Projecting PFC Emissions Using EPA’s PFC Emissions Vintage Model Presented at 9 th Annual ISESH Conference June 9 – 13, 2002 San Diego, CA C. Shepherd Burton*Scott Bartos ConsultantU. S. Environmental 2047 Huckleberry Rd.Protection Agency San Rafael, CA 1200 Pennsylvania Ave. Washington, DC

Outline Introduction Some considerations PEVM: Formulation and structure Results: 2010 Projections … Key Uncertainties, limitations and caveats Summary of findings

Introduction Semiconductor manufacturing: At threshold of low to zero emissions future Path to threshold well documented Recognition of environmental risks of PFC emissions Broad industry cooperation International government-industry cooperation Challenge ahead: Achieving aggressive WSC reduction goal by 2010 What to do? Our aim: Contribute to collective sense of what can be done using emissions modeling

Some Considerations IPCC vs. PEVM: Bottoms-up and top- down methods Are PFC emissions correlated with Si consumption? Si demand: Historical and projected patterns Trends in complexity and effect of copper on interconnects on emissions

IPCC vs. PEVM: Bottoms-up and top- down methods IPCC estimation method –Bottoms-up: Company inventories –Q(MMTCE) = usage x (1 – UF) x (1 – DE) –Tiered: Rigor varies –Arguably impractical for projecting industry emissions Si - consumption method –Top-down: Not now used – = Si x gas- process EF –“Global” EF to be developed –Partner reports and public information –Arguably practical for projecting PFC emissions

Partner Reported ‘99 MMTCE’s vs. MSI Si-Layers (2000 WFW SEMI) (Values not shown to preserve Partner confidentiality)

Correlating Partner PFC Emissions Totals (MMTCE) and MSI Silicon x Nominal Number of Layers PFC Emissions, Year Reported MSI-Si, Year WFW Published r2r

Historical and Projected Si Demand by Linewidth,

Process Complexity Trend

Reducing metal layers by switching from Al to Cu: How many?

Model Formulation, Structure and Operationalization Formulation Structure, sources of information and operationalization Comparison of emissions factors

PEVM: Formulation and Structure ≡ gas-process avg. PFC emissions in year, y ≡ Ē(y) x S (y) –Definition of Ē(y), an emissions factor for uncontrolled emissions, MMTCE per unit area of Si consumed –, from Partner emission reports – S (y),from SEMI WFW database and VLSI Research, Inc. world Si consumption figures Result: Estimate of Ē(y), does NOT account for complexity

PEVM Formulation and Structure: Accounting for “Complexity” = Ē(y) S (y) = Σ i λ i (y) s i (y) ē(y) λ i (y) = nominal no. of layers for technology node i and year y (from ITRS); s i (y) = Si consumed (from VLSI Res., Inc.) for node i ē(y) = avg. gas-process emissions per unit of area of Si consumed per average layer ē(y) = Ē(y)/ ; = (1/ S (y)) Σ i λ i (y) s i (y) Result: Est. of gas-process average uncontrolled emission factor

PEVM Formulation and Structure: Integrating Concept and Mathematics ≡ Ē(y) x S (y) = Σ i λ i (y) s i (y) ē(y) = Σ i i Result: Project uncontrolled gas-process average (BAU) emissions using = Avg( ē(y) ) and projected s i (y) and λ i (y)

Sources & Types of InformationMethod for Estimating PFC-Process Average Emission FactorsProjecting Si Demand to 2010 Projecting Emissions Partners SEM I VLSI Research, Inc. ITRS Industry Reports Annual PFC Emissions Totals, ’95-’99 World Fab Watch databases: ownership, location, capacity, geometry, etc. ’95-’99 World Si Demand, wafer size, in 2, ‘83-’05 World Si Demand, device type, ’83-’05 World Si Demand, linewidth, in 2, ’83-’05 Specs & outlook for industry Test wafers full processed, Cu effect on layers, industry capacity utilization, etc. Partner PFC-Process average emission factors (MMTCE/in 2 Si) 1995, ’96, ’97, ’98 & ‘99 Partner per-average-layer emission factors (MMTCE/in 2 Si*layer) 1995, ’96, ’97, ’98, ‘99 Linewidth-specific emission factors (MMTCE/in 2 Si), Avg. ’95-’99 Adj. for full- process test wafers Avg. number of layers/node U. S. Si demand by linewidth, in 2 Si, World Si demand (in 2 ) by linewidth, Analysis: Adoption/deca y profiles for future technology nodes World CAGR for Si demand, U. S. PFC emissions by linewidth 1988 – 2010 MMTCE World PFC emissions by linewidth MMTCE U. S. share of world layer-weighted capacity, %, Effect of Cu interconnect on number of wiring levels (layers) Compare calc. Partner avg. utilization with available world reports. Schematic of EPA’s PFC Vintage Model (v.2.14) for Projecting US PFC Emissions from Semiconductor Manufacturing

Avg. PFC-Process Emission Factors: Formation of Ē(y), ē(y) and Emission Factors Ē(y), MMTCE/billion in 2 Si ē(y), MMTCE/BSI Si-layer

Some Results: Application of PEVM Projected U. S. emissions: without and with Cu Comparison of PEVM and IPCC-like Method

Projected PEVM PFC U. S. Emissions: With Cu, No Reduction Technology

Projected PEVM PFC U. S. Emissions: Without Cu, No Reduction Technology

Comparing PEVM and IPCC-like Method IPCC-like Method: Extrap. DataQuest ’90 – ’96 survey figures ~18 MMTCEs cf. 15 and 16 MMTCEs PEVM with and without Cu, respectively CAGR: 20%, ’95 – ’10 period, IPCC-like Partnership CAGR: 13%, ’95 – ’99 period

Key Assumptions, Uncertainties and Limitations Emission factors: reflect process-gas average Silicon projections: changes by wafer size and technology generation; metal layers across markets (MPUs, memory, logic, etc.) Disaggregating avg. process-gas, node-specific emissions factors, λ i (y): formation of process/gas-specific emission factors Reduction cases: Selecting technology DEs and penetration rates for alternative control strategies Estimating U. S. share of world emissions

Findings Avg.emissions per unit of Si consumed per average layer appears a robust concept BAU projections: PEVM PFC emissions estimates compare favorably with bottoms-up IPCC-like method BAU 2010 projections with Cu: PEVM gives ~8 times corresponding 1995 emissions PFC reduction technologies: On average >90% DE implemented at 200 mm and 300 mm fabs