Post on 19-Apr-2020
Sophisticated algorithms used in the areaof defect pattern recognition on wafers
SDS - Signature Detection System Thomas Kreutzmann
AMD Fab 36 LLC & Co.KG - Dresden, Germany
AEC/APC Europe, April 2008
2 3/20/2008
� Introduction / General Overview
– Task of Inline Defect Metrology at AMD Fab 36
– Why need an automated Signature Detection System?
� Goals of Project
� System Overview
– Data Flow
– Operation and Processing
– Clustering and Classification algorithms
– Features
Content
3 3/20/2008
What is “In-Line defect metrology” ? Line and equipment monitoring / In-line quality assurance
Process steps:- Film deposition- Lithography- Etch process
Process steps......
Process steps......
Process steps...
Time of processing: 4 to 12 weeks
In-Line defect metrology step:
Bare Si wafer start
Process end- electrical test
# of die per wafer - $$$
- Defect data / trends limit
- Images
- Wafer maps
Detect issues as early as possible
Prevent trailing lots from beingaffected by same issue
Run countermeasures based on “Trouble Shooting Guide”
4 3/20/2008
What is “In-Line defect metrology” ? Challenges for quality assurance in a fully automated Fab
No need - and almost no possibilities - for „real time“manual interventions by fab personnel under standardmanufacturing conditions
Fully automated manufacturing –also for metrology / defect metrology
Defect inspection steps
1) Automatic lot selection2) Automatic FOUP delivery3) Automatic recipe select4) Automatic wafer selection
Defect Data + ADCDefect Data + ADC
Defect review steps (SEM)
1) Automatic lot selection(indicated by OOC of scan)
2) Automatic FOUP delivery3) Automatic recipe select4) Automatic wafer selection
Defect review DataDefect review Data
5) Manual classificationclassification
5 3/20/2008
MQ Series
HTTPHTTP
CORBACORBA
DB2
APF/RTD
ActivityManager
CEI
Tool
SAPPHIRE
Ab Initio
LoaderARMOR
ASPECT
APC
CatalystFDC
ORACLE
ADAPT
MQ Series
SiViewMM
SMMMMM
Sch WD XMS
AMMO
AMHS-EI
RTD_API
SAP/PM TelAlert
MACS
AMHS
FabGUI
CORBA
MDS
PCMS
EMS
EPPM
eDR
HTTP
PPR
ePPCD
eLS
FabView
Yield Manager
MQ Series
HSMS
eSPEC
What is “In-Line defect metrology” ? Challenges for quality assurance in a fully automated Fab
Defect Data Volume:
~ 6 to 8 Mio defects per day~ 80.000 defect images per day
~ 7000 wafer scans per day
Cannot cover everythingbased on defect density or
ADC limits.
Inspect all wafermapsmanually for signatures?
Can raise error actions in „real time“ if signature
recognized?
Need an automatedsignature detection
system!
6 3/20/2008
� Introduction / General Overview
– Task of Inline Defect Metrology at AMD Fab36
– Why need an automated Signature Detection System?
� Goals of Project
� System Overview
– Data Flow
– Operation and Processing
– Clustering and Classification algorithms
– Features
Content
7 3/20/2008
Goals of Project
Automated detection and classification of signatures at wafer level…
Scratches
Stacked wafermap Particlesignatures
wafer edge “ring”
8 3/20/2008
Goals of Project
- Enable reliable automated detection of signatures
- Arcuated scratches
- Straight scratches
- Clouds
- Cluster
- Rings
- Ensure scalability and flexibility
- Multiple compute node system for fast feedback
- Maintainability through powerful setup GUI
- Seamless integration into MES environment
9 3/20/2008
Signature Examples
10 3/20/2008
Signature Examples
11 3/20/2008
Signature Examples
12 3/20/2008
Signature Examples
13 3/20/2008
Signature Examples
14 3/20/2008
� Introduction / General Overview
– Task of Inline Defect Metrology at AMD Fab36
– Why need an automated Signature Detection System?
� Goals of Project
� System Overview
– Data Flow
– Operation and Processing
– Clustering and Classification algorithms
– Features
Content
15 3/20/2008
FactoryControlSystems
Fab MES
SPC Software
Data Warehouse
...
System Integration / Data Flow
SDS Main Server
(Dispatcher)
...
additional compute nodes
YMS
Wafer Inspection Equipment
SDS EI
Tool EIs
DFS
1
2
3
4
5
5
5
16 3/20/2008
System Processing / Overview
Context based Filter &
Signature Select
Clustering
SDS Server Application
Zoning
Prioritization
Write
File
Search &
Parse
Classification
SDS GUI Application (Config. & Tuning)
Process end- electrical test
# of die per wafer - $$$
Process steps:- Film deposition- Lithography- Etch process
Process steps......
Process steps......
Process steps...
Bare Si Wafer start
In-Line defect metrology step
In-Line defect metrology step
In-Line defect metrology step
C1 C2 P
17 3/20/2008
System Processing / Context Matching
C1 C2 P
check for Sig.2 and Sig. 3* / Process 1 / Layer 2
check for Sig.2Prod 2 / * / *
check for Sig.1Prod 1 / Process 1 / *
ActionContext
Process end- electrical test
# of die per wafer - $$$
Process steps:- Film deposition- Lithography- Etch process
Process steps......
Process steps......
Process steps...
In-Line defect metrology step
In-Line defect metrology step
In-Line defect metrology step
1. Context Matching
SDS Context System
• Which wafer / KLAR files to check?
• What types of signatures to check?
• Context configuration based on product,
product group, process, layer
� Enhances system classification
performance
18 3/20/2008
System Processing / Context Matching
C1 C2 P1. Context Matching
SDS Context Configuration
• By a simple to use GUI
• Wild cards for powerful context matching
• Multiple signatures per context with
prioritization
• Different variants of signature models for
different contexts
• …
19 3/20/2008
System Processing / Clustering
2. Clustering
- Partitioning of data into subsets
- Density based DbScan algorithm by Martin Ester et al. fits in most cases
- Clustering parameters definable for every version of a Signature Model
C1 C2 P
20 3/20/2008
System Processing / Clustering
Density-Based Clustering- For each object of a cluster the neighborhood of
a given radius (ε) has to contain at least a minimum number of points (MinPts).
- An object o is direct density-reachable from an object q wrt. ε and MinPts in a set of objects D if: o ϵ N
ε(q) and Card(N
ε(q)) ≥ MinPts
- An object p is density-reachable from an object q wrt. ε and MinPts in a set of objects D, denoted as p >D q, if there is a chain of objects p1, …, pn ,p1 = q, pn = p such that pi ϵ D and pi+1is directly density-reachable from pi
- An object p is density-connected to an object q wrt. ε and MinPts in a set of objects D if there is an object o ϵ D such that both p and q are density-reachable from o.
- A cluster C wrt. ε and MinPts in D is a nonempty subset of D satisfying the following conditions:
1. Maximality ∀∀∀∀p,q ϵ D: if p ϵ C and q > D p then q ϵ C
2. Connectivity ∀∀∀∀p,q ϵ D: p is density-connected to q
Core point
Circle of a given radius ε
Noise
p
q
o
21 3/20/2008
System Processing / Classification
3. Classification
- Refinement of clustering based on model of geometric shape:
- Line segment � straight scratch
- Arc � polish scratches
- Ring � rings
- Polygon � spots, Clouds
- Two algorithms were implemented:
ACC = Adaptive Compensation Calculation
(Systema) applied to sharp contours
(like scratches and spots)
EM (Expectation Maximization) applied to
soft contours (like clouds and rings)
C2C1 P
22 3/20/2008
System Processing / Classification
EM = Expectation Maximization
- finding maximum likelihood estimates of parameters in probabilistic models, where the model depends on unobserved latent variables
- E-step computes an expectation of the likelihood by including the latent variables as if they were observed
- M-step which computes the maximum likelihood estimates of the parameters by maximizing the expected likelihood found on the E-step
- signatures to be found on wafer maps were modelled using geometric shapes
23 3/20/2008
System Processing / Classification
EM - Initialization
24 3/20/2008
System Processing / Classification
EM - Iteration
Start
Result
25 3/20/2008
System Processing / Classification
EM - Iteration
Result
26 3/20/2008
System Processing / Classification
EM - Result
27 3/20/2008
System Processing / Zoning
4. Zoning
- Possibility to attach a typical region on wafer for each signature type
- A signature will only be detected if it is mainly located inside a zone
- Useful to exclude areas with blurred defect information
- Speeds up recognition
PC1 C2
28 3/20/2008
System Processing / Zoning
4. Zone - ExamplesPC1 C2
Search here for Signature Type A only
Search here Signature Type C only
Search here for Signature Type B only
…
29 3/20/2008
System Processing / Prioritization
C1 C25. Prioritization
- Possibility to prioritize Signature Models at certain contexts
- Defects in super-posed signatures will be assigned to prioritized Signature Model (see next slide)
P
30 3/20/2008
System Processing / Prioritization
5. Prioritization – Example
- Priority of scratch model ishigher than priority of ring model at a given context
- Each defect will be assigned to one defect class only
- Thus, defects belonging to scratch will be classified as such, even though they residein the proximity of the ring
31 3/20/2008
System Integration intoFactory Control System
- Use classification data in Yield Management System
- Feed SDS results via modified Equipment Interface intoMES/SPC System
- Send instant notification via Defect Feedback System (DFS)
32 3/20/2008
System Integration intoFactory Control System
Yield Management System
„P“ indicates classdata of pattern
recognition source
33 3/20/2008
System Integration intoDefect Feedback System (DFS)
- Flexible definition of rules and recipients for notification- Instant wafer history
34 3/20/2008
Thanks...
... to all who contributed to SDS development!
Special thanks to:
Jens Klobes, Robert FenskeLead Developers Systema
Dr. Remo Kirsch, Dirk Jung, Dr. Manfred Heinz, Nico NoackKey Customers/AppAdmins AMD Fab36/CFM
35 3/20/2008
Backup
36 3/20/2008
SDS Technical Features
Server Application
� Java based
� Service Oriented Architecture
� Application Server (Apache Tomcat)
� Oracle Database (for persistence data, statistics)
� Scalable by additional compute nodes
� OS independent (Solaris, Windows, Red Hat Linux)
GUI
� Fast and easy configuration, pattern detection & performance monitoring, training & tuning
� RCP (Eclipse) based
� User roles and privileges
� LDAP security interface & management
37 3/20/2008
Trademark Attribution
AMD, the AMD Arrow logo and combinations thereof are trademarks of Advanced Micro Devices, Inc. in the United States and/or other jurisdictions. Other names used in this presentation are for identification purposes only and may be trademarks of their respective owners.
©2006 Advanced Micro Devices, Inc. All rights reserved.