Documente Academic
Documente Profesional
Documente Cultură
A P P N O T E S SM
.
Correlation of Structure Functions between Analytical/Simulation
Model and T3Ster Testing Using FloEFD
Table of Contents
Introduction.................................................................................................................................................... 1
Sample Problem ............................................................................................................................................ 2
T3Ster Structure Function ............................................................................................................................. 3
Creation of Analytical/Simulation Structure Function .................................................................................... 3
List of steps for calculated structure function creation ............................................................................. 3
Creation of structure function for sample problem ................................................................................... 4
Numerical simulation ............................................................................................................................ 4
Discretize the solution domain ............................................................................................................. 5
Create a Table of Rth and Cth values .................................................................................................. 9
Plot the calculated cumulative structure function over the experimental curve to compare .............. 11
Introduction
In the process of creating structure functions through T3Ster tests, a frequent question is how to interpret
the structure functions (in particular, the cumulative structure function) to the actual part under test. This
is a question that has a unique answer for each part, but there is a general process that you can use to
understand the physical meaning of the cumulative structure function.
This AppNote is intended to teach you one method to understand your structure function. To do this, an
analytical or analytical/simulation based model is created and then compared against the test-based
structured function from T3Ster-Master. By overlaying the two plots, you can see what part of the thermal
path contributes to each part of the structure function.
Sample Problem
To demonstrate this method, a sample problem is set up and solved in this AppNote. Although this is
overall a simple problem, it has enough unique elements to demonstrate the key points in creating an
analytical/simulation model.
For this problem, an n-MOSFET device in a TO-220 case is used as the heat source. It is assembled dry
(no grease, thermal interface material, etc.) to an aluminum cold plate. The cold plate is cooled by a
copper tube embedded in into it, and water is the cooling liquid.
A simulation model can be created in many programs, such as FloTHERM, FloTHERM XT, or FloEFD.
For this AppNote, the simulation model is created in FloEFD (Standalone) with only the relevant parts and
sections of the problem (here, the die, die attach solder, copper heatsink, aluminum cold plate section,
and copper tube section). For the T3Ster test and the simulation, 4W of heating power is applied to the
die.
This structure function data was exported from T3Ster-Master and plotted in Microsoft Excel. This is done
so the calculated structure function can be overlaid on this plot for comparison.
At this point there is a completed test, but the question is, what do the various parts of the curve
correspond to in the actual hardware? With the various curves, slopes and inflections there is difficulty
being certain what each part of the curve means. This can be done by comparing the experimental curve
with some type of model that creates the structure function.
o Note it is possible in some extremely simple cases to perform this step with analytical
and hand calculations. However, due to heat spreading, one may not find the
appropriate thermal masses involved in the heat transfer path. A numerical simulation
model is strongly recommended to avoid this error.
o Care must be used when modeling thermal interfaces, as this directly affects the amount
of spreading. If there is no direct test data of a similar interface available, a reputable
source of data should be used (for example, the GE Heat Transfer and Fluid Flow Data
Discretize the domain/parts into isothermal and spreading regions based upon simulation results
o Parts with gradients need to be examined and the active spreading region used. These
regions will have the higher gradients and have most of the heat flow through them to the
next part in the heat transfer path, and little heat spread into the remaining areas of the
part (those areas are effectively not active in the heat path). Plots of heat flux are
frequently helpful in this regard.
o Parts can be found through the simulation data or from a combination of simulation data
and hand calculations
o Interfaces are found through tests, previously known data, or suitable handbooks (such
as the GE Heat Transfer and Fluid Flow Data Handbooks). These should be the same
interface values used in the simulation as described in the first step.
Plot the calculated cumulative structure function over the experimental curve to compare
o Exercise care for setting the origin of the plots (use a non-zero starting point so a log plot
may be used)
In this sample problem, both the silicon die and the solder die attach are relatively isothermal (see Fig 6 &
7). Both of these will be entries in the structure function table to be created in Excel.
The copper heat sink of the TO-220 requires a more careful evaluation. In this case, the part isnt so
isothermal as the die or solder, and the question is whether to consider it as a single isothermal part, or
only use part of it in the heat spreading consideration.
In cases where this decision is not straightforward, you can look at the heat flux leaving the surfaces of
the part to see how much heat is spreading. In Fig 9, there is a line across the bottom surface to divide
into two faces. The total 4W of heat flow will pass through both of these faces into the cold plate, but
perhaps most of the heat is in the larger section under the die. If most of the heat passes out this smaller
surface, you could consider the heat sink to be the smaller volume rather than the overall part. In this
sample problem, the reduced area passes 3.27W out of the 4W, or about 80%. It is a judgment call to
decide if this enough or not, and in this exercise the option of the full part will be used.
The next volume in the heat path is the aluminum cold plate. This part requires an even greater amount
of analysis to determine how much of the cold plate volume is active in the heat transfer path. Once
again, the part temperature contours are plotted and examined:
In Fig 11, the contour bands are chosen so the gradient can be seen better. The heated area under the
copper heat sink is surrounded by three zones a warmer zone to the left/right of the part, away from the
copper pipe, and two cooler zones, above and below the part, near the region of the copper pipe.
The question then is, which way is the heat flowing? Recognizing that heat flow is proportional to
temperature difference, the heat is flowing away from the TO-220 along the length of the copper pipe
geometry (that is, toward the cooler regions of the heat sink). Less heat would flow left/right away from
the copper pipe region. Another method would be to plot the heat flux contours and determine the
volume through inspection.
This is shown by making an volume slice into the cold plate and examining the heat leaving the volume.
Making an elliptical cut as shown in Fig 12:
For this volume, 3.62W of the 4W entering from the TO-220 passes through the elliptical volume and into
the copper tube about 90%, which is a reasonable number. Thus this volume accounts for some 90%
of the heat spreading in this problem within the cold plate.
If the elliptical slice was turned 90 and made smaller, the spreading capture portion would be far less:
Only 1.82W is entering the copper tube from this volume, and 2.18W is leaving the volume boundary
shown in blue in Fig 13.
Finally, the last two parts of the heat flow path the copper pipe and the water are made into their own
items in the discretization. The copper pipe has a total gradient of 1.3C, and the water volume is
similarly isothermal.
For the sample problem in the AppNote, the following table is constructed:
1/h k Cp
SECTION L (cm) (Kcm^2/W) (W/cmK) A (cm^2) Rth (K/W) Rth (g/cm^3) V (cm^3) [J/(gK)] Cth (J/K) Cth
Silicon 0.005 1.49 0.0625 0.0536912 0.0537013 2.33 0.001375 0.7046 0.002257362 0.002357362
Solder 0.007 0.55 0.0625 0.2036364 0.2573376 8.67 0.0004375 0.1785 0.000677073 0.003034435
From this table, you can see the six mass discretizations are made along with two thermal interfaces. A
non-zero start point is also made because the structure function is a log plot, which cannot start at zero.
One immediate question is how does one find the thermal resistance of a complex geometry part. That is
actually found simply through the simulation model. Rather than using an L/kA geometry and material
property calculation, the software can provide the T across the part and the heat flux is known. Thus it
can be calculated directly by finding the average temperature at the heat entry surface, and the average
temperature at the heat exit surface, and the known heat flux. For example, for the spreading volume in
the cold plate, the average entry surface temperature is 24.78C and the average exit surface
temperature is 23.98C.
This leads to a Rth calculation of (24.78 23.98)/3.622W or 0.221C/W. You can also use the 4W heat
value and obtain a slightly different thermal resistance of 0.2, but this will not be significant in the overall
plot of the structure function (and again, this is one of the approximations made when creating a structure
function with this method). This is the reason why finding the volumes containing large percentages of
the thermal spreading 90% in the case of this part are important. Greater errors will be introduced for
parts of the model using small percentages, such as 10% to 50%.
The table then needs to have two columns to sum the thermal resistances and capacitances these will
be the data points plotted to create the structure function.
Plot the calculated cumulative structure function over the experimental curve to compare
The analytical/simulation structure function may now be plotted. In this sample problem, the plot looks
like Fig 17:
Now from this plot and the table used to create it, each part of the structure function is identified as to
what the part or interface is and how it affects the shape of the curve. If the test curve from T3Ster-
Master is exported and plotted on top of this curve, one can now explain the various parts of the test
derived structure function as seen in Fig 18. As mentioned before, caution must be used to set the origin
of the plots correctly so a proper overlay is constructed.
One final issue to note is that the test derived plot is highly curved, while the analytical/simulation model
is not. This is caused by the very rough discretization made for the model (six volumes, two interfaces).
The T3Ster test evaluates the data over hundreds of points in making the Cauer diagram. The test
further provides more fidelity of the spreading effect and this affects the structure function curve. Last,
any test is subject to noise in the data, and the noise is amplified by the numerical algorithms that process
the data. This causes further smoothing of the structure function plot.
However, even with these differences between the curves, the coarse discretized model is more than
adequate to explain the overall shape of the T3Ster derived structure function.