Applying LCA into Decision Making

1
Applying LCA into Decision Making
Nina Chen (yuechen_at_mit.edu) Greg
McRae (mcrae_at_mit.edu) Department of Chemical
Engineering MIT
2
Outline

• Motivation and scope of study
• Process-product Input Output (PIO)-LCA mechanism
• Case studies
• Value of information
• Conclusion

3
Key Environmental Challenges and Needs

More quantitative environmental metrics
Systematical evaluation approach to environmental
impacts Database that includes important
chemical properties Consideration of
uncertainty in the environmental evaluation
Rapid environmental evaluation Life Cycle
Assessment (LCA) beyond the factory Dynamic
simulation of the processes Advanced process
control, sensing and metrology
Multi-objective optimization and decision support
4
Scope

• Develop methodologies and metrics for rapid
  economic and environmental evaluation
• Integrate the treatment of uncertainties into
  decision making about alternative technologies
• Identify opportunities for creating win-win
  situations

Strategy

• Focus on understanding uncertainty and processes
• Use existing PIO-LCA method at different stages
• Explore value of information

5
Frequently Encountered Issues in Life Cycle
Analysis

• Large amount of data are required
• Large uncertainties are imbedded in
• Environmental information
• 1 order of magnitude in air pollutant
  emission factors
• 2 3 orders of magnitude in cancer
  toxicity indicators
• 3 6 orders of magnitude in
  non-cancer toxicity indicators
• Process information
• New technologies
• Unknown equipment
• Upstream information incomplete
• Time and resources do not allow indefinite
  refining of data and model

What Shall We Do?
6
Life Cycle Analysis Model of This Work
Upstream Downstream Emissions, Material and
Energy Usage
Weighting Factors
Flow Rates
Human Toxicity Global Warming Effect Ozone
Depletion Effect Respiratory Effect
Products Byproducts Chemical Energy Water Waste
Input Output LCA Model
Design Decisions
Emissions
Process Model
Impact Indicator
Environmental Performance
Yield Process Time
Compliance with Regulations
Human Exposure
Environmental Concentration
Fate, Transport, and Exposure Model
Environmental Properties Chemical
Properties Exposure Properties
Alternative Designs
7
Components of an Environmental Valuation Model
Characterization
Weighting
Activity Emission Factors
Factors
Factors
Eij
Hik
Greenhouse
Greenhouse
Methane Reforming
effect
effect
wk
CO2
Acid
Acid
CO
Deposition
Deposition
2
NO
x
Environmental
SO
Carcinogen
Carcinogen
Impact
2
Indicator W
Exposure
Exposure
N
O
2
Energy
CO
Cu CVD
Generation
Process
VOC
Photochemical
Photochemical
smog
smog
CH
4
PM
...
Ozone
Ozone
HCHO
depletion
depletion
...
...
å
å
Precursor Generation

W
Eij
H
w
ik
k
k
i
...
Cano-Ruiz 2000
8
Model Input One Usage Matrix (B)
Electricity
Usage Matrix
B
Electricity
Cu CVD
Cu Film
Cu1(hfac)(tmvs)
H2
9
Model Input Two Fabrication Matrix (C)
Fabrication Matrix
Electricity
Cu CVD
Cu Film
Cu1(hfac)(tmvs)
H2
10
Model Input Three Market Share Matrix (F)
Market Share Matrix
Electricity
Cu CVD
Cu Film
Cu1(hfac)(tmvs)
H2
11
Model Input Four Emission Matrix (E)
Emission Matrix
Electricity
Cu CVD
Cu Film
Cu1(hfac)(tmvs)
H2
12
Model Input Five Characterization Matrix (H)

• Characterization matrix (H)
• Large uncertainties imbedded in the values

GWP100 Respiratory Human Toxicity
Effect Potential (non-
cancer)

kg CO2 equivalent/kg
kg PM10 equivalent/kg
DALYs/kg
Unit
1 -23.3
0.15 4.21E-9 -8.3 1
CO2 kg
SO2 kg
PM10 kg
Based on willingness to pay

Valuation Factor
w
3e-2 40 85000
13
Mathematical Model

• Model Input Six Price vector (p)
• Allocation matrix (G) for multiple product
  processes
• Throughput matrix (D)
• Dji FjiGji
• Direct product requirement (qdirect)
• qdirect (I BD)d
• Total product requirements
• q (I Aprod AprodAprod AprodAprodAprod
  )d (I Aprod)-1d
• where Aprod ? BD

Gji the amount of throughput of process j that
is attributed to one unit of product i made in
process j
Dji the amount of throughput of process j that
is attributed to the demand of one unit of
product I at current price and market share
14
Mathematical Model

• Total process throughput requirements (x)
• x Dq
• Life cycle environmental exchanges inventory (e)
• e Ex
• Impact valuation by process (?process)
• ?process Diag(x) ET H w
• Impact valuation by emission (?emission)
• ?emission Diag(e) H w

15
A Smaller Case

• Eighteen processes
• Fourteen products
• One hundred and two emissions
• Seven environmental impacts

Gas-fired Plant
Nature Gas Production
Coal
Gas
Coal-fired Plant
Coal Production
Hydroelectric Plant
Electricity
Cu CVD
Cu Film
Cu1(hfac)(tmvs)
16
Case StudyCu CVD
Pressure Sensor and Controller
The process model is provided by University of
Maryland.
Film Thickness Sensor and Controller
Precursor Cu1(hfac)(tmvs)
Wafer
Heater
Temperature Sensor and Controller
Scrubber
Carrier Gas Hydrogen
Sensor Path
Control Path
17
Analysis Results of the Environmental Model

• When uncertainties are considered, power
  generation still contribute to a significant part
  of environmental impact.
• Large uncertainty in coal-fired power plant and
  oil-fired power plant is from the uncertainty in
  PM10 effect and CO2 effect in GWP

18
Second Case Study Chamber Cleaning with NF3 or
F2?
RF Power
SiO2 Deposited on Wall
NF3/F2, Ar, N2
F, NF, NF2, Ar
Plasma Generator
N2, F-, NF
SiF4
F?, F2, N2, SiF4, O2

• Merits of NF3
• High disassociation rate
• High removal rate
• High etch rate
• Drawback of NF3
• High cost

O2
SiF4
F?, F2, O2, N2, SiF4
CVD Reaction Chamber

• Merits of F2
• Low cost
• Drawbacks of F2
• High toxicity
• High reactivity
• On-site generation creates explosive H2

Comparison criterion considered Life cycle
impacts given the same cleaning performances
19
Process Modeling with Kinetics

• Lumped kinetics and Perfectly Stirred Tank
  Reactor (PSTR) model
• Key assumptions
• Free electrons are generated mainly by ionization
  Are --gt Ar2e
• Electron loss and production are linear to
  electron concentration
• Diffusion of electrons dominates the transport of
  electrons.

NF3 e ? NF2 F? e k32.06E-17
Te1.7exp(-37274/Te) NF2 e ? NF F ? e
k21.57E-17 Te1.8exp(-27565/Te) NF e ? N
F ? e k11.57E-17Te1.8exp(-27565/Te)
F2 e ? F- F? k 1.02E-5Te-0.9exp(1081.8/Te
) 4F? SiO2 ? SiF4 O2
20
Process Modeling with Stoichiometrics
Driving forces of LCA impacts Cleaning gas
usages Energy consumptions
Cleaning Gases Energy

• where for NF3 cleaning
• for F2 cleaning

• Fluorine Utilization Yield F uniform(10-5,
  0.6)
• Energy Utilization Yield ?E uniform(10-10,
  0.6)
• Cleaning Time t (s) uniform(6E-4, 1200)

21
Comparison of Relative Impacts of GWP of Two
Models
Process Model with Kinetics
1.9
3.3

Process Model with Stochiometrics
Relative GWP of NF3 Process to F2 Process

• 23 orders of magnitude of uncertainties in
  inputs does not necessarily leads to low
  confidence in decision
• Increase of modeling detail decreases the
  uncertainty of the outputs
• But the decision is still the same F2 is
  better!
• Required confidence level should determine depth
  of analysis

22
Process Modeling Hierarchy and Resource Needs
Process Model Hierarchy
Distributions of Yield
Resources Needed
1 Simple stoichiometric yield 1
2 Lumped kinetics (3 reactions) 10
3 Detailed kinetics (60 reactions) 100
4 Model based experiments 1000
23
Right Procedure of Analysis
Problem
Key Parameters
Analysis Models
Relative GWP
F of NF3 Cleaning -0.64
F of F2 Cleaning 0.46
Cleaning Time t (s) -0.28
E of NF3 Cleaning -0.20
E of F2 Cleaning 0.12
4F SiO2 ? SiF4 O2
1. Stoichiometric
LCA
Gas Usage (mol)
Decision go to next level?
Refine cleaning process model
PowerPlasma Generator 0.69
Power to the TElectron in NF3 Disassociation -0.37
NF3 in NF3 Production -0.33
Energy Used in F2 Production 0.21
Power to the TElectron in NF2 Disassociation -0.19
NF3 e ? NF2 F? e NF2 e ? NF F ? e NF
e ? N F ? e F2 e ? F- F?
2. Simple Kinetics
LCA
Further refine cleaning process model
Decision go to next level?
3. Detailed Kinetics
163 Gas Phase Reactions in Plasma Generator
24
Process Modeling vs. System Boundary
System Boundary
Life Cycle
85
gt99
Chemical Industry
gt99
Semiconductor Industry
12 Month Effort?
6 Month Effort
2 Month Effort
Downstream Treatment
gt95
N/A
gt99
Confidence Level in Distinguishing NF3 and F2
Cleaning
1 hr Effort
N/A
gt95
Cleaning Tools
gt95
Process Modeling Level
Stoichiometry
Simple Kinetics
Detailed Kinetics

• Depth of process modeling and width of system
  boundary are complementary to each other.
• Based on existing knowledge, choose appropriate
  direction

25
Framework of Decision-Making Process
Generate new alternatives
Refine model, collect more data, increase data
accuracy
Ranking and Sensitivity Analysis
No
Alternative Technologies NF3 vs. F2 Cu CVD vs.
Cu plating
Is info enough for decision?
Environ. Impacts Model
Economic Impacts Model
Process Model
Yes
Uncertainty Analysis
Do nothing, or change to alternative
26
Future Plan Value of Information (VOI)

• A simple example is it worthy to buy 1M
  equipment for testing?

More Research
Current State 50 sure COO NF3 cleaning 3
COOF2 cleaning
90 sure (p)
Continue NF3 Cleaning
Cost of NF3 cleaning
(90)
Adopt F2 Cleaning
COOF2 cleaning if works well
More Research
(10)
COOF2 cleaning if not work well
More Research
10 sure (1-p)
Continue NF3 Cleaning
Cost of NF3 cleaning
(10)
Adopt F2 Cleaning
COOF2 cleaning if works well
(90)
COOF2 cleaning if not work well
Continue NF3 Cleaning
Cost of NF3 cleaning
(50)
COOF2 cleaning if works well
Adopt F2 Cleaning
(50)
COOF2 cleaning if not work well
27
Conclusions and Key Points

• Large uncertainty in the inputs does not
  necessarily lead to low confidence in decisions.
• PIO-LCA combines both the merits of EIO and
  engineering design method
• Hierarchical modeling in combination with
  uncertainty analysis are efficient ways to
  support the decision making and resource
  allocation process.
• VOI may give direction on resources allocation.

28
Acknowledgements

• Laura Losey
• David Bouldin, Mike Kasner, Tim Yeakley, and
  Tina Gilliland Texas Instruments
• Larry Novak Novak Consulting, LLC
• Alejandro Cano-Ruiz and Pauline Ho Reaction
  Design
• Daren Dance WWK
• Joe Van Gompel BOC Edwards
• Holly Ho TSMC, Taiwan
• McRae Group MIT
• Gleason Group MIT
• Engineering Research Center for Environmentally
  Benign Semiconductor Manufacturing NSF/SRC.