A 2nm-capable wafer fab will come with a $28 billion tag, up to 50% more costly than 3nm

emorphy

Posts: 64   +0
Staff
Bottom line: That is not a typo: a 2nm-capable fab could cost around $28 billion. In fact, costs throughout the 2nm development and production process will be elevated as the tools will be more complex and the talent needed will be more expensive. One saving grace might be AI-enabled EDA tools that can streamline processes and help reduce costs.

A 2nm-capable semiconductor fabrication plant that has a capacity of 50,000 wafer starts per month, or WSPM, will cost around $28 billion, according to consulting firm IBS. That is $8 billion more than the cost for a 3nm fab and just one illustration of the exponentially higher expenses that firms can expect as the industry moves to the next generation of chips.

To be precise, 2nm chip costs will rise around 50% compared to 3nm processors, IBS says, meaning that companies such as Apple will have to spend $30,000 to process a single 300mm wafer using TSMC's N2 fabrication process when it is introduced in the next few years. Possibly, though, there is some wiggle room in these numbers, potentially bringing down the expected high cost of these chips.

There are in fact numerous approaches that semiconductor companies can take and an array of design decisions they can make during the pre-construction, construction, and operations phases that will materially alter the final cost of the fab.

Also read: How CPUs are Designed and Built

To be sure, the costs of chip development are nothing to scoff at. Software development alone accounts for $314 million and verification is another $154 million, IBS figures show. Also, designing chips at the 2nm node requires specialized talent, which is in short supply. Then there is also the increased use of photolithography, a process used to create the patterns on a chip's surface.

The smaller the features on a chip, the more precise the photolithography process needs to be, thus driving up the cost of the equipment and the materials used in the process. And that 2nm-capable fab that will clock in at $28 billion? Driving the $8 billion cost differential is the increased number of EUV litho tools required to maintain a 50,000 WSPM capacity.

But even IBS acknowledges that there are nuances behind such figures and the evolving landscape of chip design. It has estimated that it can cost a company $725 million to build a sizable 2nm chip from scratch. But that is by a company without pre-existing intellectual property and the reality is many semiconductor companies, particularly startups, pursue more efficient strategies.

IBS also points out that the role of AI-enabled EDA tools is becoming increasingly crucial in chip design, streamlining processes and reducing costs by automating complex design processes and optimizing chip performance.

Permalink to story.

 
Just make an AI monkey to build and operate the new factory, LOL!
And then AI beings to purchase CPUs since people will be jobless and too poor to afford them.
It will be pretty much 2 kinds of intelligent beings, the advance race of AI, and more primitive humans.
 
I am really looking forward to buying iPhone 16 Ultra for $2000
Even though I may not notice any difference in speed compared to my previous iPhone 15 Pro, the feeling of relevancy is important. My friends would laugh at me and banish me from their inner-inner circle and I would remain an orbiter. The increased price would also contribute towards helping Apple to innovate even more. They really need money in such harsh times.
 
I am really looking forward to buying iPhone 16 Ultra for $2000
Even though I may not notice any difference in speed compared to my previous iPhone 15 Pro, the feeling of relevancy is important. My friends would laugh at me and banish me from their inner-inner circle and I would remain an orbiter. The increased price would also contribute towards helping Apple to innovate even more. They really need money in such harsh times.

You obviously have no clue on how fabs are running into physical limits of making and designing chips.
 
In my opinion apple does have sophisticated hardware and a solid software eco-system. They do sell "normal" phones and phones with the latest tech for a premium price.

Nvidia is doing the same. It's not like your forced to buy those products either.
 
Totally moronic. AMSL themselves have shown that dual patterning with low NA 0.33 optics is far cheaper and can produce smaller features size than single pass 0.55 NA optics. High NA is only superior in cost at the 1.4nm level which isn't until 2030+ or so. I love how Intel is rushing into the high NA era and willing to purchase 6 of the new high NA machines from AMSL at over $400M each.
 
Ok. Building the fab is 50% more expensive. But if you go from 3nm to 2nm with essentially the same design, you would get more chips out of a wafer, no?
It will not be 50% more, but it wouldn't make sense to go smaller and not make more chips out of the same 300mm wafer. So, in part that would offset the extra cost?
 
You can go denser by a smaller node, which means more transistors into a single chip.

Going smaller has benefit of a lower power consumption at the same design, or higher clocks within the same power target.
 
I love how Intel is rushing into the high NA era and willing to purchase 6 of the new high NA machines from AMSL at over $400M each.
That might be one of the first good moves Intel has made under Pat Gelsinger. If they can catch up or even outdo TSMC it would do their fab ambitions a lot of good. Now would be a good time with all the grants trying to move manufacturing out of Taiwan.
 
TSMC is going to price themselves out of the market. Market leaders currently but pumping prices so much will push chip designers to go else where eventually. Samsung and intel will likely do well out of this news.
 
TSMC is going to price themselves out of the market. Market leaders currently but pumping prices so much will push chip designers to go else where eventually. Samsung and intel will likely do well out of this news.

Unlikely. TSMC have cheaper options available and those who want best and can pay, have no other options.
 
That might be one of the first good moves Intel has made under Pat Gelsinger. If they can catch up or even outdo TSMC it would do their fab ambitions a lot of good. Now would be a good time with all the grants trying to move manufacturing out of Taiwan.
Intel don't have the money and as their market share falls in server, they won't have the volume to amortize costs of the new nodes. Something has to change as the cost of each new node rises exponentially.
 
Back