TMAPPER: Ultrafast Self-Adaptive Thermal Modeling
|
|
Adaptive
Thermal Modeling
Technique |
|
A novel, self-adaptive, thermal modeling
technique has been developed to handle highly-complicated problems
whose
geometric and temporal features can vary by several orders of
magnitude. These
types of problems are very common in the electronics and
micro-electro-mechanical systems areas. |
|
- Two orders of magnitude
faster simulation than with traditional approaches
- Independent
of user expertise in meshing
- Independent
of materials, geometric features, and locations of heat sources
- Solutions are obtained to
the user’s specified error criterion
- Eliminates need for
time-consuming and expensive convergence studies
- Enables
concurrent electro-thermal design of high-performance integrated
circuits
|
|
The novel, patented* approach developed at SMU during a GOALI grant
begins by solving the corresponding
steady-state problem by the use of a grid nesting technique. The
nesting technique defines a template that is then used to solve the
transient problem in a multiple grid fashion. The advantage to solving
with multiple grids in time is that the majority of the problem domain
can be resolved in time at lower nest levels (with a coarser grid),
while the finer grid resolution is reserved for the parts of the
problem that demand the finer resolutions in space and time. Since the
physical dimensions of the various materials
used in modeling high performance electronic devices vary greatly, a
uniform mesh that resolves all of the details in three dimensions
results in a prohibitively large computational grid. A common method
for dealing with dimensional variation is to skew the mesh and
concentrate more grid points in areas where higher resolutions are
needed. The shortcoming of using a biased-mesh approach to resolve the
geometry is that the problem geometry, and not the temperature
gradients, will end up dictating the meshing.
|
|
The meshing strategy used
in the development of the current novel technique was set on ensuring
that the method is:
- automatic and adaptive,
- independent of user expertise, and
- independent of materials (including air), geometry
features, and potential locations of sources
An additional objective in defining a meshing approach was to eliminate
as much "engineering judgment" as possible in deciding how much or how
little of the problem geometry to include. The new approach allows for
the addition of the difficult problem geometry associated with top
surface features of an IC over a large area, and then letting the error
prediction technique decide if the regions need further refinement. The
strength (and novelty) of the method is that it uses effective thermal
properties that are consistent with the local grid spacing at the
particular grid level in use. As a result, dealing with air, embedded
vias, and ultra-thin multi-layered structures requires no special
treatment.
Details
of communication microwave module
The transient modeling of high
performance ICs (e.g., MMICs) is required as part of the design
process. Temperature predictions are necessary both for reliability
analysis and predictions of electrical performance. The current
limitation to integrating thermal simulation results into the
electrical design is the excessive computational requirements
associated with transient simulations, especially that, if a simulation
is to be realistic, a detailed geometry, as well as temperature
dependent thermal properties must be included. For example, the
communication microwave module shown in the figure above illustrates
the typical five orders of magnitude variations in geometric scales
that exist in transitioning from modules to a single FET junction. To
further complicate matters, the time scales for describing temperature
changes at the channel level are typically in fractions of a
microsecond (µs) while the overall module can be expected to have
time scales on the order of seconds.
Schematic
of a representative complex IC problem
* U.S.
Patent No. 6,064,810 issued May 16, 2000.
International patents pending.
|
|
Complex 3D Sample Problem |
|
A
complex, three-dimensional
problem that includes top surface metalization, multiple materials, and
temperature-dependent properties, is presented here in order to
demonstrate the
significant computational speed gain afforded by the new transient
multiple-grid approach. The features of the sample problem shown in the
figure
above include three heat sources whose characteristic dimension is four
orders
of magnitude smaller than the length of the substrate. For the purpose
of
establishing a realistic comparison of solution speeds, the problem was
also
modeled by the use of TEMP (a state-of-the-art Raytheon proprietary
general-purpose heat transfer code). The problem was evaluated for both
constant and variable properties, and under both pulsed and continuous
operation. The table below lists both the raw data and the speedup
factor.
However, the results do not factor in the two days of effort required
(on the
part of a highly experienced analyst) to mesh the problem for the
conventional
method.
Simulation
Time Comparison
|
Simulated Condition
|
Conventional
Method
(Minutes)
|
Nesting Method
(Minutes)
|
Speedup
Factor
|
|
Steady-State;
linear
|
4.4
|
0.8
|
5.6
|
|
Steady-State;
K(T)
|
27.2
|
1.6
|
16.7
|
|
Transient;
linear
|
147.8
|
1.9
|
79.2
|
|
Transient;
K(T)
|
459.6
|
2.5
|
186.0
|
|
Pulsed;
K(T)
|
5,276.0
|
34.0
|
155.0
|
The results indicate that the advantages of
the novel approach
are significantly higher for transient nonlinear problems. For these
more computationally
demanding problems, the new SMU approach affords a two order of
magnitude
increase in computational speed over conventional methods. The benefits
are
even more significant if one considers the facts that, in conventional
methods,
(i) significant effort must be expended by an experienced analyst on
mesh
generation and optimization; (ii) several mesh iterations are normally
required
to ensure that features of interest are well resolved; and (iii)
several
solutions with increasing resolution are required to ensure that grid
convergence
is achieved.
When the most complex case is solved on the newest
personal computers, the simulation time is further reduced into the
minutes.
This level of CPU time is short enough to allow the thermal design to
be done
concurrently with the electronic design process, which is a desirable
approach that
has nonetheless been avoided because it has until now been impossible
or at
best impractical.
|
|
TMapper
|
|
A software
package called TMapper
has been created by NETSL to facilitate the use of the new
self-adaptive, thermal modeling computer code by anybody through the
Internet.
Click on
the button below to download the TMapper. If you have a slow connection
you must be patient for the 2.5MB ZIP file to download. After
downloading and unzipping the files click on the TMaper.exe. For
instrunctions on how to use the TMapper please refer to the section TMapper in R&D
Services.
(ZIP file, 2.5 MB)
We provide TMapper for your use, free
of charge, and hope that it would be of significant help in your
thermal analysis work. The condition for this use, however, is
that you acknowledge our work
and code in your reports and other forms
of publication, both internal and external to your organization (see
Acknowledgments below).
In addition, we request that you provide us with reference information
on the problem that you have used TMapper to solve. Your
feedback, either in the form of a testimonial or suggestion, is
important to us as we continue to improve our ultra-fast,
self-adaptive, thermal modeling capabilities.
The version of TMapper
that is presently provided on our
web site is a limited version of our more sophisticated patented
system. The limited version is for steady-state problems and can
accommodate a maximum of ten layers of materials. The full
version has no limitations in terms of number of materials,
adjacencies, or geometric features. It is fully automatic and
adaptive, independent of user expertise, and provides, for the first
time, the ability to perform parametric analyses on fully nonlinear,
pulsed devices. The full version also provides the ability to
extract the complete set of simulation data for visualization in a
commercial package, such as Tecplot.
We hope that you
will enjoy the use of this truly
amazing capability. Let us know if we can be of assistance in your
computations or in your measurements of stacked layers of nanoscale
structures.
Disclaimer
of warranty and limitation of liability
In no event shall NETSL be liable for
any direct, indirect or
consequential damages or any damages whatsoever (including but not
limited to loss of use, data, or profits) with respect to, arising out
of, in connection with, or related to the use of TMapper.
Acknowledgments
"TMapper: A web-based, ultra-fast,
self-adaptive, transient thermal
simulation method for complex nanoscale electronic devices," Peter E.
Raad and Pavel L. Komarov, http://engr.smu.edu/netsl/tmapper.html,
2002.
Tmapper is based on "System and Method
for Predicting the Behavior of a
Component," Peter E. Raad, James S. Wilson, and Donald C. Price, U.S.
Patent No. 6,064,810, Issued May 16, 2000. The method and system
are described in "Adaptive Modeling of the Transients of Sub-Micron
Integrated Circuits," Peter E. Raad, James S. Wilson, and Donald C.
Price, IEEE Transactions on Components, Packaging, and Manufacturing
Technology - Part A, Vol. 21, No. 3, Sep. 1998.
|
|
