Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/stochastics-yield-killing-gap-no-one-wants-to-talk-about.23254/page-2
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Stochastics: Yield-Killing Gap No One Wants to Talk About

So, will it be used for increasing the throughput at same dose, or increasing the dose at same throughput? Again, the net benefit of increasing the dose depends on the resist.
Yes I think effectively at each new node they increase the dose at the same throughput to hit their dimension targets (since as you know the stochastic issues get more difficult the smaller the feature size).
 
The dose increase required to get the photon shot noise to pre-EUV levels will be exorbitant: https://semiwiki.com/lithography/357643-facing-the-quantum-nature-of-euv-lithography/

DUV vs EUV dose fluctuation table.png
 
Last edited:
Well if they didn't address the stochastics then it would hit yield, right? And if they don't address the stochastics, and it doesn't hit yield, then it's not really a problem, is it?

Despite continuous improvements in the performance of EUV photoresists, EUV masks, and post-lithography processes, stochastic defects, or stochastic failures – space bridges and line breaks - are still a major factor of yield loss in EUV production.

The detection of stochastic defects is therefore integral to EUVL. However, detecting, characterizing, and repairing stochastic defectivity using experimental methods alone is expensive, requiring a lot of wafers, metrology resources and time.

Experimental methods for measuring stochastic defectivity and validating model predictions require wafer inspection by e-beam microscopes suitable for use on large areas. But the statistics of hot spot detection in EUVL complicate experiments and drive costs up since the number of repetitive measurements needed to observe a defect can be exceptionally large depending on probability of occurrence and the number of instances of the hotspot in the layout.

A simulation-based failure probability model is proposed in this paper as a virtual inspection.

 
Last edited:
Experimental methods for measuring stochastic defectivity and validating model predictions require wafer inspection by e-beam microscopes suitable for use on large areas. But the statistics of hot spot detection in EUVL complicate experiments and drive costs up since the number of repetitive measurements needed to observe a defect can be exceptionally large depending on probability of occurrence and the number of instances of the hotspot in the layout.

A simulation-based failure probability model is proposed in this paper as a virtual inspection.


My impression is that ASML is working hard to speedup the EUVL defectivity detection by eBeam inspection of larg(er) scale. I expect more experimental progress as well in the coming years, to complement the computational approach:




Nice Figure from that 2025 SPIE paper:

1755578356024.jpeg
 
Back
Top