I struggle to follow along with this. I understand Moore's law, but at what point can we just keep using 7nm for most things instead of always moving down an integer 3>7. How many things will require a 2nm and not be satisfied by a 7nm fab?
back
2 comments
The issue is you can make this claim anywhere over the past 50 years. When you just achieve a new node, you think it isn't yet useful. But what happens is that software starts to take advantage of it and it gives you new types of capabilities you didn't have before.
For example, if we never got down to where we are now, we likely wouldn't have LLMs or viable real-time ray tracing or the ability to run various types of photo improvement algorithms on your iPhone to make all your pics look good. Those require significant processing and it wouldn't have been achievable if technology here just stopped a decade ago.
If we had stopped CPU progress 25 years ago we would have never gotten smart phones at all.
It is hard to know what will be possible in a decade with all the additional compute resources we get from moving down more nodes. But I do expect there to be a lot.
Depends how you count. Per design we already use >7nm for way more things because older processes are cheaper. But per $ (and perhaps also in volume) the leading-edge processes get the lions share.