Transcript Document 7357349

CAN PLASMA MODELING BE A PREDICTIVE
TOOL IN PROCESS DEVELOPMENT?
ETCHING OF VERY HIGH ASPECT RATIO
FEATURES AND GATE STACKS
Mingmei Wang, Juline Shoeb,
Yang Yang and Mark J. Kushner
University of Michigan
Ann Arbor, MI 48109 USA
[email protected] http://uigelz.eecs.umich.edu
October 2008
AVS08_MJK
CONGRATULATIONS
 Congratulations to Prof. Herb Sawin on his retirement.
 With apologies to www.despair.com
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
AGENDA
 Past, present and future expectations for process modeling.
 How do we get from here to there?
 We model equipment well, why not processes?
 Extremely high aspect ratio features
 HfO2 gate-stack etching
 Concluding Remarks
 Work supported by Semiconductor Research Corp., Applied
Materials Inc., Tokyo Electron Ltd. and Intevac.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
1991 NRC REPORT: “PLASMA PROCESSING OF MATERIALS”
 In 1991, the US National Research
Council assessed the status of
plasma processing of materials.
 Findings:
 “Currently, computer-based
modeling and plasma simulation are
inadequate for developing plasma
reactors.”
 “There is no fundamental obstacle
to improved modeling…nor to the
eventual creation of computer aided
design tools for designing plasma
reactors.”
 Ref: “Plasma Processing of Materials”, NRC, 1991.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
1991 NRC REPORT: WHAT IS NECESSARY TO ACHIEVE GOALS
 To achieve the modeling goals set out in the 1991 NRC Report:
 A reliable and extensive plasma data base against which the
accuracy of simulations of plasmas can be compared.
 A reliable and extensive input data base for calculating plasma
generation, transport and surface interaction.
 Efficient algorithms and supercomputers for simulating magnetized
plasmas in 3 dimensions.
 Ref: “Plasma Processing of Materials”, NRC, 1991.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
2007 NRC REPORT: PLASMA SCIENCE: ADVANCING
KNOWLEDGE IN THE NATIONAL INTEREST
 NRC Decadal Study on Plasma Science
(http://www.nap.edu/catalog.php?record_id=11960)
 Low Temperature Plasmas:
Extreme challenges face modeling and the allied
sciences to develop comprehensive and validated
theories, computer models and databases that place
predictive capabilities in the hands of technologists.
“This represents the highest level of challenge and the highest potential
return…to both quantify and advance our understanding of low
temperature plasmas, and to leverage that understanding by speeding
the develop of society benefiting technologies.”
Ref: Adapted from “Plasma Science: Advancing Knowledge
in the National Interest”, NRC, 2007.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
DOE/OFFICE OF SCIENCE
WORKSHOP
 Low Temperature Plasma Science: Not
only the Fourth State of Matter but All of
Them (September 2008)
 Priorities in Modeling and simulation
 1 - Expand plasma capabilities to combine
theory, simulation, and reacting flow
equations to model closely-coupled,
stochastic processes.
 4 - Develop multi-scale methods describing interactions of
plasmas with nanoscale features such as nano-particles and
nano-textured surfaces.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
1991 to 2008…..Have we accomplished anything?
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
EQUIPMENT AND PROCESS MODELING
 The ability to craft structures and functionality on surfaces
 
ultimately depends on the ability to control f v , r , t  of charged and
 
neutral species (and specify f v , r , t  to produce a feature).
 Can we achieve this degree of sophistication?
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
MODELS FOR DESIGNING PLASMA EQUIPMENT
 We do well in modeling plasma equipment
 Vary power, pressure, geometry, gas mixture in the model…We can
predict the change in fluxes.
 Parametric variations better
represented than absolute values.
 Subtleties of equipment (e.g.,
materials, gaps) can be captured.
 Instabilities and long term
transients are problematic.
 High frequency and non-standard
excitation are challenging but
doable.
 In principle there is a path forward
for most of these issues.
AVS08_MJK
 3D CCP simulation showing
plasma penetration through
confinement rings (Ref: Kenney,
Rauf, Collins, 2008)
University of Michigan
Institute for Plasma Science
and Engineering
MODELS FOR DESIGNING PROCESSES
 We do less well in modeling processes.
 Vary power, pressure, geometry, gas mixture in the model…
Quantitative, first principles predictions of profiles (or better yet,
specify the reactor conditions to produce a profile) are lacking.
 In principle, this is a “chemistry issue”. If we knew the reaction
mechanism, we could represent the process well.
 Subtleties abound (e.g., mixed layers, reflection from non-planar
local topology) that are not known or that are not captured.
 Subtleties may not be resolvable from first principles (e.g., MD)
on large enough scale lengths for process design without
massively parallel computing approaching weapons design.
 Even then fundamentals parameters are not known…..
 Is there a path forward? Less clear….
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
WHY IS IT DIFFICULT?: TIMESCALE AND CHEMISTRY
 Technological plasmas have vastly different timescales that must
be addressed in models.
 Integrating timestep
(numerical stability): t
 Dynamic timescale
(resolve phenomena): T
 Plasma transport:
 Dielectric relaxation
t = /  1 ps – 10 ns
 T = ns - ms
 Surface chemistry:
 t = s, T = 10 s
 Will not be resolved
“conventionally.”
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
WHY IS IT DIFFICULT?: TIMESCALE AND CHEMISTRY
 Computationally representing the gas phase and surface
chemistries (but primarily surface) are now as large an intellectual
challenge as addressing timescales.
 PR morphology weakly
dependent on
temperature with only ion
bombardment.
 Add UV…hugely
dependent.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
WHAT IS THE PATH TO
INTEGRATED DESIGN
 Tool design: Advance
model infrastructure
on par with weapons
design.
 Process design:
Advance plasma
chemistry knowledge
base on par with
combustion and
catalysis.
 Most direct:
Understand linkages.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
ONE-PATH UP THE MIDDLE:
HYBRID MODELING
 Hybrid models resolve multiphysics over multi-scales.
 Compartmentalize physical
processes into modules
having minimum of overlap.
 Establish interdependencies
 Time slice on physics timesscales.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
TOOL DRIVEN PROCESS DESIGN:
TWISTING IN
EXTREMELY HIGH
ASPECT RATIO
FEATURES:
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
HIGH ASPECT RATIO CONTACT (HARC) ETCHING
 Processes for HARC etching with aspect ratios > 50-100 are being
developed for capacitors and through wafer vias.
 Twisting, bowing and curvature of
features is randomly observed.
 Extremely high selectivity for
mask emphasizes controlling PR
erosion profiles.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
CHARGING IN HARC
 Features are so small that random
fluctuations of fluxes of radicals, ions and
electrons into holes produces variations.
 Closely related to charging
 High energy ions penetrate deep into feature.
 Electrons charge top of feature.
 Charging of features produce internal E-fields
that affect trajectories.
 Randomness of charging leads to erratic
paths.
 Scaling law: Flux/cm2 = constant is prohibitive
as feature sizes shrink
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
MONTE CARLO FEATURE PROFILE MODEL
 Psuedoparticles representing neutrals, ions and
electrons are directed towards surface.
Potential
 IEADs from HPEM - All ions neutralize striking
surface, depositing charge. Reflect as neutrals.
 EEADs of high energy secondary electrons from
HPEM – Electrons scatter from surface f(,).
 Maxwellian fluxes for bulk electrons with
Lambertian distribution.
 Charge neutral fluxes averaged over many rf
cycle but randomly not-neutral.
 Electric potential is solved using Successive
Over Relaxation (SOR) method.
 The charge of pseudo-particles is adjusted to
account for finite sized particles.
AVS08_MJK
-6
0
151
University of Michigan
Institute for Plasma Science
and Engineering
SiO2 / Si HARC ETCH: EFFECTS OF CHARGING
 Etch rate higher with
increasing power.
 Without charging:
 Generally straight
profiles.
 With charging:
 Ion trajectories
perturbed.
 Overcome with voltage.
 Some evidence of
randomness due to small
contact area.
 Without Charging  With Charging
AVS08_MJK
 10 mTorr, Ar/C4F8/O2 = 80/15/5,
10/ 40 MHz, 500 W.
University of Michigan
Institute for Plasma Science
and Engineering
SiO2/Si HARC ETCH: RANDOMNESS OF CHARGING?
 6 trenches receiving
“same” fluxes.
 Stochastic nature of
fluxes produces random
twisting.
 Similar behavior
observed experimentally.
 Effect is amplified by
finite size of particles
and mesh.
 10 mTorr, Ar/C4F8/O2 =
80/15/5, 300 sccm, LF 4
kW, HF 500 W.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
MELD TOOL AND PROCESS DESIGN
DC-AUGMENTED RF
 Single (or dual) frequency CCP…with external, negative dc bias
on opposing electrode.
 DC ion current produces dc e-beam current incident onto wafer.
 dc e-beam, mono-energetic and narrow in angle, penetrates
deep into feature to neutralize excess positive charge.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
10 MHz LOWER, DC UPPER: PLASMA POTENTIAL
 LF electrode passes rf current. DC electrode passes combination of
rf and dc current with small modulation of sheath potential.
 Ar, 40 mTorr, LF: 10 MHz, 300 W, 440V/dc=-250V
 DC: 200 W, -470 V
AVS08_MJK
ANIMATION SLIDE-GIF
University of Michigan
Institute for Plasma Science
and Engineering
10 MHz LOWER, DC UPPER: [e], ION ENERGY DISTRIBUTIONS
 Ion energy distribution to wafer is
many degrees, 150 eV in width.
 Electron energy distributions
onto wafer is narrower in angle
and broader in energy.
 E-beam reflects instantaneous
potential difference between
electrodes.
 Ar, 40 mTorr, LF: 10 MHz, 300 W,
440V/dc=-250V; DC: 200 W, -470 V
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
HEE EFFECTS on TWISTING:
 E-beam current neutralizes
sufficient charge to prevent major
twisting.
 Difference in etch depth results
from randomness of fluxes.
Without HEE
 Bowing occurs at later stages due
to reflection from sloped profile of
eroded PR.
 40 mTorr, Ar/C4F8/O2 = 80/15/5, 300
sccm, RF 5 kW at 10 MHz, DC 200 W.
Aspect Ratio = 1:20
With HEE
University of Michigan
Institute for Plasma Science
and Engineering
ENERGY DISTRIBUTIONS vs DC VOLTAGE
 Electrons
 Ions
 RF voltage constant while varying DC voltage.
 Slightly IEAD due to increase in plasma density with VDC.
 At low energy, low VDC broadens EEAD.
 40 mTorr, Ar/C4F8/O2 = 80/15/5,
AVS08_MJK
300 sccm, RF 1.5 kV at 10 MHz.
University of Michigan
Institute for Plasma Science
and Engineering
TWISTING ELIMINATION: DC
VOLTAGE
Different random seeds
 Two group of profiles are selected from
21 cases with different random seed
number generators.
 HEE neutralizes positive charge deep
into the trench.
 Higher HEE energy and flux produce
better profiles and higher etch rates:
 VDC=0 V,
twisting probability=7/21.
 VDC=500 V, twisting probability=5/21.
 VDC=750 V, twisting probability=3/21
 40 mTorr, Ar/C4F8/O2 = 80/15/5, 300 sccm,
RF 1.5 kV at 10 MHz.
MINGMEI_AVS08_12
Aspect Ratio = 1:20
University of Michigan
Institute for Plasma Science
and Engineering
PROCESS DRIVEN TOOL DESIGN:
HfO2 GATE STACK
ETCHING
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
HfO2 GATE STACK
MODELING
 High-k metal oxides are being used as SiO2
replacements to minimize gate leakage.
 For process integration and speed, desirable to
simultaneously etch entire gate
stack…Success with Ar/BCl3/Cl2 plasmas.
 Challenge:
 Modeling is needed to speed process
development and optimization.
 No fundamental database in the open
literature for plasma-surface interactions.
 Must develop mechanism based on
experience, sparse data from literature…
 Perform sensitivity studies to calibrate.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
ETCHING MECHANISM IN Ar/BCl3/Cl2 PLASMA
 HfO2 Etching
 Bond Breaking
 Adsorption
 Etching
M+(g) + HfO2(s)
 HfO(s) + O(s) + M(g)
M+(g) + HfO(s)
 Hf(s) + O(s) + M(g)
Cl(g) + Hf(s)
 HfCl(s)
BClx(g) + O (s)
 BClxO(s)
M+ (g) + HFClx(s)
 HfClx(g) + M(g)
M+(g) + BClxO(s)
 ByOClx(g) + M(g)
 Selectivity with respect to Si results from deposition of BClx
polymer on Si
AVS08_MJK
BClx(g) + Si(s)
 SiBClx(s)
BClx(g) + SiBClx(s)
 SiBClx(s) + Poly-BClx(s)
University of Michigan
Institute for Plasma Science
and Engineering
PR TRIMMING AND GATE STACK ETCHING
 PR trimming (80
to 32 nm) and
BARC removal in
Ar/O2 plasma.
PR
 Features are small
enough that
random fluxes
and micromasking are
important.
BARC
TiN
HfO2
SiO2
Si
Initial Feature
AVS08_MJK
 High selectivity to
Si required.
Trimming
Ar/O2 plasma
ANIMATION SLIDE-GIF
Stack Etching
Ar/BCl3/ Cl2
Plasma
University of Michigan
Institute for Plasma Science
and Engineering
HfO2 ETCH RATE AND BIAS VOLTAGE
PR
BARC
TiN
HfO2
SiO2
Si
30V Bias 60V Bias 100V Bias
 The etch rate of HfO2 increases
with bias voltage, thereby
increasing polymer sputtering
and causing lower selectivity.
ANIMATION SLIDE-GIF
 Ar/BCl3/Cl2 = 5/40/55,
AVS08_MJK
5 mTorr, 300 W ICP
University of Michigan
Institute for Plasma Science
and Engineering
SELECTIVITY: CALIBRATION
 Without fundamental data, reaction
mechanism requires calibration.
 Selectivity of HfO2 over Si depends on
a layer of Poly-BClx maintained by a
competition between bondingdeposition and sputtering.
Bonding:
BClx(g) + Si(s)  SiBClx(s)
Polymer Deposition:
BClx(g) + SiBClx(s)  SiBClx(s) + P-BClx(s)
Sputtering:
M+(g) + P-BClx(s)  M(g) + BClx(g)
Sputtering
 Calibration is obtained through
sensitivity studies.
AVS08_MJK
University of Michigan
Institute for Plasma Science
and Engineering
WHAT IS NEEDED FOR PREDICTABILITY?
 Become self-aware of the proper
algorithms and resolution for the local
physics.
 Transparently generate data and
reaction mechanisms.
 Integration of spatial and time scales,
and physical processes, by improving
computational techniques and
harnessing high performance
computing..
 Extreme challenges face
modeling…to place predictive
capabilities in the hands of
technologists ...highest level
of challenge and the highest
potential
AVS08_MJK
 Adopting standards for exchange of
algorithms, modules and data.
 Dedicated experimental collaborative
efforts – Take a lesson from
combustion sciences.
University of Michigan
Institute for Plasma Science
and Engineering