a huge post that is "other costs and mark up". Hmm. Plus depreciation assumptions, plus...The figures are nonsense, and are made up by an analyst paper using dubious investment costs and formulas: https://cset.georgetown.edu/wp-content/uploads/AI-Chips—What-They-Are-and-Why-They-Matter.pdf
We know for a fact the costs (from actual cost data) are far below those figures.
Unless TSMC have changed their business model, these prices do not include the cost of the wafers - that would be a separate cost. These are TSMC's foundry costs. The wafer choice that customers specify (and pay for) must meet TSMC's requirements but if you want to use cheap wafers at the risk of lower yields, that's up to the customer. I would image for consoles, the choice of wafer is likely something determined between AMD and TSMC based on trial runs. This is the expertise that Microsoft and Sony are paying for.
Iterating on a previous comment, if you get permission to state (processed but not including mask cost) wafer costs, it would be great. Figures circulating on the web is typically ”analyst estimates” which vary both in quality and in underlying assumptions. Once you plow through enough of those, it may yield a decent feel for where costs lie. Or not, if the numbers all rely on some dubious or simply old data point.The figures are nonsense, and are made up by an analyst paper using dubious investment costs and formulas: https://cset.georgetown.edu/wp-content/uploads/AI-Chips—What-They-Are-and-Why-They-Matter.pdf
We know for a fact the costs (from actual cost data) are far below those figures.
Wasn't it reported already a while back that TSMC has ordered crapload of tankers to combat the water shortages?
They've contracted for water capacity transport by tankers if needed but AFAIK, they aren't using it. They just have the contracted capacity that they can exercise in the future if needed. IIRC, they've also got contracts in place to be able to FLY in water (expensive as hell) if absolutely needed.
Regards,
SB
Easy to see why they're putting foundries elsewhere, though I wonder if "we have weeks long blackouts" Texas will really get the nod from anyone.
This news comes from several Taiwanese business news publications citing sources at, or in business dealings with, the world's biggest contract foundry - TSMC. Depending on the product/process, TSMC's customer can expect to pay up to 20 per cent more for their chips.
...
A wide range of influences have come together to inspire the price rises. What you might call and 'imperfect storm' of factors have lead TSMC to its price rise decision. There are so many factors listed by the sources that I've decided to bullet point them:
The various sources have slightly different stories about the impending TSMC foundry price hikes. Mostly the reports are quite similar, and seem to point to the TSMC price increases being applied to contract work from Q1 2022 onwards.
- TSMC's gross margin has fallen under the industry standard 50 per cent
- TSMC has started several large and costly expansion initiatives (large new factories abroad, and advanced facilities in Taiwan) and has to service investor return
- Demand is up and shortages are intensifying
- Raw materials prices have increased
- Freight rates have increased
- Other inflationary and geopolitical pressures at play
Another area where the reports seem to agree is that the price rises will apply differently across different processes. Specifically, customers of advanced processes (sub 12nm) will be facing price increases of up to 10 per cent. Meanwhile, customers reliant on the more mature processes (16nm+) will be looking at 10 to 20 per cent price increases.
TSMC is targeting Intel on its home turf by constructing a facility in Arizona. The next generation iPhone will use chips manufactured on 3nm technology, and Apple will be the first company to use wafers from this factory. NVIDIA is getting early wafers for the first time, however it will likely get 4nm instead of 3nm.
...
TSMC plans to construct six factories in Arizona, each with a monthly output of 100,000 wafers. Taiwan Semiconductor Manufacturing Company officials will officially break ground on the new facility this week in Arizona. Attendees will include Vice President Joe Biden, Apple's Tim Cook, NVIDIA's Jensen Huang, and AMD's Dr. Lisa Su, all of whom will be present to mark the arrival of the first machines on campus.
...
TSMC founder Chang has confirmed that TSMC will bring the latest 3-nm chip production to the U.S. but also said that manufacturing in the U.S. will be 50% more expensive than in Taiwan. TSMC's current production base for 3-nm chips is in the southern Taiwanese city of Tainan, where production using the technology only recently began. Samsung and Intel are also massively expanding chip production footprints in the U.S. Samsung is building a $17 billion chip facility in Texas, while Intel is constructing a new $20 billion plant in Arizona and another $20 billion plant in Ohio. Charles Shi, a semiconductor analyst with Needham & Co., said TSMC should put a higher priority on its new plants and expansions in Arizona and Japan.
Erm, the first Arizona "4nm" fab is supposed to come online 2024 and the "3nm" one in 2026.Apple and NVIDIA first customers of TSMC 3nm/4nm chips made in new USA based Arizona Plant
TSMC is targeting Intel on its home turf by constructing a facility in Arizona. The next generation iPhone will use chips manufactured on 3nm technology, and Apple will be the first company to use waf...www.guru3d.com