Shift? add? Rumor.
1 Like
I don’t see AMD buying trouble. The WCCFTECH article has a guess at AMD going for the new 4LPP+ process instead of the current 4LPP node. AMD used to be a company which would grab for the best node they could get. But they have learned. AMD was already developing at both 4 and 3 nm nodes since TSMC had trouble with its 3 nm node. If you look at what both Samsung and TSMC, they are hitting the limit, which was expected at 5 nm. You need a certain number of atoms in a gate or other parts to have a transistor. Think of it this way, you need enough atoms to get doping atoms where needed. Z has to go up to get the X and Y dimensions where you want. Going from tungsten to copper vias bought some breathing room.
A different way to see it is that you need at least N dopant atoms. Statistically, you can overload the dopants, but with billions of transistors per chip, you will still have a lower overall yield due to at least one critical transistor not having enough dopants.
Is it possible to do better? Sure. Start with undoped silicon, and inject the dopant atoms one at a time. The technology exists, but it will be a long time before the cost is low enough.
1 Like
So now it sounds like Samsung 3nm may have something to offer?
You raise an interesting point about AMD having learned from grabbing the top node and finding it wasn’t really ready… “Offer me a path to production that will work.” Not miracles or swinging for fences.
The original Zen kind of looked like a swing for the fences… or I guess maybe that was Zen 2 with the jump from 14nm or 12nm to 7nm instead of 10nm. After that it’s held back a bit, I thought for reasons of cost and not wanting to battle Apple or whoever for the most expensive node and lowest capacity. I suppose that also protects them from FAIL on the top node.
3 Likes