Professional Documents
Culture Documents
Venkat R. Gaurav
Credence Systems Corporation
Hillsboro, Oregon 97124
Phone: (503) 466-7823
Fax: (503) 466-7454
E-mail: venkat_gaurav@credence.com
KEY WORDS
Figure 2
Six fans, three across and two deep, were used to generate the
required airflow and overcome the system impedance.
Figure 4 shows a typical pin card used in the test head. The key
components at risk for lack of cooling were the timing and power
chips located near the card connector. These included eight TQFP
IC’s with an exposed pad (four on each side) dissipating close to
7.0 W each and sixteen timing chips, 208 pin BGA’s (eight on
each side) dissipating 2.5 W each. The target junction temperature
of the IC’s was not to exceed 95o C at the worst-case allowable
ambient air temperature of 30o C.
The modeling approach exploited the cyclic symmetry of the card Figure 5- Cyclic symmetric geometry with two pin cards
layout. Fig. 5 shows the geometry of this wedge-shaped domain. embedded within the flow volume. All power generating and flow
Two adjacent cards, the test head centerline, and the outer wall of obstructing chips are represented on the pin cards.
The model shown in fig. 5 was created using a commercially Later, mock-up pin cards were made and a flow bench was used to
available CAD package [1]. This CAD representation was measure actual system impedance, figure 7.
directly imported into the CFD solver [2]. With very few steps,
this model was then converted into a fully meshed finite element
CFD model.
This meshed model would be considered quite coarse by any
normal CFD modeling standards. For example, flow passages
through the domain were meshed with no regard to the number of
elements spanning across the flow channel. In most cases only one
element was used to define the edge-to-edge separation. Normally
CFD theory, with regard to FEA methods, would recommend a
minimum of 5 nodal points (four 4-node tet elements) from edge-
to-edge across the flow channel [3]. The CFD solver used in this
investigation has the unique ability to inflate or enhance the coarse
mesh near walls. Walls, in this context, refer to surfaces of the
FEA model where airflow encounters turbulence. The mesh
inflation creates a multi-layered mesh of six-node wedge elements
using the initial surface mesh as a template. The number of layers
and their thickness’ are user-defined. The end result allows the
user to create and use an incredibly coarse FE mesh (minimizing
computational pre-processing and data storage tasks) and yet Figure 7 - Measuring system impedance using mock-up pin cards
obtain high quality CFD results. and flow bench.
Thermal and Flow Modeling Flow resistances due to the enclosure side panels and fan plenum
were easily computed [5].
Only those chips, which generated over 1.0 W, were modeled in
sufficient detail using manufacturer’s geometry and material EXPERIMENTAL VALIDATION
specifications [4]. The thermal load on each chip was applied as a
volumetric heat generation. For the circuit board a standard 12- Airflow and temperature measurements were made on a fully
layer FR-4 board was assumed and then its thermal conductivity operational test head to validate the results of the CFD simulation.
was modified using the stack-up information.
Airflow Measurements
The air inlets and fan outlets were defined next as shown in Fig. 5.
A zero static pressure boundary condition was applied at the inlet. An array of airflow sensors was designed to fit across a pin
At the fan outlet, measured air velocity values as well as fan curve module (see figure 8) to measure the air velocity entering the card
data were applied in separate simulations. slots. Additionally another airflow sensor was placed between a
pair of pin cards such that it measured the air velocity between the
Determining System Impedance cards.
Static pressure losses through the complex flow channels inside
the test head could not be easily arrived at by calculation. First, an
estimate was made using a CFD model to size the fans, figure 6.
CONCLUSIONS
ACKNOWLEDGEMENTS
REFERENCES
1. Autodesk Inventor Release 3.0, Autodesk, Inc.
Figure 9 – Heat sink optimized for the 100 lead TQFP chips.
2. CFDesign 5.0, Blue Ridge Numerics, Inc
Additionally, the case temperatures of the timing chips as
predicted, ranged from an average of 45oC to 75oC and compared 3. Building Better Products with Finite Element Analysis,
closely with the measured results. Vince Adams and Abraham Askenazi, OnWord Press,
1999.
The airflow tests indicated consistent and uniform air velocities at
all locations in the test head as predicted. The average air 4. Advanced Electronic Packaging, Edited by William D.
velocity between pin cards past thermally significant Brown, IEEE Press, 1999.
components as predicted by simulation was within 1% of the
actual measured values. Figure 10 is a graph of the measured 5. Flow Resistance – A Design Guide for Engineers, Erwin
values. The average velocity predicted by simulation was 600 Fried and I.E. Idelchik, Taylor & Francis,
ft/min.