The
Yield
Management
Direct-Write Inspection
David Adler
BACUS
September 12, 2003
Company
What Will We Be Inspecting In 2005
-6?
2005-6?
z
Today: compare 4x reticle Vs. design database
y 90 nm node -> 360 nm DR
z
Direct-Write: Inspect 1x wafer image
y 45 nm node -> 8x smaller DR than today!
2 ·
z
Nano-Imprinting: Inspect 1x template
z
X-ray, LEEPL: Inspect 1x mask
z
NGL’s: Tool qualification (inspect image) at 1x
z
Cannot afford to develop separate inspection tools!
1X Inspection Requirements
z
Tough Requirements for Inspection
y Catch 100% of defects
y Very low false-positive rate
(fewer than 100 per terapixel!)
z
Tough Imaging Capabilities
y 25 nm resolution or better
y Image uniformity
y “Simple” images—can be modeled easily
z
3 ·
High Throughput
The 50k
-Foot View of
50k-Foot
Inspection
If a single chip was as big
as North America
(3000 mi):
Chip
N. America
45 nm line
Freeway
Defect today Car
4 ·
Defect 2006
Pothole
Atom
Baseball
The 50k
-Foot View of Inspection
50k-Foot
Therefore, the problem of inspecting
for defects at 1X
Is Similar to ...
Finding and classifying every pothole
in N. America in ~1 hour
5 ·
1x Inspection Basic Requirements
z
High Throughput
y Inspect 1 field (1E12 pixels) in about an hour
y => process 200 million pixels per second
y Image template or wafer
y Simulate image from database
y Compare real image to database image
y Detect and classify defects
y Good news: no MEEF
y fewer pixels on template or wafer
than on 4x mask!
6 ·
R&D Required for 1x Inspection
z
Develop high-speed e-beam microscope
y At least 200 million pixels / second
y Stable images—no charging, distortions
z
Develop e-beam rendering algorithms
y Very high accuracy required
y Take into account physics of e-beam imaging
y Include processing effects on wafer/template
z
Develop detection algorithms for 1x
y Different for various lithography technologies
7 ·
Key Contributors to KT PERIS Program
z
DARPA—Dave Patterson
z
KLA-Tencor (partial list):
y Marian Mankos, Luca Grella Matthew Marcus,
Harald Hess, David Soltz, Stuart Friedman,
Eugene Shifrin, Don Parker, David Walker,
Salam Harb, Gershon Perelman, Tom Tyrie
z
Special Thanks
y
y
y
y
8 ·
Gertrude Rempfer (Portland State Univ.)
David Joy (U. Tenn.)
Robert Reams (ARL)
Lee Veneklasen
What is PERIS ?
z
Parallel Electron Reticle Inspection System
z
E-Beam 1x Inspection at Optical speeds
y 100-1000X faster than a traditional SEM
y Imaging system--like “Light” microscope
z
For 100nm and smaller design rules
z
Novel Technology Developed at KLA-Tencor
z
“Proof of Principle” has been demonstrated
PERIS: CCD-Based E-Beam System
Similar architecture to
optical microcscope
except uses electrons
instead of light.
Image Converter and CCD Detector
2D electron image collects
> 10,000 pixels in parallel
Imaging Optics
SEM
Flood Gun
Illumination Optics
Continuous X Motion
PERIS is Better than SEM
z
Gets around electron interactions
z
Capture >104 pixels in parallel
y Power density is <10,000 times lower with PERIS
y lowers possibility of damage to resist
y 104 x greater integration time reduces noise
y Low speed electronics (relative to SEM)
z
No scanning deflectors -> reduced electronic noise
z
Very high data rates possible (200 MPPS and higher)
z
Extensible to higher resolutions and data rates
z
Don’t need a “bright” source (light bulb, not laser)
z
Lower current density means less space charge effect.
z
No “edge” or “scan” effects that appear in SEM images
y Makes D:DB Inspection simpler, faster.
SEM Images Can Be Complex
• Edges are enhanced
• Enhancement depends on
other nearby structures
• These effects must be
accurately modeled in
D:DB rendering program
12 ·
PERIS Image Is Simple
High-resolution SEM
image of particle
PERIS image 97um FOV
13 ·
Cell-to-Cell Detection of Contact Defects
Cell-to-Cell
Difference
Original Image
Partially-closed contact: S/N = 14.5
Fully-closed contacts: S/N = 75
Expanded
view of
partiallymissing
contact
Electron Image vs. Database Rendering
Raw Image from
Electron Microscope
Database Image
Its easy to see that the contrast is different and the corners are different.
These differences must be removed before defects can be located.
15 ·
Images are Processed and Subtracted
Residual
background
noise = 2.7
grey levels
Difference image
without new
algorithms
16 ·
Difference image
using new
algorithms
Conclusions
z
E-Beam D:DB inspection is required for many NGL’s:
y
y
y
y
z
E-beam direct write lithography (DBDW)
Nano-imprint lithography (NIL)
1X x-ray, LEEPL
Tool qualification
1X Inspection will be much harder than inspecting masks
y Higher resolution < 25 nm
y Higher throughput > 200 MPPS
y Processing must be included in rendering models
z
KT is investigating a new technology for 1x: PERIS
y Sensitivity of a SEM
y Speed of an optical system
y Die-to-Database inspection