TSMC killed 450mm wafers for fear of Intel, Samsung

theregister.com

19 points·oumua_don17··6 comments

From a competitive perspective this is probably accurate. However the technological roadblocks of 450mm wafers were sizable and not easily overcome. Wafer flatness across an 18 wafer is extremely important. The industry handled this same issue when going from 200mm to 300mm, but the machines were significantly more complex and cost ~10x their 200mm counterparts. 450mm machines were expected to be on the order of another 10x their 300mm counterparts. That means a EUV tool that is currently ~$100M/each would be closer to a billion dollars. Metrology tools were way behind in development too...since KLA was not keen on doing 450mm and only TSMC and Samsung were pushing them to develop.

Spending 10x to get double the yield (not really true, just double the chips/wafer) is not economically feasible either.

What exactly causes the increase in complexity when dealing with larger wafers?

Outside of the production of the larger wafers (larger crystals need to be grown which is difficult, and then sliced into wafers), factories need to be retooled to physically move and transport the larger and heavier wafers.

https://www.youtube.com/watch?v=0s5TO9h6fco

This video was recently published that talked about going from 200 to 300mm wafers, and there is some further discussion in the comments about the transition as well as going to 450.

> Wafer flatness across an 18 wafer is extremely important

Could they correct it by adaptively deforming the mounting surface, assuming the changes in thickness are smooth, adaptive optics style?

If it's going to cost a billion to make tools for 450mm wafers, why not just jump to 600mm, they'd get more chips which should make it economic. We just aren't making enough chips at the moment.

Reply on news.ycombinator.com