Hi
In the book "ULSI Process Integration 5" I've found some detailed description 
of the usage of dummy-gates for gate-last manufacturing.

https://books.google.com.hk/books?
id=o3zXAiskEYUC&lpg=PA319&ots=sD2_WDEV6A&dq=gate-last%20cmos%20dummy-
gate&pg=PA319#v=onepage&q=gate-last%20cmos%20dummy-gate&f=false

There are also pros and cons listed for gate-last vs. gate-first

What do you folks think? What's the smarter choice?

Cheers

On Tuesday, 6 March 2018 10:14:05 PM HKT David Lanzendörfer wrote:

Hi

How do you want to handle that later with smaller nodes?

For that it would be interesting to have access to the full IEEE document
here: http://ieeexplore.ieee.org/document/5984621
They apparently made it work.

I think they all have the problem of aligning gates, isn't it?

This document thematizes that very well.
http://electroiq.com/blog/2010/03/integrating-high-k

Gate-first-gate-last problem, is exactly what we have here right now.
If we use poly for the gate first we will have a problem with grainyness but
if we do gate-last with Aluminum we will have a alignement issue.

Or that method for Aluminum:
http://km2000.us/franklinduan/articles/images/high_k88.jpg

We can use a dummy gate and then use the damazene method to add the actual
Aluminum gate electrode.
I'm actually eager to go for this, because poly silicon is very grainy as
I've explained to George today over some beers.
It will be a problem to have exact geometric structures with 28nm when using
poly for the gate electrode.
Also we loose the "4.1 volt advantage" when it comes to threshold drops.
(E_Al = V_M*q = 4.1 eV -> V_M = 4.1V)

Cheers
	David