SEM Based Defect Classifier For VSB Mask Writer

A SEM-based DL diagnosis system
For identifying VSB mask writer defects

- Ajay Baranwal, et. al.
© Copyright 2021 Center for Deep Learning in Electronics Manufacturing, Inc.Confidential and Proprietary to Center for Deep Learning in Electronics Manufacturing and its IP Members.
VSB writer system is a complex & reliable system
2
© Copyright 2021 Center for Deep Learning in Electronics Manufacturing, Inc. NuFlare confidential and proprietary
Rarely, something can go wrong
Need to diagnose and fix defects ASAP
3
A quick overview of the VSB mask writer
4
A quick overview of the VSB mask writer
5
Mask vendors to diagnose and fix defects ASAP
To provide best customer service and support
Writer defects need to be diagnosed and fixed
ASAP
Root cause the defect; repair and/or replace parts
Shape errors
Position errors
Dose errors
There are two SEM defect classification methods
To identify the writer’s defects
Note: Data not to scale
For illustrative purpose
Single Die reticle example 2x2 Die-array example
Die-to-database method Die-to-die method

Compares with corresponding CAD data Compares two Dies
8
D2DB example D2D example
Reference CAD & defect SEM Reference SEM & defect SEM
Note: Data not to scale
For illustrative purpose
CAD & SEM images SEM images
Die-to-database classifier Die-to-die classifier

Compares with corresponding CAD data One Die acts as a reference
9
Hard for humans to classify the error
A Die-to-database example
Reference CAD Defect SEM
10
Hard for humans to classify the error
Needs overlaying, zooming-in/out; still hard due to noise, LER
Reference CAD Defect SEM
11
Yet another case: hard to classify
A Die-to-die example
Reference SEM Defect SEM
12
Yet another case: hard to detect and classify
Needs overlaying, zooming-in/out; still hard due to noise, LER
Reference SEM Defect SEM
13
We built DL models for classification methods
DL works great on image patterns ‒ match, differentiate, group, etc.
Reference CAD images
Deep NN Position error Dose error

Trained on millions of
DT created synthetic SEM
Shape error Normal
Type of error probability
Defect SEM images
© Copyright 2021 Center for Deep Learning in Electronics Manufacturing, Inc. NuFlare confidential and proprietary 14
We built DL models for classification methods
DL works great on image patterns ‒ match, differentiate, group, etc.
Reference CAD images
e rs
e t e s
a m
r im win a g
p a t
N M
E ita l
2 M S i g
, K
1Deep 5NN0 a d
e r s 8 h
Trained on millions of
t
Position error Dose error
a y th i
DT created synthetic SEM
i w
6 l w l y
2 ne d o n
r a i l e
T sib
Shape error Normal
o s Type of error probability
Defect SEM images

P
Though, design CAD and SEM need alignment
A new DL model; algorithmic methods limited for different domain images
Image alignment
Misaligned CAD & SEM Aligned CAD & SEM
D2DB DL model shows ~ 93% accuracy
For different CD-SEM machines; visualization shows confidence
D2D DL model also show ~ 91% accuracy
For different CD-SEM machines; visualization shows confidence
A SEM-based defect classifier system: D2DB & D2D
Possible with several DL models and Digital Twins to synthesize data
Reference CAD images Reference SEM images
a. Position error b. Dose error
c. Shape error d. Normal

Defect SEM images Defect SEM images
Die-to-database Die-to-die
classification method Error classification
classification method
Mask vendors to diagnose and fix defects ASAP
To provide best customer service and support
1% 9% 2%
92. acy 85. acy 90. acy
c c ur c c ur c c ur
A A A
Dose error Shape error Position error
Mask vendors to diagnose
ef e c t and fix defects ASAP
d
To provide best customer
No uracyservice and support
cAc 2%
96.
1% 9% 2%
92. acy 85. acy 90. acy
c c ur c c ur c c ur
A A A
Dose error Shape error Position error
Mask vendors to diagnose
ef e c t and fix defects ASAP
d
To provide best customer
No uracyservice and support
c Ac 2%
96.
e rs
e t e s
a m
r im win a g
p a t
N M
E ita l
2 M S i g
1% , K
1 855.90% y a d 2%
92. acy
ur e r s 8 rac
cc ith
90. acy
ur
A c c
a y i th A
u
w A c c
6 l w l y
2 ne d o n
r a i l e
T sib
Dose error
o s Shape error Position error
P
DL is great for SEM analysis!
DL augments human taking faster and accurate decisions with large data
SEM denoising
CAD2SEM digital twins SEM2CAD digital twins
Filter good quality SEM CAD to SEM Image alignment VSB writer defect classification
23
24

SEM Based Defect Classifier For VSB Mask Writer

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SEM Based Defect Classifier For VSB Mask Writer

Uploaded by

Copyright:

Available Formats

A SEM-based DL diagnosis system

For identifying VSB mask writer defects

Single Die reticle example 2x2 Die-array example

Die-to-database method Die-to-die method

CAD & SEM images SEM images

Die-to-database classifier Die-to-die classifier

Reference CAD Defect SEM

Reference CAD Defect SEM

Reference SEM Defect SEM

Reference SEM Defect SEM

Deep NN Position error Dose error

Shape error Normal

Type of error probability

Defect SEM images

o s Type of error probability

Defect SEM images

Misaligned CAD & SEM Aligned CAD & SEM

a. Position error b. Dose error

c. Shape error d. Normal

Dose error Shape error Position error

Dose error Shape error Position error

You might also like