Documentos de Académico
Documentos de Profesional
Documentos de Cultura
J. Electron. Packag. / Volume 130 / Issue 4 / THERMAL ISSUES IN EMERGING TECHNOLOGIES THEORY AND...
The construction and usage of compact thermal models (CTMs), for the thermal analysis as well as the design of cooling
devices for electronic systems, are reviewed. These models have many advantages over the so called detailed models
based on 3D simulations, mainly being a convenient and simple quantitative description of the modeled object, when
constructional details are either unavailable or too detailed to be of use at the desired level of analysis. However, CTMs
have manifested some deficiencies in many cases, in particular, multiple chip modules (MCM) and stacked dies. The
opposite approach, detailed modeling, is more reliable, although extremely heavy. A new approach is proposed that solves
this dilemma by bridging the gap between compact and detailed models. While retaining all advantages of CTMs, i.e., having
a limited number of degrees of freedom and not requiring detailed constructional features, it can attain any required precision
level depending on the degree of complexity adopted. It gives reliable results covering all operating conditions including
MCM and stacked dies. Moreover, it gives access to data on surface temperature gradients that were never obtained before
by compact models and are highly important for reliability issues.
History: Received 28 September 2007; revised 17 August 2008; published 14 November 2008
doi: http://dx.doi.org/10.1115/1.2993132
Keywords
PACS
89.20.Kk
Engineering
81.70.Pg
Thermal analysis (materials testing)
YEAR: 2008
RELATED DATABASES
PUBLICATION DATA
Coden:
JEPAE4
ISSN:
Publisher:
ASME
REFERENCES (26)
1 of 2 29/10/2009 15:38
Welcome to IEEE Xplore 2.0: Compact Thermal Models: A Global Approach
Home | Login | Logout | Access Information | Alerts | Purchase History | Cart | Sitemap | Help
Abstract
Construction and usage of compact thermal models (CTM) for the thermal analysis as well as the design of
cooling devices for electronic systems are reviewed. These models have many advantages over the so called
detailed models, mainly being a convenient and simple quantitative description of the modeled object, when
constructional details are either unavailable or too detailed to be of use at the desired level of analysis. However,
CTMs have manifested some deficiencies in many cases, in particular MCM and stacked dies. The opposite
approach, detailed modeling, is more reliable, although extremely heavy. A new approach is proposed that solves
this dilemma by bridging the gap between compact and detailed models. While retaining all advantages of CTM's,
i.e. having a limited number of degrees of freedom and not requiring detailed constructional details, it can attain
any required precision level depending on the degree of complexity adopted. It gives reliable results covering all
operating conditions including MCM and stacked dies. Moreover, it gives access to data on surface temperature
gradients that were never obtained before by compact models and are highly important for reliability issues
Index Terms
Inspec
Controlled Indexing
cooling electronics packaging reliability thermal analysis
Non-controlled Indexing
compact thermal models electronics cooling stacked dies surface temperature gradients
thermal analysis
Author Keywords
Not Available
PACS Codes
Not Available
References
Citing Documents
1 of 1 19/10/2009 15:23
Thermal Issues in Emerging Technologies, ThETA 1, Cairo, Egypt, Jan 3-6th 2007
34 ThETA01/048
Compact Thermal Models: a global approach; Sabry & Saleh
qs = h (T ∞– T) (2.1c)
∑
N
j =1
Yij T j = Qi (2.5a)
(where h is the heat transfer coefficient by convection) as
well as radiation and non-linear convection.. In the sequel,
Ti = ∑ j =1 Rij Q j
N
(2.5b)
λ will be considered constant in order to have a linear
PDE (2.1a). This is not a limitation, since the case where
Any CTM, regardless of how it is built (a network of
λ is function of T can be easily transformed to a linear
resistors, star shaped or Delphi type) is only a special case
equation using Khirchhoff transformation. Note that
of the general matrix representation (2.5). However,
boundary conditions may still be a nonlinear function of T
matrices appearing in either forms of (2.5) (whether Y or
through h, which may represent both nonlinear convection
R) cannot have any arbitrary form. As they represent a
and radiation.
physical model relating heats and temperatures, they must
Equation (2.1a) describes the relation between respect laws of physics. In particular, they must:
temperature and heat fluxes within the domain Ω. Since - Respect first law of thermodynamics
this is a partial differential equation of the elliptic type, it
must be supplied with boundary conditions (2.1b) - Respect second law of thermodynamics
describing interactions between the modeled object and
- Respect reciprocity or the auto-adjoint nature of
outside world. They can be either of the Dirichlet type on
the governing differential equation.
the sub-domain ∂ΩD of the boundaries or Neumann on
∂ΩN (or even of Robin type if (2.1c) was used). Equations It has been shown earlier [3] that this leads to the
(2.1) can be simplified by assuming that λ is constant that following constraints that are easier to express in terms of
may have different values over different sub-domains Ωi matrix Y:
∑
N
of Ω. Two adjacent domains of different λ can be Y ij = 0 ∀ j (2.6a)
i =1
matched using the following conditions at the interface:
∑ Y ij (Ti − T )(T j − T ) ≥ 0
N
T1 = T2 ; λ1n1 ⋅ ∇T1 + λ 2 n 2 ⋅ ∇T2 = 0 (2.2) i =1
(2.6b)
Yij = Yji (2.6c)
where ni is he unit outward normal to domain Ωi. System
(2.1), can be rewritten in a dimensionless form in each where T is the average temperature. All above
sub-domain of constant λ as: constraints are respected by a network of resistors,
although this is not the most general form.
∇2T (r ) = −qv (r ) r∈Ω (2.3a)
Two main methodologies are used to construct compact
T = Ts ; n ⋅ ∇T = qs (2.3b) models (reviewed in [1, 2]): structural [4-13] and
∂Ω D ∂Ω N
behavioral [14-23]. A global approach to each of these
Characteristic length a, temperature difference [T] and methodologies will be briefly presented below.
heat flux [Q] (in Watts) can assume any set of values
2.1 Behavioral methodology
selected from boundary conditions (2.1b) provided that
the following relation is respected: In this methodology, internal constructional details are
irrelevant, only the response of the system to a set of
[Q] = λ a [T] (2.4) excitations. Take for the moment node 1 to be a “special”
Compact models are expected to be based on equation node. Suppose now we perform N-1 experiments in which
(2.3a) only, if they are to respect the BCI condition. all nodes are insulated except node 1 as well as another
Hence they must be linear, which does not preclude the node, but never the same one in different experiments.
possibility of having nonlinear expressions in (2.3b) (see Obtaining temperatures (by direct measurement or by
above note on Kirchhoff transformation). This equation numerical or analytical methods) of all nodes in each of
relates temperatures and heat fluxes at any point in the these experiments would give us sufficient data to
domain or at the surface, a relation that depends only on calculate all elements in the matrix. In fact, by
the object itself. It does not depend on applied boundary superposition of the results of the N-1 cases, we can
conditions. A compact model should not involve all such reproduce any “realistic” case in which all nodes
details, i.e. relations between ALL points, but rather a participate in heat transfer. We could have also started
relation between a very limited set of state variables with any other set of N-1 “realistic” test cases to obtain all
describing the system behavior. This set may be for matrix elements, provided the set is “complete”. This
example average temperatures and heat fluxes at different property has been defined and corresponding
“nodes” of the modeled object. When combined with mathematical criteria were derived earlier [3]. The
applied boundary conditions, this relation will allow us to procedure seems simple and straightforward. However,
get both temperatures and heat fluxes. Suppose now that due to the distributive nature of heat transfer, temperature
temperatures associated to each node are grouped in an profiles would have a large impact on the results. Two
array of size N, the total number of nodes, as well as experiments with the same average temperature on each
another array of size N for heats exchanged. With this node, but with different temperature profiles on each node,
rather simplified set of state variables, the most general would produce different heat transfer on each node, or
form of a CTM would be a matrix relating both arrays, vice-versa. This means that we may pick 2 different sets
which can take either of the following equivalent forms: of complete N-1 experiments and get 2 different models
for the same object. The Delphi solution was as follows:
ThETA01/048 35
Compact Thermal Models: a global approach; Sabry & Saleh
- Select a relatively large set (much greater than compact model of the form (2.5), but with different values.
N-1) of “plausible” experiments of practical So we are back again with the same problem, as for the
relevance, called the generating set. behavioral approach: model, CTM depends on selected
(or assumed) profiles. We have to be clever enough to let
- Optimize matrix elements (or equivalently
these profiles be as close as possible to “practical”
network resistors) to give the least possible error
situations for the model to be relevant.
in predicting results of the generating set.
- Confirm precision by using the obtained model
to predict cases in another set, called the test set, 3. THE PROPOSED APPROACH
which are outside the generating set The proposed approach is a global solution to the compact
2.2 Structural methodology model generation issue that can be applied to both
behavioral and structural approaches. Although it can also
In the structural methodology, model generation is based be applied to convective problems [24], only conduction
on model internal constructional features. In fact, the problems will be addressed here. In fact, it is mainly
behavior is obtained by solving the governing differential about how we treat the profile issue. In a previous work
equations for a more or less general case, described by [3], expected errors due to different profiles were studied
more or less “general” boundary conditions. There are in details and an error bound was derived. Error
many different ways to solve PDE’s but they are all estimation was so close to reality, which encouraged
equivalent and can be cast in a single general procedure proposing a new approach that uses these estimates to
using the Green’s function G(r,r') satisfying: improve model precision. The basic principle is very
simple:
∇2G (r , r ') = −δ(r − r ' ) r, r' ∈ Ω (2.7)
Instead of associating only one global value to T and q
(where δ is the Dirac distribution) together with a set of over each node, let us associate a finite set of constants
boundary conditions that need not be specified at this for T and q, which can be expansion coefficients of a
level. Using it will allow us to represent the solution as: truncated series (Taylor, Fourier …) representing the
T (r ) = ∑ j =1 ∫
N
{G (r ,r' )q(r' ) − n ⋅ ∇G T (r' )}dr' (2.8) profile.
r'∈Ω j The number of expansion coefficients retained depends
where Ωj is the j’th sub-domain at which heat is on the precision needed. We can virtually go to infinity in
exchanged, i.e. the sub-domain belonging to compact order to get the exact solution. Fortunately, we never need
model “node” j ranging from 1 to N. This sub-domain can to go that far. It has been shown earlier [3] that highly
be either a volume or a surface depending on the nature of oscillating modes can be safely neglected. The application
heat source. of the flexible profile method to the structural approach
has been presented earlier [25] for rather simple cases. It
At that point, every thing is still analytic. We did not will be briefly reviewed here before giving results for real
make any assumption so far. This means that (2.8), which life cases. A modified Green’s function technique is used
relates temperatures with heat fluxes, can be taken as the satisfying the following conditions:
basis for ANY compact model. Temperatures and heat
∇ 2G (r , r ') = −δ(r − r ' ) r, r' ∈ Ω (3.1a)
fluxes are still “detailed” in the sense the equation relates
local values of T and q, and not global or average values − 1 r ∈∂Ω1
over nodes. The method by which we relate local values n ⋅ ∇G (r, r ' ) = (3.1b)
0 r ∉ ∂Ω1
with global values, and hence construct the CTM out of
(2.8) would constitute the most important difference ∫r∈∂Ω Gdr = 0
1
(3.1c)
between one proposed approach to build compact models
and the other. We may not be able to obtain G in a closed where ∂Ω1 is the outside surface sub-domain
corresponding to an arbitrarily chosen “reference” node.
form, hence an approximation is needed, which is also
Using the above Green’s function, as well as the Green’s
another, although minor, factor in distinguishing between
theorem, equation (2.3a) can be transformed into:
different model generation procedures. This has an impact
T (r ) − Tav1 = ∑ j =1 ∫ G (r , r ' )q(r ' )dr '
N
on how fast and accurate model elements can be (3.2)
r '∈Ω j
calculated, which will not receive any further attention
here. But the first difference is about the philosophy and where Tav1 is the average temperature over node 1 and Ωj
ability of the obtained model to reflect real situations. is the domain of “node” j. Using any convenient
Suppose for instance that we assume that all profiles are orthonormal set φiu (r ) over each node i, we can express T
uniform, which means we simply replace local values by and q profiles as:
their global counterparts. Substituting in (2.8), taking out
T (r ) r∈Ω = ∑u = 0 Tiu φiu (r )
∞
(3.3a)
of the integral any uniform profile and performing i
may also assume any profile in the form of a function (f Hence, by substituting in (3.2) we get after multiplying
(r) say) having a unity average value. Local value of T, both sides by φiu (r ) , integrating using the orthonormal
over node k for example, can be expressed as: T = Tk f(r).
(where Tk is the corresponding global value at node k). property of φiu (r ) and truncating the series after U terms:
Substituting in (2.8) and integrating, would still give us a Tiu = ∑ j =1 ∑ v = 0 Rijuv q vj i ∈ [1, N ], u ∈ [0, U ]
N U
(3.4a)
36 ThETA01/048
Compact Thermal Models: a global approach; Sabry & Saleh
Rijuv = ∫ ∫ φiu (r )G (r, r ' )φvj (r ' )dr 'dr (3.4b)
r∈∂Ω i r '∈∂Ω j T-TOP-M0 T-TOP-M2
T-TOP-M4 T-TOP-FEM
This is the Flexible Profile Compact Model involving as
state variables at each node, not only one single value, but
rather the coefficients of the expansion of T and q profiles 45
over a given complete set. Taking a sufficient order of 40
precision over each node, typically 0 to 4, one can
approximate reasonably well any T or q profile with one 35
and the same model. It is worth noting that classical
T
approaches are explicitly or implicitly equivalent to the 30
proposed approach with the restriction of the number of
terms of the series (3.3) to only one term. 25
20
4. RESULTS 0 0.5 r 1 1.5
The proposed method was applied to a BGA package Figure 2. Temperature predictions T versus r (the distance
proposed in an earlier benchmark [26]. Constructional along diagonal) of Flexible Profile applied to the BGA
details of the package are shown in figure 1. The upper problem (test case 15) for the zero’th (M0), second (M2)
part of the figure represents the cross section given in the and forth order (M4) approximations
benchmark, the lower one being the breakup of the
original body into pieces each being a parallelepiped of a In order to show the Flexible Profile capabilities to treat
given material. Results for the test case 15 of the MCM cases, with general boundary conditions, the
benchmark are shown in figure 2 (Top node temperature structure shown in figure 5 was considered. Lateral
as a function of the radial distance r along the diagonal) as dimensions are not the same for all blocks. Lateral walls
compared to finite element results of the same problem. are not adiabatic with boundary conditions of both
Figure shows the zero’th, second and forth orders of Neumann and Robin types. Two volumetric heat sources
approximation. The second order is already quite close to were inserted inside the structure.
the FEM solution. The forth is almost indistinguishable
from FEM results. Figures 3 and 4 give results for test
cases 40 and 01 respectively, with 4 modes, showing the T-TOP-Fp T-TOP-FEM
same level of accuracy. T-BOT-Fp T-BOT-FEM
26
Mould
Die 24
T
Substrate top 22
Substrate 1 Substrate 2 Substrate 3
20
Inner balls Air Outer balls
18
0 0.5 r 1 1.5
Figure 3. Test case 44 for the BGA package proposed in
the Benchmark [26]
ThETA01/048 37
Compact Thermal Models: a global approach; Sabry & Saleh
130 109.2
125 109
108.8
T
120
108.6
T
115 108.4
110 108.2
108
105
0 0.2 0.4 X 0.6 0.8 1
0 0.5 r 1 1.5
Figure 7. Results of the hypothetical package, with only
Figure 4. Test case 01 for the BGA package proposed in the right heater energized
the Benchmark [26]
5. CONCLUSION
The new Flexible Profile approach, designed for reduced
order thermal modeling, has for the first time been applied
to a real life package model: the BGA proposed in an
earlier benchmark [26]. Results were conclusive. With a
bit more parameters than standard compact thermal
models (CTM), the Flexible Profile approach is capable
of predicting temperature profiles that are very close to
those that would have been obtained by a detailed model
(requiring significantly more parameters). These results
were obtained with only 4 modes in each direction. Hence,
the Flexible Profile approach is a global one that bridges
Figure 5. A hypothetical package structure to test the the gap between CTM and detailed models, with any
ability to treat MCM with general boundary conditions desired accuracy.
Figure 6 shows the results if only the left heater is
energized, while in figure 7, only the right heater is Moreover, the approach was capable of predicting correct
energized. results for MCM packages, which usually constitute a
problem for classical CTMs.
Last, but not least, profile data for temperature and heat
flux in this approach are a calculation outcome. This has a
T-Bot-Fp-V1 T-Top-Fp-V1
large positive impact on reliability studies.
T-Bot-Cos-V1 T-Top-Cos-V1
109.2 6. REFERENCES
109 [1] M.N. Sabry, “Dynamic compact thermal models used for
electronic design: a review of recent progress,” Proceedings of
108.8
IPACK03, International Electronic Packaging Technical
T
38 ThETA01/048
Compact Thermal Models: a global approach; Sabry & Saleh
Packaging, and Manufacturing Technology, Part A, Vol. 18, no. diode and IGBT)”, 13th International Symposium on Power
4, pp. 781 – 787, 1995 Semiconductor Devices and ICs, ISPSD '01, pp. 381 – 384, 2001
[6] G. Digele; S. Lindenkreutz and E. Kasper, “Fast electro- [21] G. Noebauer, “Creating compact models using standard
thermal circuit simulation using reduced order models”, 4th spreadsheet software”, 17th Annual IEEE Symposium
International Workshop on Thermal Investigation of ICs and Semiconductor Thermal Measurement and Management, SEMI-
Microstructures, THERMINIC, Cannes, France, pp. 115 – 120, THERM XVII, pp. 126 – 133, 2001
1998 [22] M. Rencz and V. Szekely, “Dynamic thermal multiport
[7] V. Phanilatha; M.S. Nakhla; Zhang Qi-Jun; Liu Da-Guang, modeling of IC packages”, IEEE Trans. on Components and
“Finite element transient thermal analysis of electronic boards Packaging Technologies, Vol. 24, no. 4, pp. 596 – 604, 2001
and packages using moment-matching techniques”, The 5th [23] K.L. Pandya, and W. McDaniel, “A simplified method of
Intersociety Conference on Thermal Phenomena in Electronic generating thermal models for power MOSFETs”, 18th Annual
Systems, ITHERM V, pp. 391 – 398, 1996 IEEE Symposium on Semiconductor Thermal Measurement and
[8] M.N. Sabry, “Static and dynamic thermal modeling of IC’s”, Management, SEMI-THERM XVIII, pp. 83 – 87, 2002
Microelectronics Journal, Vol. 30, pp 1085 – 1091 (1999) [24] M.N. Sabry, , Compact Thermal Models For Internal
[9] Y.C. Gerstenmaier and G. Wachutka, “Time dependent Convection, Transactions of IEEE / Components, Packaging
temperature fields calculated using eigenfunctions and and Manufacturing Technology, Vol. 28, No.1, pp 58-64, 2005
eigenvalues of the heat conduction equation”, 6th International [25] M.N. Sabry, High Order Compact Thermal Models,
Workshop on Thermal Investigation of ICs and Systems, Transactions of IEEE / Components, Packaging and
THERMINIC, Budapest, pp. 55 – 61, 2000 Manufacturing Technology, Vol. 28, No.4, pp 623-629, 2005
[10] M. Zubert; A. Napieralski and M. Napieralska, “The new [26] C. Lasance, “Two Benchmarks for the Study of Compact
general method for thermal and electro-thermal model Thermal Modelling Phenomena”, Proc. 6th THERMINIC
reduction”, 6th International Workshop on Thermal Investigation Workshop, Budapest, Hungary, pp. 217–222, 2000
of ICs and Systems, THERMINIC, Budapest, pp. 62 – 67, 2000
[11] W. Batty; C.E. Christoffersen; A.J. Panks;S. David; C.M.
Snowden and M.B. Steer, “Electrothermal CAD of power
devices and circuits with fully physical time-dependent compact
thermal modeling of complex nonlinear 3-d systems”, IEEE
Trans. on Components and Packaging Technologies, Vol. 24, no.
4, pp. 566 – 590, 2001
[12] L. Codecasa; D. D'Amore and P. Maffezzoni, “Moment
matching model order reduction of discretized thermal
networks”, 7th International Workshop on Thermal Investigation
of ICs and Systems, THERMINIC, Paris, pp 22 – 26, 2001
[13] L. Codecasa; D. D'Amore; P. Maffezzoni and W. Batty,
“Multi-Point Moment Matching Reduction of Distributed
Thermal Networks”, 8th International Workshop on Thermal
Investigation of ICs and Systems, THERMINIC, Spain, Madrid,
pp. 231-234, 2002
[14] J.V. Motto, Jr.; W.H. Karstaedt; J.M. Sherbondy. and S.G.
Leslie, “Modeling thyristor and diodes; on-state voltage and
transient thermal impedance, effective tools in power electronic
design”, Industry Applications Conference, IAS '97, Vol. 2, pp.
1182 – 1189, 1997
[15] C. Lasance; D. Den Hertog and P. Stehouwer, “Creation
and Evaluation of Compact Models for Thermal
Characterisation Using Dedicated Optimisation Software”, 15th
Annual IEEE Semiconductor Thermal Measurement and
Management Symposium, SEMI-THERM, p.1, 1999
[16] V. Sunde;Z. Bencic and Z. Jakopovic, “A temperature-
dependent electrothermal MOSFET model for calculating its
current loadability”, IEEE International Symposium on
Industrial Electronics, ISIE '99., Vol. 2 , pp. 579 – 583, 1999
[17] S. Schroder and R.W. De Doncker, “Physically based
models of high power semiconductors including transistor
thermal behavior”, IEEE 7th Workshop on Computers in Power
Electronics, COMPEL 2000, pp. 114 – 117, 2000
[18] V. Szekely and M. Rencz, “Thermal dynamics and the time
constant domain”, IEEE Trans. on Components and Packaging
Technologies, Vol. 23, no. 3, pp. 587 – 594, 2000
[19] M. Busani; R. Menozzi;M. Borgarino and F. Fantini,
“Dynamic thermal characterization and modeling of packaged
AlGaAs/GaAs HBTs”, IEEE Trans. on Components and
Packaging Technologies, Vol. 23, no. 2, pp. 352 – 359, 2000
[20] P.M. Igic; P.A. Mawby and M.S. Towers, “Physics-based
dynamic electro-thermal models of power bipolar devices (PiN
ThETA01/048 39