Yes, it’s not cheap. More typical to switch on the next node iteration. Rare to see fabless guys switch production fabs with the same product or to split production of same product design between multiple fabs..
Last edited:
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
Chinese semiconductor industry
- Thread starter Hendrik_2000
- Start date
- Status
- Not open for further replies.
This is also why you don’t see people swapping back and forth between fabs that frequently for advanced node for the same exact product.
Apple and Qualcomm did source the same chip to both TSMC and Samsung for advanced node from time to time
Apple switched but not mid-product.Apple and Qualcomm did source the same chip to both TSMC and Samsung for advanced node from time to time
Qualcomm did mid-product with Snapdragon 8 Gen1 with Samsung to Snapdragon 8+ Gen 1 with TSMC, which I mentioned earlier. It is the first time we had an actual Apple-to-Apple comparison between foundries. But this is due to Samsung’s low yield…if their yield had been good, switching in middle of the same product wouldn’t make economical sense.
Last edited:
Are EDA tools for different nodes very different from each other, e.g. 14nm vs 7nm or is it just a matter of integrating new foundry PDKs with a new node?Yes, it’s not cheap. More typical to switch on the next node iteration. Rare to see fabless guys switch production fabs with the same product or to split production of same product design between multiple fabs..
Just like manufacturing equipment, it’s not exactly the same but just new features added on top of what we used for previous node. But some times more complex additions needed to address revolutionary changes such as going from planar to finFET, finFET to GAA, application of Contact Over Active Gate (COAG), DDB to SDB, these require more significant changes to the EDA.Are EDA tools for different nodes very different from each other, e.g. 14nm vs 7nm or is it just a matter of integrating new foundry PDKs with a new node?
this is mostly how everybody market their product so to have steady revenue stream going from one node to the next. Anything that is good enough across multiple nodes means no steady stream of revenue. Few equipment or software are ever capable across multiple nodes without some sort of extra upgrade.
like when people here thinks a 28nm capable litho tool cam be used for 7nm, that’s just outright incorrect. A 7nm capable litho tool could be used for 28nm, but not vice versa. A 7nm capable scanner has much more extra capability than a 28nm capable one; so much so that the former is priced almost 2 to 3 times more than the latter.
I'm not sure you've explained that well. A 22nm single pass machine is what SMIC is using for 7nm from my understanding. It does this with multiple passes and masks.Just like manufacturing equipment, it’s not exactly the same but just new features added on top of what we used for previous node. But some times more complex additions needed to address revolutionary changes such as going from planar to finFET, finFET to GAA, application of Contact Over Active Gate (COAG), DDB to SDB, these require more significant changes to the EDA.
this is mostly how everybody market their product so to have steady revenue stream going from one node to the next. Anything that is good enough across multiple nodes means no steady stream of revenue. Few equipment or software are ever capable across multiple nodes without some sort of extra upgrade.
like when people here thinks a 28nm capable litho tool cam be used for 7nm, that’s just outright incorrect. A 7nm capable litho tool could be used for 28nm, but not vice versa. A 7nm capable scanner has much more extra capability than a 28nm capable one; so much so that the former is priced almost 2 to 3 times more than the latter.
I think what your saying is a machine like the 1980i which is single pass of 28nm couldn't make a 7nm chip, but a 2050i could do 28nm.
But it's confusing because lithography machines ratings and what they're capable of isn't a simple 1:1 like your comment implied.
A NXT1950 is a 28nm rated scanner. It can handle only the most critical layer of 28nm, but can’t handle the most critical layer of 7nm.I'm not sure you've explained that well. A 22nm single pass machine is what SMIC is using for 7nm from my understanding. It does this with multiple passes and masks.
I think what your saying is a machine like the 1980i which is single pass of 28nm couldn't make a 7nm chip, but a 2050i could do 28nm.
But it's confusing because lithography machines ratings and what they're capable of isn't a simple 1:1 like your comment implied.
A NXT2000 is a 7nm rated scanner because it is capable of handling the most critical layer of 7nm process. A NXT2000 can be used for 7nm and 28nm.
what I’m saying is, if a scanner is capable of doing 28nm AND 7nm, we would’ve call it a 7nm litho tool.
So when SMEE SSA800, the so called 28nm litho scanner, it’s exactly that, a 28nm scanner. If it’s intended to have all the bells and whistle to handle 7nm, they would’ve called it a 7nm scanner instead of calling it a 28nm scanner. My main point is there are quite a big difference between the so called 28nm scanner vs 7nm scanner.
It’s not only about resolution limit or how many litho passes. If that’s the case NXT1950 and NXT2000 could be used interchangeably…both have same resolution capability on paper and both could be used for multiple litho pass.
the thing is people here think it’s all about the resolution so double patterning with NXT1950 is the same as double patterning in NXT2000. But it’s not that simple. And “I did not explain it well” because I didn’t explain the details at all. I merely pointed out 28nm vs 7nm scanners are not the same.
and scanner model to node rating IS that simple 1:1. With each more advanced nodes comes tighter specification in order to reach satisfactory yield, each progressive model was designed with features and capabilities to fulfill each node’s requirement. So it is indeed really that simple of 1:1 scanner model to node.
Last edited:
A NXT1950 is a 28nm rated scanner. It can handle only the most critical layer of 28nm, but can’t handle the most critical layer of 7nm.
A NXT2000 is a 7nm rated scanner because it is capable of handling the most critical layer of 7nm process. A NXT2000 can be used for 7nm and 28nm.
what I’m saying is, if a scanner is capable of doing 28nm AND 7nm, we would’ve call it a 7nm litho tool.
So when SMEE SSA800, the so called 28nm litho scanner, it’s exactly that, a 28nm scanner. If it’s intended to have all the bells and whistle to handle 7nm, they would’ve called it a 7nm scanner instead of calling it a 28nm scanner. My main point is there are quite a big difference between the so called 28nm scanner vs 7nm scanner.
Wait a minute. ASML did not call their machine XXnm machine，they never say NXT1950 is 28nm machine and NXT2000 is 7nm machine.
BTW, ASML already stopped producing NXT1950，the lowest grade machine in ArFi category they offer now is 1980 series. That is according to ASML website
That's just not true. Let's look at ASML website. No where does it say rated for 7nm. Instead it says less than 38nm.A NXT1950 is a 28nm rated scanner. It can handle only the most critical layer of 28nm, but can’t handle the most critical layer of 7nm.
A NXT2000 is a 7nm rated scanner because it is capable of handling the most critical layer of 7nm process. A NXT2000 can be used for 7nm and 28nm.
what I’m saying is, if a scanner is capable of doing 28nm AND 7nm, we would’ve call it a 7nm litho tool.
So when SMEE SSA800, the so called 28nm litho scanner, it’s exactly that, a 28nm scanner. If it’s intended to have all the bells and whistle to handle 7nm, they would’ve called it a 7nm scanner instead of calling it a 28nm scanner. My main point is there are quite a big difference between the so called 28nm scanner vs 7nm scanner.
It’s not only about resolution limit or how many litho passes. If that’s the case NXT1950 and NXT2000 could be used interchangeably…both have same resolution capability on paper and both could be used for multiple litho pass.
the thing is people here think it’s all about the resolution so double patterning with NXT1950 is the same as double patterning in NXT2000. But it’s not that simple. And “I did not explain it well” because I didn’t explain the details at all. I merely pointed out 28nm vs 7nm scanners are not the same.
and scanner model to node rating IS that simple 1:1. With each more advanced nodes comes tighter specification in order to reach satisfactory yield, each progressive model was designed with features and capabilities to fulfill each node’s requirement. So it is indeed really that simple of 1:1 scanner model to node.
And no one is calling the 2050i a 7nm rated lithography machine, particularly since no one uses it for such besides SMIC. So the ratings are definitely not 1:1. Unless ASML itself is lying to us. Which would be super weird since that would be saying their machines do worse than they can.
Agree. But it doesn’t contradict with the fact I shared about the minimum configuration used for each node.Wait a minute. ASML did not call their machine XXnm machine，they never say NXT1950 is 28nm machine and NXT2000 is 7nm machine.
BTW, ASML already stopped producing NXT1950，the lowest grade machine in ArFi category they offer now is 1980 series. That is according to ASML website
using something much more capable for a given application like NXT1980 for 28nm is allowed. Not efficient and an overkill, but no one would stop you from doing so.
everything was created for a specific purpose, no law that says you can use it for something less critical.
- Status
- Not open for further replies.