Advanced Manufacturing

using Virtual Metrology

and Equipment
Intelligence ®

New techniques to solve difficult metrology

and process optimization challenges

David Fried, Ph.D.

Vice President, Computational Products, Lam Research
Cautionary statement regarding
forward-looking statements
This presentation and the accompanying discussion contain “forward-looking statements” under U.S. securities laws. Forward-looking
statements include any statements that are not statements of historical fact. Examples of forward-looking statements include, but are
not limited to: (1) anticipated business, balance sheet, cash flow and financial measures and results, including guidance, whether on a
GAAP or non-GAAP basis; (2) economic, market, industry and industry segment expectations; (3) product performance and changes
in market share or customer demands; (4) our ability to successfully execute business, capital allocation, product and growth plans or
strategies, or otherwise deliver value for customers and stockholders; (5) the impact of the COVID-19 pandemic on our operations and
financial results, and our ability to mitigate operational and business impacts caused by it; and (6) the impact of trade regulations,
export controls and trade disputes. Forward-looking statements speak only as of the date they are made and are subject to risks and
uncertainties that could cause actual results to differ materially from those expressed, including: the severity, magnitude and duration
of the COVID–19 pandemic (and the related governmental, public health, business and community responses to it), and their impacts
on our business, results of operations and financial condition, are evolving and are highly uncertain and unpredictable; business,
political and/or regulatory conditions in the consumer electronics industry, the semiconductor industry and the overall economy may
deteriorate or change; the actions of our customers and competitors may be inconsistent with our expectations; trade regulations,
export controls, and trade disputes may inhibit our ability to sell our products; and widespread outbreaks of illness may impact our
operations and revenue in affected areas; as well as the other factors discussed in our filings with the Securities and Exchange
Commission (“SEC”), including specifically the Risk Factors described in our annual report on Form 10–K for the fiscal year ended
June 28, 2020 and our quarterly report on Form 10-Q for quarter ended September 27, 2020. You should not place undue reliance on
forward-looking statements. Lam undertakes no obligation to update any forward-looking statements.
Agenda 01 Next Node & Metrology Challenges

02 Process Modeling & Virtual Metrology

03 Mass Metrology

04 Equipment Intelligence®

05 Example - In-Situ Metrology

06 Conclusion
Next Node Technologies
and Metrology Challenges
Challenges to Process Integration and Metrology

LAM RESEARCH
Critical Technology Inflections
EUV
Patterning Patterning Variability, Cost
Additive Patterning

3D NAND Stacked 3D NAND Density

New Materials & Architectures

DRAM Density, Power
Non Volatile Fast Memory
Memory
Storage Class 3D Vertical Architecture
Density, Cost
Memory New Materials

Magnetic, Phase Change & Speed, Power,

Embedded Memory
Ferroelectric Memory Density

Gate All Around

Transistor Power, Speed
Negative Capacitance FET
Logic
New Materials
RC Management Power, Speed
Barrierless

Packaging Chip Integration Heterogeneous Integration Footprint, Speed

Metrology Challenges
Next Node Technologies

Smaller 3D Device Measurement Hybrid

Dimensions Geometry Frequency Metrology
Requires metrology with Difficult to measure inside HAR Lack of low-cost, high- Need to combine fragmented
greater sensitivity and structures, opaque thick films, quality measurements metrology data from different
precision other “hidden” structures can lead to process sources
excursions
Process Modeling & Virtual
Metrology
Using Unit Process Models, Virtual Fabrication, Virtual Metrology
and Domain Expertise to Optimize Processes and Pathfinding

LAM RESEARCH
Virtual Fabrication Platform
Virtualize the entire fabrication process flow (or a subset, patterning sequence, etc.)

Process Flow  Step-by-Step Process, 3D Model  SEMulator3D

Applications
Behavioral description, process library Virtual Fab Modeling Engine
…
PATHFINDING
Hard mask dep.
Lithography • Process integration validation
Mask open etch • Big Branch decisions
Mem. hole etch
…
• Process assumptions

DEFECT ANALYSIS
Front End Logic BEOL • 3D Failure Checks
Design  Editor and Viewer to support
GDS layout and other input types • Defect Kill Ratio
Design • R/C, Leakage, Device, IV, CV,
GDS etc.
OPC
YIELD: RAMP AND HVM
Contours
… • Design Process Optimization
Power MEMS
• Process Window Optimization
• Cross-Wafer Uniformity
Virtual Metrology
Measure Critical Structural Attributes during
Simulation
Similar to in-line metrology in the fab

Can measure items such as film thickness, critical dimension

(width), step height, sidewall angle, contact area and more

Advantages
• Measure parameters that could only be determined destructively in
the fab
• Applicable to any device type - no special metrology structures or
equipment required
• Useful in validating process assumptions and effect of process
changes on design rules.
• Validate process model by comparing wafer-based metrology data
to virtual metrology
Process Model Calibration: Virtual & Physical Metrology
SPACER 1 OXIDE FIN CD STEP – XSEM COMPARISON OF XSEM WITH VIRTUAL CALIBRATION RESULTS
IMAGE FROM IMEC METROLOGY

GATHER TARGET MEASUREMENTS PROCESS MODEL CALIBRATION AFTER COMPARISON OF ACTUAL VS. CALIBRATED
MULTIPLE AUTOMATED TRIALS MODEL RESULTS

Enter target measurements Actual XSEM values and Comparison of actual device
from an actual semiconductor predicted target and SEMulator3D model built
• Critical dimensions from XSEM values converge after process from optimized parameters
or TEM model calibration optimization
study
Mass Metrology
Angstrom-Level Measurements for Advanced Process Monitoring
and Control of 3D Device Structures

LAM RESEARCH
Metryx® Mass Metrology
Monitoring HAR/Hidden Processes with Å-Level precision

Liner
Dep
Etch
Descum

Accurate & Application

High Precision 3D/ Hidden
Non-Destructive Flexibility
Mass metrology enables Delta-mass is non- Measure etch, dep, bevel, Enables quantification of
Å-level precision process destructive & accurately clean, curing & all wafer- HAR, hidden/buried
monitoring with direct represents wafer state/ processes, from processes where
measurement at high TPUT process (incl. oxidation) R&D through to HVM conventional metrology fails
Metryx® Metior®

Challenges in DRAM and 3D-NAND HVM Detect Process Excursions not possible with Optical Metrology

Monitoring and control of ultra-thin films, thick

films or stacks, and complex 3D geometries (HAR • Sensitive & Non-Destructive
structures) is difficult with traditional optical
metrology. • Direct Measurement
• Correlation to Electrical performance
• Independent of Device Patterning
DRAM 3D-NAND
• Low Cost of Ownership
CD

Direct correlation to Capacitance High-k Predictive of Leakage

1.4
1.3

Leakage Current (au))

1.2
1.1

Capacitance (au)
1
0.9
Memory Hole Sidewall
& Conformal Process & 0.8
Capacitor Deposition Fill 0.7
Etch 0.6
0.5
0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3
ΔMass Etch Composite ΔMass
3D-NAND SONO Punch: Metryx with SEM3D process model
192 class
Pre-SONO Punch SONO Punch (Final) Validating Model to Measured ΔMass

Tier 2

Si Gouging

Deposition

Visualising Process ΔMass with SEM3D Etch cycles

Transition from Si liner etch to Oxide liner

Tier-2 CH
Tier 1  Delta-Mass enables monitoring of SONO Punch process
(optical alternatives are limited due to low signal response)
 ΔMass response to Tier-2 Channel Hole CD (left), aligns
well with process model expectation
Punch here
 SEM3D model and measured ΔMass results (above) show
good correlation & trend rate with process cycles
Mass Measurement of Gate-All-Around Fin Etch/Spacer
Mass Change of SiGe Recess/Spacer Process
-50
Lot
Target Distribution
-50.5 Nominal w/ Outlier
Cumulative ∆Mass (mg)

-51 Outlier

-51.5

-52

-52.5

-53 Over Etch

-53.5 Over-Etch
outlier
-54
Fin Recess Inner Spacer
Etch Spacer EB

 Mass Metrology enables quick & accurate targeted

Courtesy of imec

development for process monitoring in R&D & HVM

 Excellent ability to monitor isotropic-etch and ALD

conformality in trench/recess
Equipment Intelligence ®
Enables advanced adaptive modeling using tool and sensor data

LAM RESEARCH
Lam Equipment Intelligence® is focused on addressing
customer needs

Experts connected Data optimization

6 1

Fleet optimization & Equipment Digital worker

5 2
predictive maintenance Intelligence® optimization

4 3

Maintenance optimization/ Tool optimization

human variability elimination through simulation

Smart tools for early learning to HVM productivity

Lam Equipment Intelligence® is focused on addressing
customer needs
Bringing value out
Leverage the best of data
experts available Experts connected Data optimization

6 1

Fleet optimization & Equipment Digital worker

5 2
predictive maintenance Intelligence® optimization

4 3

Maintenance optimization/ Tool optimization

human variability elimination through simulation Improving productivity
Transitioning from
and quality
reactive to predictive

Smart tools for early learning to HVM productivity

Lam Equipment Intelligence® - Enables Tool-to-Tool
(Fleet) Matching & Process Performance Improvements
FLEET MAINTENANCE FLEET OPTIMIZATION EI-DA + ON-WAFER PREDICTIVE CAPABILITY
INITIAL FLEET OPTIMIZATION (EI-DA) RESULTS FEEDBACK FOR APC TUNING

Center the process window

Broad distribution within the Reduce chamber variability Reduce fleet variability using Improve process Cpk using
using on-wafer results to tune
process window due to human variability machine learning predictive capability for APC
the fleet

Improved fleetwide process control

 Illustration of how maintenance and fleet optimization
combined enables higher level of control
MAINTENANCE OPTIMIZATION (ELIMINATION OF HUMAN ERROR) + FLEET OPTIMIZATION BASED ON TOOL SENSOR DATA

Improved MTBC (and availability) by eliminating Reduced variability in sensor signal output increases Reduced yield risk due to higher
unscheduled cleans due to poor clean quality signal to noise for tool health monitoring models sensitivity models with low false alarms
TIME (RFH) BETWEEN CLEANS (MTBC) EI-DA CHAMBER HEALTH MODEL EI-DA WET CLEAN QUALITY VALIDATION MODEL

Model Variable 1
MTBC (RF Hours)

Time / Wafers Wet Clean procedure issue (no on-wafer shift detected):
30 - 80% reduction in
unscheduled cleans
Non-WCO
WCO Incomplete Top heater screw was found to be loose (10/8)
Non-WCO WCO WCO Complete
WCO Wet Cleans After tightening, the trend returned to normal

Reducing chamber to chamber variation enables useful detection & prediction models
= higher equipment availability, higher yield
Equipment Intelligence® Data Analyzer prediction:
monitor chamber shift & drift by Y-data regression
Advanced adaptive modeling system using tool and sensor data to monitor and control on-wafer performance
Metrology Low frequency CHAMBER 1 CHAMBER 2 CHAMBER 3 CHAMBER 4
Measurement

Tilt Angle

Tilt Angle

Tilt Angle

Tilt Angle
Time Time Time Time

Every Wafer CHAMBER 5 CHAMBER 6 CHAMBER 7 CHAMBER 8

Measurement
• Tool data

Tilt Angle

Tilt Angle

Tilt Angle

Tilt Angle
EI-DA
• Sensor data

Regression is based on: Time Time Time Time

Tool and sensor parameter data to
Measurement data (CD, Angle, Thickness …) CHAMBER 9 CHAMBER 10 CHAMBER 11 CHAMBER 12

Ability to monitor every lot and flag out-of-control wafers

Tilt Angle

Tilt Angle

Tilt Angle

Tilt Angle
• Reduce scrap and excursion events
• Flag OOC wafers for additional disposition
Time Time Time Time
Using sensors for Equipment Intelligence®

Lam’s latest generation etch product, Sense.i®, is powered by Equipment

Intelligence® which includes almost 400 more sensors than previous-
generation tools. It autonomously self-calibrates, has self-maintenance
capabilities and uses machine learning to adapt to process variations.
In-Situ Metrology: Etch
Depth Control Solutions
Broadband Spectral Reflectometry to Achieve Repeatable and
Accurate EndPoint Control without Stop Layers

LAM RESEARCH
Traditional in situ Metrology Inadequate for Endpoint at NM Scale
Spectra of Si1 vs Frame

• Pre or post measurement

• Lot-to-lot (L2L) Control
Standalone
• High Accuracy
metrology
• Flexible sampling

Signal for 6 nm
• Pre or post measurement
Integrated • Wafer-to-wafer (W2W) or within
metrology wafer advanced process control
Intensity at 400nm vs OCD R2=0.16484
• Process tool integration

• Realtime continuous measurement

• In situ active control
In situ
• Multi-step
metrology • Poor sensitivity & accuracy with
conventional endpoint methods 6nm

Reference Depth
In Situ Etch Depth Metrology with Machine Learning
Feature monitoring & endpoint detection
Challenges Differentiated Solution: Lam Spectroscopic Reflectometry (LSRa)
LSRa: Stable in situ reflectometry with advanced signal analysis
• Repeatable etch end point
Signal processed with Repeatable endpoint control
for etch-to-depth In situ signal collection Advanced reflectometry
machine learning algorithm for target depth
application without stop
layers
• Etch compensation for Etch Time Machine Learning
End Point

Reflectance
Target
wafer to variability due to Test error

Error
Depth

Depth
incoming variations Wfr 1
Training error Wfr 2
Model complexity
Wfr 3
Wavelength

Repeatable metrology enables precision etch depth matching

Metrology quality data

designed for HVM Wafer
Wafer to
to Wafer
Wafer Depth
Depth Variation
Variation

• High SNR Sensor

Example: endpoint detection
of buried gate etch • Long term stability

• Reliable ML Algorithm

Without In-situ Control With In-situ Control

Machine Learning of in situ Spectra
Predicted Depth

Actual Performance
(Tested for > 6 months in HVM)

Reference > 50% reduction

Depth > 45% reduction
Predicted Depth

System A System B
ER, incoming,
Timed Etch ML EP
chamber
contributions
well separated In situ spectral endpoint provides
Etch Time significant improvements to W2W
control compared to timed endpoint
Ramp to Yield Faster

Virtual Metrology and

Fleet Matching: Chamber 1 3 5

2 4 6

Machine learning to correlate

system behavior to end of
line performance to enable
the rapid identification of
Day 0 Day 65
chamber mismatch and
process variability
Summary
• Machine learning enables in situ metrology In situ Control System
where
EP
• No clear endpoint signal (no etch stop Control
layer) In-Situ
Depth Evolution Optical Spectra
• Low signal to noise ratio; tight control
spec
• Complex incoming variations & patterning

• Lam has successfully demonstrated In situ

End Point
endpoint control on etch systems in HVM

• High-quality sensor data, process expertise,

and intelligent control opens up new APC
Pre Post
possibilities
Conclusion

LAM RESEARCH
Conclusions
Smaller dimensions, 3D structures, and measurement & integration requirements
will create metrology challenges for next node architectures.

Data from in-situ and standalone metrology , using ML/AI, calibrated models and
advanced analytics, will drive real-time feed-forward and feedback optimization.

Process domain expertise is required to meet these challenges. The Equipment

Intelligence® product uses Lam's domain expertise, unit process excellence and
advanced analytics to add value to data.

Result: Better process and integration decision making and results.

Call to Action: Fabs, metrology vendors and tool vendors need to share data and
expertise to solve these issues, with secure data sharing solutions needed.
Q&A