Significant challenges for various Critical Dimension (CD) measurement matching procedures are reaching a comparable complexity as result of negative effects of roughness on the features. Due to the constant trend of integrated circuit in features reduction, impact of roughness start to be more destructive for various sets of measurement algorithms. Commonly used attempts to increase magnification for pattern recognition in measurement mode could in turn detect higher deviation from predefined patterns and thus initiate shift in placement of measurement gate. The purpose of this paper is to discuss how to reduce measurement gate (MG) placement variation impact and filter acquired data using edge correlation approach. The essence of listed above approach is to create set of width correlation function represents particular feature under test and compare it to “golden” one as a mean of detection of uncorrelated scans, which in turn should be excluded from overall computation of matching results. We describe general approach for algorithm stepping and various techniques for judgment of measurement comparison validity. Presented approach also has particular interest in determination of specified tool performance for predefined pattern recognition feature as well as for pattern recognition algorithm robustness study - direct interest for manufacturer. Precise matching estimation as part of Round Robin (RR) routines creating possibility to work with restricted amount of data and perform quick reliable qualification procedures. This paper concentrated on practical approach and used both simulation and actual data measurements data before and after proposed optimization taken by various generation tools by Hitachi (S-8840, S-9300, S-9380) in production environment

1. Image Resolution Monitor
Vladislav Kaplan, September 2005

2. 2
Resolution monitor
Proposition:
Replace daily repeatability/matching monitor for relevant tools (CDBup/CDDli/CDDCu)
with resolution monitor. Add resolution monitor for CDBCu/CDAxxtools. Prepilot
perform on CDBcu toolset.
Description and theoretical background:
Resolution estimation approach based on image processing techniques involved
spatial transforms, spectral FFT and basic morphology operations. As mean for image
sharpness (resolution) estimation chosen a STP carbon targets attached to every
Hitachi toolset, require for beam alighnmentprocedure.
1. High density of the targets and relatively high resemblance virtually permits as to perform statistic sampling on non unique targets -thus remove charging impact.
2. Second stage require FFT transform of the images to define spectral distribution.
3. Low pass filter applied for high frequency noise reduction
4. Threshold turn 8 bit image to binary one in order to allow morphological operations
5. Basic morphology operations such as erosion and dilation require for closure of isolated regions.
6. Radius of binary circle (in pixels) will be representation of image resolution and upon conversion result could be estimated in nm, even possible stigmationimpact will be detected.
Resolution reading is relative number and should be treated accordingly -by defining baseline of tool performance.
Taking in account that STP patterns have high similarities, but some of them has carbon structures which could impact recorded resolution, statistical approach should be used for resolution determination. 9 sites within radius of 7 micron chosen to represent given beam alignment.

3. Good resolution-2.49 nm
Bad resolution. -2.69 nm
Resolution monitor

4. 4
Resolution monitor – technical
justification
IDW created as singe die wafer. This letting us possible to
include STP location coordinates into the wafer map
IDP creation is strait forward – 9 sites with step size of 5 um (free to
choose any other reasonable step)
Alignment point chosen around wafer center with coordinates
about 100k/100k for X and Y respectively.
Pattern recognition acceptance limit should be minimal – but
preferable not 0.
Low magnification SEM pattern recognition should be
performed by “Struct” algorithm with sufficiently low
acceptance level in order to alternate sites which were
affected by charging. (around 150-200). High magnification
SEM pattern recognition around 200k should be performed by
“Entire” or ”Hole” algorithms with medium acceptance level
in order to support flexible patter recognition (small changes
in STP should not impact acceptance of image as target).
Measurement execution is not activated, just “Image save”
option.

5. Image Resolution monitor
Completed preliminary testing:
GRS test were run on 199 sites with represent good and bad resolution. Bad resolution
defined is shift from good resolution (best possible beam alignment) for 2 steps of
stigmation for X and Y axes (about 80 units). This miss is common for SSTE with
low/medium training level.
0.021
0.022
0.023
0.024
0.025
0.026
0.027
0.028
0.029
0.030
0.031
Good results STP
1
15 30 45 60 75 90 105 135 165 195
Sample
Avg=0.02531
LCL=0.02224
UCL=0.02838
Individual Measurement of Good results STP
0.024
0.025
0.026
0.027
0.028
0.029
0.030
0.031
0.032
0.033
0.034
Bad results STP
1
15 30 45 60 75 90 105 135 165 195
Sample
Avg=0.02766
LCL=0.02453
UCL=0.03078
Individual Measurement of Bad results STP
Control Chart
Stack Good vs Bad
0.023
0.024
0.025
0.026
0.027
0.028
0.029
0.03
0.031
0.032
Bad results STP Good results STP
Iteration
Bad results STP
Good results STP
Level
198
199
Number
0.027656
0.025313
Mean
0.001196
0.001096
Std Dev
0.00009
0.00008
Std Err Mean
0.02749
0.02516
Low er 95%
0.02782
0.02547
Upper 95%
Means and Std Deviations
Oneway Analysis of Stack Good vs Bad By Iteration
GRS show two separate sets with UCL LCL with separation
over 95% confidence level.

6. ImageResolutionmonitor
Implementation Gain:
1.2.8% in tool availability
2.Implementation of numerical approach –reduce person related estimation of resolution.
3.Early alarm –resolution impact is more obvious than measurement using proposed strategy.
4.Plug and play approach –no dependencies for other tools from same toolset or/and other areas -Gasonic)
5.Slight reduction in TW consumption.
6.Easy control from Eng/tech for tool performance – reduction Eng time 1 WD in month for chart change.
7.PM flexibility –no restriction on number of tools serviced by PM –parallel mode enabled. Project status: Implementation completed 400v and 800v

7. Some images:
CDA03 CDA04 CDA05 CDA11
CDB103 CDB107 CDB101 CDB104
CDD102 CDD105
Tools CDA03 CDA04 CDA05 CDA11 CDB103 CDB107 CDB101 CDB104 CDD102 CDD105
Upper 3.61 3.25 3.25 3.69 3 3.02 2.9 2.3 2.88 2.72
Med 3.46 3.1 3.1 3.54 2.85 2.87 2.75 2.15 2.73 2.57
Lower (not flagging) 3.31 2.95 2.95 3.39 2.7 2.72 2.6 2 2.58 2.42
Image Resolution monitor

8. Some live data:
Case Study: CDA04 before
CDA04 after
ImageResolutionmonitor

9. Some live data: CDD105
ImageResolutionmonitor

10. Some live data: CDD106
ImageResolutionmonitor