10/9/2003 1
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Influence of Gas Composition on Wafer Temperaturein a W CVD Reactor: Experimental Measurements,Model Development, and Parameter Identification
H. -Y. Chang* and R. A. AdomaitisISR and Department of Chemical Engineering
J. N. Kidder, Jr. and G. W. RubloffISR and Materials and Nuclear Engineering
University of Maryland
Contact information:www.isr.umd.edu/[email protected]
*Currently at Novellus Systems
10/9/2003 2
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Overview
ULVAC Tw
True Wafer Temperature
•Research motivation: significant mismatch between single available temperature measurement and true wafer temperature;
• Experimental observations of strong influence of gas composition on wafer temperature;
• Gas phase simulations and assessing the applicability of global eigenfunction expansion methods and other OO-MWR;
• Process recipe developmentusing a validated dynamic process model; parameter ID issues.
10/9/2003 3
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Tungsten CVD
Reactor chamber details
ULVAC cluster tool
10/9/2003 4
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
ULVAC W CVD Operating and Control Structure
•Look-up table: static and not always activated
10/9/2003 5
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Data Acquisition and Experiment Setup
• Signals captured synchronously
• Hardware and software integration
• Challenges: – Signals are of different types and different ranges
– Noise reduction of TC signal
•Offset wafer to assess gas/susceptor thermal conduction effects
10/9/2003 6
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Experimental Results at 0.5 Torr
• Constant lamp powerΤemp change due to gas composition change
• Temp increased as N2increased κH2 ≈ 6κN2 at 0.5 Torr
• TC5 had the largest temp change Both sides of wafer contact gas mixture
– System TC is outside reactor chamber
– Look-up table is inactive in I/O mode
10/9/2003 7
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Gas Flow and Heat Transfer Model
Assumptions
• Fully-developed, laminar gas flow
• Transport and gas thermodynamic properties in bulk gas phase are constant
• No buoyancy or wafer rotation induced flow (Gr = 1.99; Ra = 1.35)
• Heat generated by viscous dissipation and gas and surface reactions are negligible
(Chang & Adomaitis, 1999 Int. J. Heat & Fluid Flow)
Gas velocity, temperature
Gas temp BCs
10/9/2003 8
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Wafer/Susceptor Thermal Dynamics Model
Estimated parameters: Qlp,0, βwf , hwf,0, α0 (steady state/Gauss-Newton)
γ0 , ∆w,∆s (colloc int/G-N, physical arguments)
10/9/2003 9
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Gas Phase Model Solution
Global (eigenfunction) expansion
•vx by Galerkin’s method•Iterative solution procedure
•Projection operations carried out by quadrature•Object-oriented version of MWRtools (www.ench.umd.edu/software/MWRtools)
•Trial function objects for 3D temperature field•SD objects to represent modeling equations/solution; collocation time integrator•PARAM objects to facilitate Gauss-Newton procedure
10/9/2003 10
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Model Solution / Parameter Identification Procedure
Simulation Approach
Main script[rhs, J, drdp] = ulvac_fun(t,s,p)
•unpack state s and parameter p objects•evaluate diff-eq “rhs”, Jacobian, and derivatives w.r.t parameters to be identified; define in terms of SD objects
•Set initial state s and parameter p objects•s = newtraph(s,p)•S = odaepc(s,p)•p = gnstep(s,p)
MWRtools: www.ench/umd.edu/software/MWRtools
Parameter Estimation Sequence1) Using TC5 data estimate Qlp,0 and βwf; steady-state model and data
2) Identify hwf,0 and α0 with TC1-3 measurements; steady-state model/data
3) Identify γ0 , ∆w,∆s with TC1-3 measurements; transient model/data
10/9/2003 11
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Representative Gauss-Newton Iterations
Identify hwf,0 andα0 with TC1-3
measurements:
Results: hwf,0=3.0 W/m2/K α0= -0.03 W/m2/K Qlp,0=30.3 kW/m2 βwf=17.8
Physically reasonable values: Chang et. al. (2001), JVST B, to appear
10/9/2003 12
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Model Prediction for Temperature vs Gas Composition
Inte
rior R
egio
nO
utsi
de R
egio
n•Mean model prediction error is less than 3 K for each data set
•Significant model nonlinearities
10/9/2003 13
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Experimental Results at 0.5 Torr (cont.)
• Temp differences were less than 5 K at corresponding s.s. pointsrepeat gas composition change in reverse order
• TC5 responded to the initial ramping faster than other TCs
• Insignificant temp changes for N2 100 → 60 sccm and H2 40 → 100 sccm
Gas convective heat transfer modeling term relatively unimportant in the low pressure condition of the ULVAC system
10/9/2003 14
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Heat Flux Across Wafer Top/Gas Boundary
∆q = q N2=100 - q N2=60N2 = 100 sccm and 0.5 Torr
• The difference of energy flux ∆q < 7 W/(m2K) is small compared to heat transfer rate q itself; convective heat transfer is negligible compared to gas conduction.
• Relative insensitivity of the wafer temp to gas velocity field justifies omission of combined side inlet and showerhead inlet streams in flow simulation.
10/9/2003 15
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Parameter Estimation and Dynamic Simulation
1) Three different temperature setpoints
2) Collocation-based time integration
γ0 = 12 kW/m2
∆w = 0.8 mm (cf. 0.5 mm wafer)
∆s = 0.65 cm (cf. 2 cm showerhead)
Initial mismatch due to initial u(t) spike
10/9/2003 16
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Dynamic Simulation and Model Validation
Model validation
•Re-check dynamic model validity with final set of parameter values;•Tsh(0) depends on time since previous reactor operating cycle.
10/9/2003 17
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Deposition Process Recipe
Time(sec.)
Ion
curr
ent f
or H
2(A
mp)
Ion
curr
ent f
or H
F(A
mp)
H2
HF
H2 depletion
HF generation
0 400 800 1200 16000.00E+000
2.00E-011
4.00E-011
6.00E-011
8.00E-011
0.00E+000
3.00E-012
6.00E-012
9.00E-012
1.20E-011
Temperature.Step 1 Step 2 Step 3 Step 4 Step 5
H2 flush Cold wafer cycle Hot wafer cycleHeating Cooling
H2 flushing Cold waferbackground
Heating Hot wafer deposition
Quadrupole Mass Spectrometer data by Dr. Yiheng Xu
Process Prep.
Dep.H2 40sccm 40sccm 40sccm 40sccm 200sccmWF6 0sccm 10sccm 0sccm 10sccm 0sccmPressure 0.5Torr 0.5Torr 0.5Torr 0.5Torr 0.5Torr
Cool down
10/9/2003 18
W CVD Simulation Chang, Adomaitis, Kidder, and Rubloff, 2000 Annual AIChE Meeting
Conclusions and Current Work
•Simulator validation: excellent agreement between model predictions and measured wafer temperature;
• Extrapolation of wafer temperature predictions to true process operating gas composition;
•Current simulation research to investigate wafer temperature nonuniformity, gas phase reduced-basis discretization methods; validation of blanket W deposition model;
•Extension of simulation methods to CVD reactor design and other semiconductor manufacturing processes.
Process Prep.
Dep.