correlation between whisker initiation and compressive...
TRANSCRIPT
Kato et al. 1
Correlation between Whisker Initiation and Compressive Stress in Electrodeposited
SnCu Coating on Cu Leadframes
Takahiko Kato, Haruo Akahoshi, Takeshi Terasaki, Tomio Iwasaki, Masato Nakamura
Hitachi, Ltd., JapanTomoaki Hashimoto, Asao Nishimura
Renesas Technology Corp., Japan(E-mail: [email protected])
Tin Whisker Workshop, Reno, NV, May 29, 2007
Kato et al. 2
OutlineOutlineSubject: whiskers formed at R.T. over long term
from SnCu coating on Cu leadframe.Part 1: Correlation between whisker formation, microstructure, and stress
- Two samples with same SnCu coating on two different Cu leadframes that demonstrate drastically different whiskerinitiation tendencies
- Whisker formation: SEM- Microstructures: FE-TEM/FE-STEM, EBSP- Stress: X-ray diffraction method - Stress gradient: calculation by FEA
Part 2: Correlation between whisker initiation sites and Sn diffusion sites
- Sn diffusion sites under compressive stress: calculation by molecular dynamics
- Whisker initiation sites: planar slice method
Kato et al. 3
Part 1Part 1
Correlation between WhiskerFormation, Microstructure,
and Stress
Kato et al. 4
The same SuCu coating on two different Cu leadframesThe same SuCu coating on two different Cu leadframes
(a) SnCu-CUCR lead
52 months No Whiskers
SEM
100 µm
(b) SnCu-CUFE lead
34 months
Whiskers: max. > 200 µm
SEM
- Electrodeposited ∗- Matted- ∼2 mass% Cu –
bal. Sn- Thickness: 10 µm
Coating
Cr: 0.3, Sn: 0.25, Zn: 0.2, Cu: bal. (mass%)CDA number: 18045SnCu-CUCR
Fe: 2.4, Zn: 0.13, P: 0.08, Cu: bal. (mass%)CDA number: C19400SuCu-CUFE
Commercial Cu leadframeSample
∗ Electrodeposition was done in a commercial environment
Kato et al. 5
Age of samples for each examination (periodic inspection)Age of samples for each examination (periodic inspection)
(Optical inspection)
SEM Observation of whiskerFE-TEM, FE-STEM analysis
SEM observation of whiskerEBSP measurement
M: months
24 M
42 M
27 M
45 M 52 M
34 M
65 M
47 M
- Two sets of samples were prepared for each examination.
- Each exam was conductedduring the same inspectionperiod.
- SnCu-CUCR sample withno whiskers is always older than SnCu-CUFEsample at eachinspection period.
SnCu-CUFE
SnCu-CUCR2005 2006 2007JFY
Made sample
X-ray stress measurement
163 days
98 days
JFY 2001 2002 2003 2004 2005 2006 2007
Made sample
SnCu-CUFE
SnCu-CUCR
Made sample
Kato et al. 6
Crystalline orientations (EBSP measurement)Crystalline orientations (EBSP measurement)
- SnCu coatings exhibit columnar structure buthave no oriented grains.
- Cu leadframes have cold-rolling texture.- Cu leadframe does not create epitaxial SnCu coating growth for either samples.
Cross-sectionof whisker
5 µm
Cross sections of leads
SnCu coating: β-Sn tetragonalstructure
Cu6Sn5 IMC:hexagonalstructure
Cu leadframe:FCC structure
65 months 47 months
(b) SnCu-CUFE lead(a) SnCu-CUCR lead
Standard triangle of stereogram
Kato et al. 7
Grain size distributionGrain size distribution
Evaluated by EBSP measurement
SnCu-CUFE (47 months)SnCu-CUCR (65 months)
1 2 3 4 50Grain diameter (µm)
0.15
0.1
0.05
0.0
0.25
0.2
Num
ber f
ract
ion
SnCucoating
Grain diameter (µm)1 2 3 4 50
0.15
0.1
0.05
0.0
0.25
0.2
Num
ber f
ract
ion
SnCucoating
1 2 3 4 50Grain diameter (µm)
0.6
0.4
0.2
0.0N
umbe
r fra
ctio
n
Cu leadframe
Grain diameter (µm)1 2 3 4 50
0.4
0.2
0.0
Num
ber f
ract
ion Cu
leadframe
0.6
- No special difference in the distribution between the two samples can be distinguished for either coating or leadframe.
Kato et al. 8
Cu leadframe characteristics (TEM/EDX)Cu leadframe characteristics (TEM/EDX)
200 µm
Cr-rich particles
200 µm
Fe particles
Cr-richparticles 1. 9 x 10 20 m -3Density
Mainly 10-20 nmSize (D) Feparticles 5.0 x 10 19 m -3Density
Mainly 50-200 nm (Max. 2 µm)
Size (D)Energy (keV)
Counts
(a. u.)
0 20 40 60 80CuFe
Fe
Fe
EDX resultEx. (at%)Fe : Cu =97.7 : 2.3
(a) CUCR leadframe (b) CUFE leadframe45 months 27 months
Energy (keV)
Counts
(a. u.)
0 20 40 60 80
Cu
CuCr
Cr
CuEDX result
Ex. (at%)Cr : Cu =39. 6 : 60. 4Cr
Kato et al. 9
Microstructural characteristicsMicrostructural characteristics< SnCu coating - CUFE leadframe >
Fine-grained intermetallic compound (FGIMC) layer formed between coating and leadframe. Large-grained intermetallic compound (LGIMC) built up on the FGIMC layer with atriangle configuration (a wedge-shaped structure) along GBs in coating.
1.5 µmFGIMC layer
Sn/Cu coating
Cu leadframe
LGIMC
Grain boundaries
FE-STEM Bright Field Image EDX Mappingsame field
27 months Cu-K
Sn-L
1.5 µm
Kato et al. 10
Microstructural characteristicsMicrostructural characteristics< SnCu coating - CUCR leadframe >
same field
FGIMC layer
Cu leadframe
LGIMC
SnCu coating
LGIMC
1.5 µm
GB
FE-STEM Bright Field Image
45 months
Sn-L1.5 µm
Cu-KEDX Mapping
GB GB GBGB GB
As in the SnCu-CUFE sample, an LGIMC formed on the FGIMC layer. Note though, that LGIMC has a comb-teeth structure. GBs in the coating keep acting as LGIMC formation sites in this sample, too.
Kato et al. 11
Identification of IMCsIdentification of IMCs
pts.: number of analyzed points
SnCu-CUFE
(27 months)
(Av. of 5 pts.)
Cu Sn
at%6050403020100
Cu6Sn5
(Av. of 3 pts.)
Cu Sn
at%6050403020100
Cu6Sn5
(Av. of 5 pts.)
Cu Sn
at%6050403020100
Cu7Sn5
(Av. of 3 pts.)
Cu Sn
at%6050403020100
Cu7Sn5
LGIMC:hexagonalCu6Sn5
FGIMC:hexagonalCu7Sn5
LGIMC:hexagonalCu6Sn5
FGIMC:hexagonalCu7Sn5
00001011-1121--
0110-
Cu6Sn5 [1011]-Hexagonal
00000111-1012--
1103- -
Cu6Sn5 [2111]-Hexagonal
Cu6Sn5 [1232]-Hexagonal
0000
2110-2021-
0110-
EDX analysis
FGIMC Determinedphase
structureand C.C.
LGIMCSample(Age) Electron
diff. pattern EDX analysis Electrondiff. pattern
SnCu-CUCR
(45 months)
Cu6Sn5 [1102]-Hexagonal
00001101-
2021-1120-
Kato et al. 12
Relationship between the distribution of Fe particles and the coRelationship between the distribution of Fe particles and the configuration of IMCsnfiguration of IMCs< SnCu Coating - CUFE Leadframe >
Fe particles are present in leadframe and FGIMC layer but not in the LGIMC and coating regions.> Interface between FGIMC layer and LGIMC is found to be the original surface position of the
Cu leadframe before electrodeposition.LGIMC growth is partially suppressed by the Fe particles in dense Fe areas like circles A to E.The top of the triangle-shaped LGIMC is always located at GBs.Superimposition of these two effects of Fe particles and GBs for IMC formation results in the triangular configuration of the LGIMC.
Trace of IMCs on EDX mapof Fe particles
1.0 µmFe - K
A B C D E
Cu leadframe
LGIMC
FGIMC layer
InterfacebetweenFGIMC layerand LGIMC
SnCu coating
LGIMC
GBsGBs
27 months
Kato et al. 13
< SnCu Coating - CUCR Leadframe >Distribution of CrDistribution of Cr--rich particlesrich particles
As with the Fe particles, Cr-rich particles are distributed in FGIMC and Cu leadframe but are not seen in LGIMC and coating region. > Interface between the FGIMC layer and the LGIMC is found to be the original
surface position of the Cu leadframe before electrodeposition in this case, too.Cr-rich particles have no obvious trace where LGIMC growth is suppressed, and only the GBs in the coating keep acting as LGIMC formation sites (see also page 10). > This causes the LGIMC to form the comb-teeth structure in this case.
200 nm Cr-K45 months
SnCuLGIMC
FGIMC layer
Cu leadframe
GB GB GBGB GB
Cu leadframe
SnCu coating
FE-STEM
FGIMC layer
Trace of IMCs on EDX map of Cr-rich particles
LGIMC
Kato et al. 14
Comparison of FGIMC configurationsComparison of FGIMC configurations
Distance along longitudinal direction of leads (µm)
Thi
ckne
ss o
f FG
IMC
laye
r (n
m)
Av:385 nm
0 2.5 5.0 7.00
200
400
600
800
1000
12001400 (a)
Av:859 nm
0200
400
600
800
1000
1200
1400 (b)
Cr-K
Fe-K1 µm
SnCu-CUCR
SnCu-CUFE
Thickness of FGIMC layer
Thickness of FGIMC layer
The FGIMC layer in SnCu-CUCR is more than twice as thick as that in SnCu-CUFE. > Diffusion of Sn from coating to leadframe is suppressed by Fe particles for SnCu-CUFE.
Sn diffusion is easy
Sn diffusion issuppressed byFe particles
Longitudinal direction of leads
Kato et al. 15
Comparison of LGIMC configurationsComparison of LGIMC configurations
Cu diffusion along GBs in coatingfrom Cu leadframe could be easy.
Cu diffusion along GBs in coating could be suppressed by Fe particles.
Average height of LGIMC: 4,220 nm
Cu-K1 µm
Cu diffusion could be easy
GB
GB
GB GBGB GB
LGIMC
1 µmCu-K
Cu diffusion could be suppressed byFe particles
GB GB GBGB
GBGB
LGIMC
Average height of LGIMC: 2,870 nm
Cross-sectional configuration of the comb-teeth could be given for LGIMC.
Cross-sectional triangular (wedge-shaped) configuration could be given for LGIMC.
<Assumption>
SnCu-CUCR SnCu-CUFE
Kato et al. 16
Prediction of stress induced in coatingsPrediction of stress induced in coatings
SnCu-CUFESnCu-CUCR
Cu-K1 µm
GB GB GBGB GB
LGIMC
1 µmCu-K
GB GB GBGB
LGIMC
LGIMC configuration oftriangle-shaped structure
LGIMC configuration ofcomb-teeth structure
Small compressive stresscould be induced.
Large compressive stresscould be induced.
Kato et al. 17
Residual stress measurement by XResidual stress measurement by X--ray diff.ray diff.sin2ψ method
Iso inclination methodClassification in detector scanning planeFixed ψ methodClassification in X-ray incident methodSnCu-CUFE: 98, SnCu-CUCR: 163Age of samples (days)
ψ=0 to ψ=45ºOffset angle(312) plane of β-Sn structureEvaluated peak
0.35Poisson’s ratio ν
5 mm x 4 mmStress-measured area
X: longitudinal direction of leadY: width direction of lead
43.5 G Pa
Eleven data points
Measured direction
Young’s modulus E
Data in 2 θ vs. sin2ψ diagram
Fundamental equation for stress measurementE π δ 2θ
2(1+ ν) 180 δ sin2 ψσ = cot θ0. . .
Kato et al. 18
Stress data from sinStress data from sin22ψψ--22θθ diagram diagram
Examples of sin2ψ−2θ diagram
SnCu-CUFESnCu-CUCR(Age: 98 days)(Age: 163 days)
0.0 0.2 0.4 0.6sin2ψ
144.2
144.0
143.82θ (d
eg.)
0.0 0.2 0.4 0.6sin2ψ
2θ (d
eg.)
144.2
144.0
143.8
X direction -19 MPa X direction -29 MPa
- All data in the sin2ψ diagram reveal good linear relationships.- Residual stress is reflected in the slopes.
Kato et al. 19
Residual stress measurement resultsResidual stress measurement results
- Compressive stress of SnCu-CUFE is roughly double that of SnCu-CUCRfor both directions of X and Y.
- Drastic difference in whisker initiation tendencies between the SnCu-CUCRand SnCu-CUFE samples can be explained in terms of difference in residualstress in the coatings.
Stre
ss in
Sn/
Cu
coat
ing
(MP
a)
SnCu-CUFE(98 days)
SnCu-CUCR(163 days)
Y directionX directionY directionX direction
0
-10
-20
-30
X direction
Y directio
n
(1) Measurements for same samples were repeated twice for each direction X and Y.
(2) Error bar for each data point indicates confidence limit.
(3) X is longitudinal direction and Y is width direction of lead.
Kato et al. 20
Model of whisker initiation (intermediate summary)Model of whisker initiation (intermediate summary)
SnCu-CUCR
SnCu-CUFE
SnCu coating
Grain boundaries
Cu leadframe
Fine Cr-rich particles
Fe particles
Cu diffusion
No whiskers
Whisker initiation
Large compressive stress is induced
Fe particles prevent Cu diffusion
Cu diffusion is easyCu leadframe
Compressive stress is small
Sn diffusion is suppressed by Fe particles
Sn diffusion is easy
LGIMCFGIMC
LGIMC
FGIMC
Formation morphology of LGIMC is a wedge-shaped structure, that induces large compressive stress in coating.
Formation morphology of LGIMC is a comb-tooth structure, which diminishes compressive stress in coating.
Sn diffusion is easy due to smallness of Cr-rich particles in CUCR frame, so a thick FGIMC layer is formed.
Fe particles exist in CUFE frame. Sn diffusion is suppressed by Fe particles, so FGIMC layer is thin.
SnCu coating
Grain boundaries
Kato et al. 21
Calculation of stress distribution in coating Calculation of stress distribution in coating
- To determine the essential contribution of stress to whisker initiation
- FEA (Finite Element Analysis) to evaluate the stress distribution- Configurations of LGIMCs in the two samples were taken into account
in the FEA calculation. - To clarify the difference in stress distribution between two samples.
SnCu-CUCR SnCu-CUFEGB GB GBGB GB
Cu-K1 µm
LGIMC
LGIMC configuration ofa comb-teeth structure
1 µm Cu-K
GB GB GBGB
LGIMC
LGIMC configuration ofa wedge-shaped structure
Kato et al. 22
FEA FEA ((FFinite inite EElement lement AAnalysis)nalysis) modelmodel• A two dimensional plane strain condition
Analytic area
x
y
10 µ
m
3 µm
GB
LGIMC
2.25
µm
0.5 µm
Columnar crystal
LGIMC
GB
2 µm
3 µm
10 µ
m
x
y 1 µm
Analytic areaColumnar crystal
CUCR (excluded in model) CUFE (excluded in model)
x,y fix
x fix
y fix
x,y fix
x fix
SnCu-CUFESnCu-CUCR
y fix
Kato et al. 23
Material constantsMaterial constants
• SnCu coating: bi-linear elastic-plasticdeformation behavior
– Young’s modulus: 43.5 GPa– Poisson's ratio: 0.35– Yield stress: 30 MPa – Strain hardening coefficient: 700 MPa
(all data were estimated from bulk solder)
• LGIMC: elastic deformation behavior
– Young’s modulus: 100 GPa (speculation)– Poisson's ratio: 0.01 (speculation)
S-S curve
S-S curve
Kato et al. 24
SetSet--up conditions of stress in coatingup conditions of stress in coating
SnCu-CUCR SnCu-CUFE
GB
x
y
SnCu SnCu SnCu
SnCuSnCu
GB
x
y
SnCu SnCu SnCu
SnCuSnCu
Initial: 2.25 µm Growth: 2.86 µm Initial: 1 µm
Growth: 1.28 µm
Initial: 0.5 µmGrowth: 0.64 µm
Initial: 2 µm Growth: 2.56 µmLGIMC
Surface
LGIMC
- Initial stress in coating including LGIMC with initial configuration is zero.- Distribution of induced stress is calculated when LGIMC grows by up to25% in both x and y directions, as shown above.
Kato et al. 25
FEA results for XFEA results for X--directional stress distribution directional stress distribution X-directional stressσx is normal to GBSnCu-CUCR SnCu-CUFE
Tension
Compression
-130
20
- 50- 70
-100
σx (MPa)
- 20GB
Tension
Large compressive stress fieldx
y
LGIMC
GB
Tension
Tensile stress fieldx
y
LGIMC
Compressivestress gradienttoward surface
along GB
- Two-directional compressive stress gradient toward surface and toward leadframe.- The compressive stress gradient toward surface in CUCR sample is smaller than
that in CUFE sample.
Kato et al. 26
Stress distributions along GBStress distributions along GB
- Stress gradient along GB (slope) toward surface in CUCR sample is smallerthan that in CUFE sample.
- From a relationship between stress gradient and atom flux along GB (next page), atom flux toward surface in CUFE is thought to be larger than that in CUCR.
-60
-40
-20
0
20
40
60
0 2 4 6 8Distrance from A to C(C') (µm)
σx o
n G
B
(MP
a)
CUCRCUFE
A
B
B’
Surface Leadframe
Atom fluxtoward surface
Atom flux towardtwo directions
C’
C
σ x
alon
g G
B (M
Pa)
Distance from A to C (C’) (µm)(σ x : X-directional stress normal to GB)
CUCR
B
A
C
CUFE A
B’
C’x
y
Kato et al. 27
Relationship between stress gradient and atom fluxRelationship between stress gradient and atom flux
µ = µ0 − σn Ω (2)
: Atom flux: Coefficient of GB diffusion : Boltzmann constant: Absolute temperature: Reference value of chemical potential : Normal stress to the GB: Atomic volume: Direction along GB
JBDBKTµ0σnΩs
DB δµ
kT δsJ B = along GB (1)
(1) A. Needleman and J.R. Rice, Acta Metallurgica, 28 (1980), 1315.(2) C. Herring, J. Applied Physics, 21 (1950), 437.
Columnar crystal
Com
pres
sive
str
ess
Com
pres
sive
str
ess
GB
Crystal growth Whisker
Stress gradientnormal to GB
s
Atom flux along GB
Grain boundary (GB) diffusion
Kato et al. 28
Conclusions of FEAConclusions of FEA
(1) Gradient along GB of compressive stress normal to GB in SnCu-CUCR sample is smaller than thatin SnCu-CUFE sample.
(2) Difference between whisker initiation tendenciesof two samples is attributable to difference inamount of atom flux toward surface along GB, which is due to compressive stress gradient.
Kato et al. 29
Part 2Part 2
Correlation between Whisker Initiation Sites and
Sn Diffusion Sites
Kato et al. 30
Simulation of atom diffusion by molecular dynamicsSimulation of atom diffusion by molecular dynamics
- Periodic boundary conditions applied in x, y, and z directions- Diffusion coefficient for y direction calculated from Einstein’s relation
D = lim D (t), D(t) = < [yi(t +t 0)- yi(t 0)]2 >/2t
yi(t +t 0)- yi(t 0): displacement of atom i in y direction< >: average over diffusing atoms
t → ∞
- Atomic diffusion simulated by solving Newton’s equation of motion: mid2ri/dt2=Fi
- To clarify grain boundary diffusion as compared with bulk diffusion- Stress dependence of GB-diffusion coefficient by applying stress σx
1nm
Grain BGrain A
Stress σx
Coincident grain boundary Σ 5 (110)
Stress σx
A bi-crystal model (2x104 atoms) Bulk model (1x104 atoms)
x
y
x
y
Single crystal
Cu atoms(1at%Cu-Sn)
1nm
Kato et al. 31
Simulation results of interface and bulk diffusionSimulation results of interface and bulk diffusion
Diffusion coefficient calculated from molecular dynamics
1.E-55
1.E-50
1.E-45
1.E-40
1.E-35
1.E-30
1.E-25
1.E-20
1.E-15
0 0.5 1 1.5 2 2.5 3 3.5
Diff
usio
n co
effic
ient
(m
2 /s)
1/T (103/K)
Appliedcompressivestress σx
200 MPa50 MPa0 MPa
200 MPa50 MPa0 MPa
Bulk
20ºC100ºC
200ºC
GB: Σ 5 (110)
- Grain boundary diffusion coefficients are much larger than bulk- diffusion coefficients under same temperature and stress conditions.
- Grain boundary diffusion could be a more dominant factor in whisker initiation than bulk diffusion.
- Whiskers could preferentially initiate immediately on the GBs of the coating.
Kato et al. 32
Confirmation of whisker initiation sites Confirmation of whisker initiation sites -- planar slice method planar slice method --
1
Direction 1side elevation view
4Direction 4side elevation view
side elevation view
2
3
Direction 3side elevation view
Whiskers from front view (tilt=0º)
2 µm
A
B
2 µm Direction 2(tilt=45º)
A B
2 µm Direction 3(tilt=45º)
A
B
2 µm Direction 1(tilt=45º)
AB
2 µm Direction 4(tilt=45º)
A
B
SEMLocation of whisker roots
Direction 2
SEM observation of whiskers from various angles in SnCu-CUFE sample with an age of 47 months
Kato et al. 33
Planar slice methodPlanar slice method- To confirm correlation between whisker root and coating microstructure,sample was sliced horizontally using a planer slice method.
Direction of SEM observation Location of whisker roots
LGIMC10nmFGIMC
Cu leadframe
SnCu coating
Surface
Whiskers
GB
Location of whisker roots
A
B
SEM
Whisker AWhisker B
Whiskers in a front of view (tilt=0º)
Kato et al. 34
Planar slice methodPlanar slice method
Step 1: Protecting whiskers with resin coating
10nm
Thinned resin
Horizontal section of whisker
10nm
Resin
Whisker
Whisker AWhisker B
A
B
Backscattered electron image in a front of view
Location of whisker roots
Thinned transparent resin coating
Direction of BEI observation(BEI: backscattered electron image)
Protect whiskers by plastering resin on surface
Cutting whiskers by mechanical reduction of resin
Kato et al. 35
Standard triangle of stereogram
Step 2: Thinning of SnCu coating
1 µm
EBSP IPF (orientation) map: normal direction
SnCugrain
IMC
Location of whisker roots
1 µmBackscattered electron image
IMC
SnCugrain
- Whisker roots always located at intersections of GBs.- Whisker initiation sites must be grains immediately on top of GBs.
Direction of BEI observation and EBSP measurement
∼ 1 µm
10nm
Thinning of SnCu coating surface
Location of whisker roots
Planar slice methodPlanar slice method
Kato et al. 36
Direction of SEM observation
Step 3: Removal of SnCu coatingby chemicaldissolution
1 µm
Location of whisker roots
LGIMCs with a shape of pyramid A front view of the shape of LGIMCs in SEM image
Cu leadframe
Location of whisker roots
Planar slice methodPlanar slice method
Coating was completely removed.
LGIMC
Whisker roots always located above peaks and ridge lines of LGIMC-pyramids.
Kato et al. 37
Discussion of Part 2 Discussion of Part 2 (1) Whisker initiation sites correlate with Sn diffusion sites (i.e.,
GBs), which is explained by the finding that whisker roots arealways located on top of the intersections of GBs (a result ofplanar slice method) and that dominant Sn diffusion sites are GBs when compressive stress is applied from the normal direction (a result of molecular dynamics calculation).
(2) Whisker initiation sites are located above the peaks of LGIMC-pyramids (a result of planar slice method). So, taking into account the results of FEA and X-ray stress measurement, it is reasonable to assume that LGIMC-pyramids could induce a large compressive stress field above the pyramids, resulting in the enhanced Sn diffusion at GBs immediately on top of the pyramids.
(3) It can be concluded that whiskers could initiate from the grain located immediately on top of the GB due to the compressive stress field normal to the GB, which is induced by the LGIMC-pyramid.
Kato et al. 38
SummarySummary(1) A considerable difference between the whisker initiation
tendencies of SnCu-CUCR and SnCu-CUFE samples is explainable through the correlation between whisker formation, microstructure, and stress.This correlation was supported by evaluation of the stress gradient in the SnCu coating deduced by FEA.
(2) Whisker initiation sites in the SnCu-CUFE sample are preferentially located on top of the intersections of grain boundaries in the SnCu coating and above the peaks of the LGIMC-pyramids. This fact corresponds to findings for dominant tin diffusion sites (GBs) calculated by molecular dynamics.