You are on page 1of 4

Predicted Thermal Stresses in a TSV Design

Ephraim Suhir, Sung Yi,


Depts. of Mechanics and Materials Engineering and Dept. of Mechanical and Materials Engineering.
Electrical and Computer Engineering, Portland State University, Portland, OR, USA
Portland State University, Portland, OR, USA syi@pdx.edu
suhire@aol.com

Abstract— Physically meaningful and easy-to-use analytical if the aspect ratio of a particular TSV design of interest is
predictive stress models are developed for a through-silicon-via between 0.25 and 2.5, the stress data for such a design can be
(TSV) design using theory-of-elasticity based approach. Two obtained by interpolation between the calculated data for disc-
extreme cases of the TSV height-to-diameter ratios are like and rod-like vias.
considered: disc-like vias, with the aspect ratios below 0.25 The objectives of the analysis are to assess the induced
(plane stress approximation can be employed in this case), and
rod-like-vias, with aspect ratios, above 2.5 (plane strain
stresses and the pressure on Si, when the TSV structure is
approximation is applicable in this case). The objectives of the heated up, as well as the role and effectiveness of a surrogate
analysis are 1) to evaluate the effect of the size of the opening in material (i.e., an additional material that is not needed from
the silicon (Si) material on the pressure at its boundary with the the standpoint of the functional performance of the design) as
disc/rod, when the TSV structure is heated up, and the Si a possible stress relief means. While the pressure at the Si
material is in tension because of the elevated circumferential opening determines the reliability of the Si material, the
stresses, and 2) to assess the incentive for using a compliant shearing stress at the Cu-Si interface is critical from the
surrogate layer to relieve the shearing stresses at the Cu-Si standpoint of the adhesive and cohesive strength of the TSV
interface. While the pressure at the Si opening is important from design, including the barrier, seed and surrogate materials.
the standpoint of the strength of the Si material, the level of the
interfacial stress is critical from the standpoint of the adhesive
and cohesive strength of the TSV structure. II. ANALYSIS
Keywords—3D packaging; through-silicon-vias; modeling; thermal A. Radial Pressure
stress; theory-of-elasticity; design-for-reliability; reliability physics In the case of a disc-like via, one can proceed, using the
plane stress approximation, from the expression [6]
I. INTRODUCTION 1 1  r12  (1)
Through-silicon-via (TSV) designs have attracted during u(r) = (1 − ν )(γ 2
p − p ) r − (1 + ν )( p − p )
r 
1 2 2 1
the last years significant attention of 3D IC researchers and E 1−γ 2 
reliability engineers [1-4]. TSV structure, as any other IC for the radial displacements in a disc subjected to the radial
design, experiences thermally induced stresses. These stresses, pressures p1 and p2 acting on its inner outer radii r1 and
if high, could compromise the short- and long-term strength of
the structure. Design for reliability, including thermal stress r2 , respectively. Here E and ν are the elastic constants of
modeling, is therefore an important part of the 3D IC TSV r
engineering effort. Analytical modeling [5] has numerous the material, and γ = 1 is the radii ratio. Assuming that the
advantages over the widely used finite-element-analyses
r2
(FEA) in terms of compactness, time savings and clear forces due to the pressures p1 and p2 are equal, so
indication on the underlying reliability physics and clear
that p 2 = γp1 , the formula (1) yields:
indication on what affects what in the design of interest. Such
modeling should always be considered therefore in addition to 1 p2  rr 
u(r) =  − (1 − ν ) r + (1 + ν ) 1 2  . (2)
the FEA simulations. Analytical and FEA modeling are based, E 1+γ  r 
as a rule, on different assumptions, and if the computed data Applying this general expression to the Cu/Si interface, the
obtained using these approaches is in good agreement, then following compatibility condition for the radial interfacial
there is a reason to believe that trustworthy information is displacements at the Cu/Si boundary can be obtained:
obtained.
1 −ν1 1+ν2
Accordingly, a simple, easy-to-use and physically α 1 Δ tr1 − p 1 r1 = α 2 Δ tr2 + p 2 r2 − Δ u . (3)
meaningful analytical predictive stress model using theory-of- E1 E2
elasticity approach is developed in this analysis for a typical Here E1 and ν 1 , and E 2 and ν 2 are the elastic constants of
TSV design. Plane stress approximation is used for disc-like
vias, with the height-to-diameter ratio below 0.25, while the Cu and Si, respectively, α1 and α 2 are their CTEs, and Δt is
plain strain approximation is used for the rod-like vias, with the change in temperature. The first terms in the left and in
the height-to-diameter ratio above 2.5 [6]. Although disc-like the right parts of the condition (3) are unrestricted (stress-free)
via designs are viewed as not very practical today, the results displacements, the second terms are the displacements caused
obtained for such vias can still be useful, both because there is by the thermally induced pressures, and the term Δu is due to
a definite incentive for employing such vias, and also because the compliant bond, if any. This term considers the difference
in the displacements at the outer and the inner boundaries of ∂ σ r ∂ τ rz σ r − σ θ
+ + =0
the bond. With the relationship p 2 = γp1 , the condition (3) ∂r ∂z r
yields: is always fulfilled, and the second condition
γα 1 − α 2 ∂ ( r τ rz ) ∂σ z
p2 = Δt , (4) +r =0
1 −ν1 1 +ν 2 2 1−γ ∂r ∂z
+ + yields:
E1 E2 E0 1 + γ
1 d ( r τ rz ) dσ z .
where E 0 is Young’s modulus of the compliant bond. = −
r dr ∂z
In the case of a rod-like via, using the plane strain Since the left part of this relationship is z independent, and its
approximation, and the formula [6] right part is r independent, each of these parts must be a
constant value:
1 1 +ν  r12  (5)
u(r) = (1 − 2ν )(γ 2
p − p )r − (1 + ν )( p − p ) 1 d ( r τ rz ) dσ z
E 1 − γ 2  r  =− = 2C 0 .
1 2 2 1
r dr ∂z
for the radial displacements, and considering that p2 = γp1 , Then
we have: C1
τ rz = C 0 r + ,
p 1+ν  rr  r
u (r ) = 2  − (1 − 2ν ) r + (1 + ν ) 1 2  (6) and
E 1+ γ  r 
and σ z = −2C0 z + C 2 .
(1 + ν )( 2 − ν ) 1 − γ The boundary conditions
Δu = − p 2 r2 . (7)
E 1+ γ τ rz ( r1 ) = τ 1 , τ rz (r2 ) = τ 2
The compatibility condition for the radial displacements at the lead to the following formulas for the constants of integration:
Cu/Si boundary is τ r − τ 1 r1 , C = τ 1 r2 − τ 2 r1 r r .
(1 + ν 1 )(1 − 2ν 1 ) C 0 = 2 22 1 1 2
(1 + ν 1 )α 1Δ tr1 − p1 r1 = r2 − r12 r22 − r12
E1 (8) Since there are no external forces acting at the end cross-
1 +ν 2 sections of the tube, the condition τ 2 = γτ 1 for the shearing
= (1 + ν 2 )α 2 Δ tr2 + p 2 r2 − Δ u
E2 stresses should be fulfilled. Then
Solving this equation for the p2 pressure we have: C0 = 0, C1 = τ 1r1 = τ 2 r2 ,
and
(1 + ν 1 )γα1 − (1 + ν 2 )α2
p2 = Δt . (9) C1
(1 + ν 1 )(1 − 2ν 1 ) 1 + ν 2 (1 + ν 0 )(1 − 2ν 0 ) 1 − γ τ rz = , σ z = C2 .
+ + r
E1 E2 E0 1+ γ Because of the taken assumption of zero normal stresses, the
B. Interfacial Compliances constant C2 should be zero: C 2 = 0. Then the axial strain
Since in this analysis the interfacial shearing stress is ∂w 1
addressed using the concept of the interfacial compliance [7,
εz = = [σ z − ν (σ r + σ θ ) ]
∂z E
8], one should determine first the interfacial compliances of should be zero as well, and the axial displacements w could be
the TSV structural elements with respect to shear loading
a function of the radius r only: w = w(r ) . The shearing
applied to the inner and/or to the outer surfaces of the
compliant bond. (angular) strain γ rz is determined as [6]
A simplified theory-of-elasticity solution is obtained in this ∂ u ∂ w dw
section for the interfacial compliances. γ rz = + =
Consider a thick-wall tube loaded by constant longitudinal ∂z ∂r dr
by the Cauchy formulas and as
shearing forces τ 1 and τ 2 acting along the longitudinal z
τ rz
axis in the opposite directions and distributed over the inner γ rz =
surface, r = r1 , and the outer surface, r = r2 , of the tube.
G
by the Hooke’s law, where
The following major assumptions are used: E
G =
1) The radial, σ r , and the circumferential (tangential), σ θ , 2 (1 + ν )
normal stresses are zero everywhere; is the shear modulus of the material. Hence,
2) The shearing stresses, τ rz ( r ), depend on the radius r dw C
= 1 ,
only; dr Gr
and the axial displacements are
3) The axial normal stress σ z = σ z (z ) depends on the C1
longitudinal coordinate z only. w(r) = ln r .
G
With these assumptions, the first equilibrium condition [6]
The difference between the axial displacements on the outer The first term in the right part of (10) is the stress-free
and the inner boundaries of the tube is displacement. The second term is the displacement due to the
C r τ r r τ r r thermally induced compressive force T (z ). This
Δ w = 1 ln 2 = 1 1 ln 2 = 2 2 ln 2 .
G r1 G r1 G r1 displacement is evaluated based on the Hooke’s law and
If the shearing force is applied at the inner boundary, the reflects an assumption that the longitudinal thermal
corresponding interfacial compliance is displacements are the same for all the points of the given
Δw r r cross-section. The third term in (10) is, in effect, a correction
κ1 = = 1 ln 2 .
τ1 G r1 to this assumption. This correction considers that the
If the shearing force is applied at the outer boundary, this interfacial displacements are somewhat larger than the
compliance is displacements of the inner points of the cross-section.
Δw r r Since the cross-sectional area and Young’s modulus of the
κ2 = = 2 ln 2 . Si material are significantly greater than those of the via, its
τ2 G r1
axial compliance is considerably smaller than the compliance
In the case of a copper rod, in a situation, when τ1 = 0 and of the via and need not be accounted for. We assume that this
r1 = 0, the formulas for the constants C0 and C1 yield: is true for the interfacial compliance of the Si wafer as well.
Then one could represent the interfacial axial displacement of
τ2 the Si material as
C0 = , C1 = 0,
r2 u2 ( z ) = α 2 Δtz .
and the shearing stress is In other words, the longitudinal interfacial displacements in Si
τ2 are not different of its stress-free displacements. Then the
τ rz ( r ) = r. displacement compatibility condition is
r2
 r r 
The Cauchy’s formula and the Hooke’s law equations yield: u1 ( z ) = u 2 ( z ) +  1 ln 2 τ ( z ) . (11)
dw τ 2 r ,
=  G0 r1 
dr G r2 The second term in the right part of this condition is the
so that displacement due to the compliant bond, if any,
τ2 r2 . E0 is the shear modulus of the bond material,
w(r) = G0 =
2 G r2 2 (1 + ν 0 )
The difference between the axial displacements at the and r2 is the outer radius of the bond. Then the equation (10)
boundaries r2 and r1 is results in the following equation for the sought shearing stress
τ 2 r2 2 − r1 2 . function τ (z ) :
Δw =
2G r2  1 1 r2  1 −ν 1
z

2 
r1  + ln τ ( z ) − T (ς )dς = − ΔαΔtz , (12)
τ 2 r2 2
 1G G 0 r1  πE r
1 1 0
When r1 = 0, Δw = , and the axial interfacial
2G Where Δα = α1 − α 2 . By differentiation we find:
compliance is 1 −ν1
Δw r2 κ τ ′( z ) − T ( z ) = − Δ αΔ t , (13)
κ= = . π E1 r12
τ2 2G where
C. Interfacial Shearing Stress  1 1 r 
κ = r1  + ln 2  (14)
The longitudinal interfacial displacements on the copper  2 G 1 G 0 r1 
surface can be found, in an approximate analysis, as is the longitudinal interfacial compliance of the TSV
1 −ν1
z
r1 structure, with consideration of the surrogate material.
u1 ( z ) = α 1 Δ tz −
π E 1 r12  T (ς ) dς + 2G τ ( z ) . (10) Assuming that the butt ends of the via are stress free, i.e., that
0 1
T ( ± a ) = 0, we obtain the boundary condition
Here τ (z ) is the longitudinal interfacial shearing stress,
ΔαΔt
z τ ′(a) = −
T ( z ) = 2π r1  τ (ς ) d ς κ
−a for the shearing stress function τ (z ) . Considering this
is the thermally induced force acting in the copper rod cross- condition, the following solution to the equation (13) can be
sections, and a is half the TSV height. The origin of the axial obtained:
coordinate z is at the mid-cross-section of TSV on its axis. ΔαΔt sinh kz
τ ( z ) = −k , (15)
λ cosh ka
where Thus, the carried out numerical example for a 1μm thick
1 −ν1 λ layer of Indium indicated that such a thin layer of highly
λ=2 , k = compliant buffering material resulted in the reduced the
E1r1 κ pressure at the silicon boundary by about 39% for a disc-like
are the axial compliance of the via, and the parameter of the TSV and by about 14% for a rod-like TSV. As to the
interfacial shearing stress. interfacial shearing stresses, the reductions were even larger:
For disc-like vias ( ka ≤ 0.25) about 69% and about 41%, respectively.
ΔαΔt ΔαΔ t . IV. CONCLUSIONS
τ (z) = − z , τ max = a (16)
κ κ Simple, easy-to-use and physically meaningful analytical
Thus, the shearing stress is distributed linearly along the TSV, models have been developed for the evaluation of the
and its maximum value is proportional to the height of the via
thermally induced pressure (radial stresses) and the interfacial
and is inversely proportional to the interfacial compliance of shearing stresses cause by the thermal expansion mismatch of
the TSV structure. the dissimilar materials (copper and silicon) in through-
For rod-like vias ( ka ≥ 2.5) silicon-via (TSV) structures.
Δ αΔ t Δ αΔ t Δ αΔ t The carried out numerical example for a 1μm thick layer of
τ ( z) = −k e − k ( a − z ) , τ max = k = (17)
λ λ λκ Indium indicated that such a thin layer of highly compliant
Hence, the shearing stress concentrates at the TSV ends and The developed models enable one to assess the role of the size
drops exponentially when moving away from the ends. The of the opening in the Si wafer and to assess if there is an
maximum shearing stress becomes TSV height independent incentive to bring in a “surrogate” compliant bonding layer for
and is inversely proportional to the square root of the product lower interfacial stresses at the Cu/Si boundary by about 39%
of the axial and the interfacial compliances. Comparing the for a disc-like TSV and by about 14% for a rod-like TSV. As
obtained formulas for the maximum stresses, we conclude that to the interfacial shearing stresses, the reductions were even
the compliant bond has a substantially larger effect on disc- larger: about 69% and about 41%, respectively.
like TSVs than on rod-like TSVs. Our calculations show also that, as far as the induced stresses
are concerned, there is an incentive for employing disc-like
III. NUMERICAL EXAMPLE TSV structures rather than rod-like structures. In the absence
Input Data: of the compliant bond the pressure in the disc-like structure
was only about 58% of the pressure in the rod-like structure,
r1 = 10μm = 0.010mm; E1 = E 2 = 12000kg / mm 2 ; and the maximum shearing stress was about half of the stress
ν 1 = 0.34; G1 = 4478kg / mm 2 ; ν 2 = 0.22; in the rod-like structure.
Future work will include FEA calculations, as well as
α1 = 20 x10 −61 / 0 C ; α 2 = 2.5 x10 −61 / 0 C ; experimental investigations.
Δt = 3000 C ; E0 = 1300kg / mm 2 ; ν 0 = 0.45; REFERENCES
G0 = 448kg / mm 2 (Indium). [1] D. Witarsa, et al, “Through Wafer Copper Via for Silicon
Based SiP Application”, 7-th EPTC, 2005.
Calculated Data: [2] C. Okoro, et al, “Prediction of the Influence of Induced
Stresses in Silicon on CMOS Performance in a Cu-Through-
r2 , μm 10.0 11.0 12.0 13.0 14.0 15.0
Via Interconnect Technology”, EuroSimE, 2007.
Predicted pressure at the Cu/Si boundary, kg / mm 2 [3] N. Tanaka, et al, “Mechanical Effects of Copper Through-
Disc-like 33.503 20.459 14.33 10.81 8.559 6.998 Vias in a 3D Die-Stacked Module”, 52-nd ECTC, 2002
TSV [4] W. Worwag and T. Dory, “Copper Via Plating in Three
Rod-like 51.856 44.806 39.21 34.68 30.952 27.833 Dimensional Interconnects”, 57-th ECTC, 2007.
TSV [5] E. Suhir, “Analytical Thermal Stress Modeling in
Predicted shearing stress at the Cu/Si boundary, kg / mm 2 Electronic and Photonic Systems”, ASME App. Mech.
Disc-like Reviews, invited paper, vol.62, No.4, 2009.
TSV, 23.509 7.356 4.217 2.894 2.174 1.521 [6] E. Suhir, “Structural Analysis in Microelectronics and
a = 5 μm Fiber Optics”, Van-Nostrand, New-York, 1991
[7] E. Suhir, “Stresses in Bi-Metal Thermostats”, ASME
Rod-like Journal of Applied Mechanics, vol. 53, No. 3, Sept. 1986.
TSV, 47.374 27.792 21.98 18.95 17.042 15.698 [8] E. Suhir, “Interfacial Stresses in Bi-Metal Thermostats”,
a = 100μm ASME J. Appl. Mech., vol. 56, No. 3, September 1989

You might also like