Challenges of Heat Removal From Flex Circuit Technology
Jim Von Bank
General Dynamics, Bloomington, MN
ABSTRACT
Removing the heat from Chip Scale Packaging (CSP) implemented on flexible circuits presents a challenge for this
technology. Heat generation is a certainty for all ICs and especially with high performance logic ICs. Dynamic power
dissipation is proportional to CV2f and low power CMOS die operating at 1.2 to 3.3v at 100 Mhz and driving 50 pf loads will
nevertheless generate 7 mw to 54 mw of heat per output. If adequate conductive paths from these point heat sources are not
provided then the technology will never gain wide spread support and usage. This paper explores the traditional methods and
design options available to manage the thermal load. The paper bounds the parameters that are necessary to achieve
acceptable temperature rises and/or thermal impedances within the constraints of the materials selected.
Heat generated on printed circuits is removed by convection, conduction, and radiative processes. How do these play out with
CSP and flex circuits? The most popular flex circuit substrate materials are polyimide and polyester, that by themselves are
not good thermal conductors but rather are thermal insulators. Consequently, the job of heat conduction falls on the metallic
conductors. Just as in rigid printed circuits it will be necessary to install the equivalent of thermal vias for a given level of
power dissipation. Perhaps this type of conduit would be better named as thermal streets with radiating stubs. This paper will
parameterize the thermal problem of flex circuits with point heat sources.
1.0 Framing The Problem
The trend in electronics, and portable electronics in particular, is toward lower power dissipation. The dominant silicon
technology for integrated circuits (ICs) is CMOS. Heat generated in CMOS is composed of a static quiescent factor
(Vdd*I_quiescent) and a dynamic factor (Cpd*Vdd2*f). To keep the power dissipation low the operating voltages have been
dropping, but at the same time, in many chips, the operating frequency is increasing, so that power dissipation must still be
carefully managed. Even a few milliwatts of heat can turn into a reliability issue if no thermal path is provided. For purposes
of bounding the thermal problem herein, ICs generating anywhere from 10 mw to 1000 mw will be considered part of the
design space.
The thermal conductivity of polyimide is 0.1 to 0.4 W/m-K making it essentially a thermal insulator. In contrast, the metallic
conductors of copper and silver are excellent thermal conductors having thermal conductivities of 429, and 401 W/m-K
respectively. Heat removal from the heat sources (ICs on a flex circuit) can occur by convection if there is moving air
surrounding the die. Heat removal can also occur by conduction through the metallic conductors if there are heat sinks at the
end of the path. Additionally, heat removal can occur by radiation and varies as the 4th power of the temperature difference
between heat generator and radiative receptor.
The types of flex circuit structures evaluated in this paper are: a) single sided flex circuits, b) double-sided flex circuits, or c)
multi-layer flex circuits. Within these generalized structures there are a plethora of possible configurations that may or may
not include thermal vias to enhance the heat removal from the sources. Table 1 bounds the design space for this analysis.
Examine the columns of table 1. Flowover is a heat exchange mechanism where there is a fluid moving at a given flow rate.
A flex circuit inside a vented enclosure that relies upon thermal conduction and natural convection as the heat exchange
mechanism is denoted by the heading ‘Vented NC’. When there is no vent and thermal conduction is augmented by radiative
heat exchange the column heading ‘NV with radiation’ is used in table 1. These three heat exchange mechanisms are
evaluated herein.
Table 1 - Pared Down Design Space
Single-Sided Flex
Flow-over
Vented
NC
Copper
Copper with
thermal vias
Silver
Silver with
thermal vias
NV with
radiation
Double-Sided Flex
Flow-over Vented
NC
NV with
radiation
Multi-Layer Flex
FlowVented
over
NC
NV with
radiation
2.0 Analysis Tool
There are many thermal analysis tools available today ranging from detailed finite element based tools to simple hand
analysis. Analyzes presented here are based on the CALCE_PWA thermal analysis tool. This tool is based on control volume
theory and uses a finite difference approach. This tool was developed for the purpose of analyzing temperature distributions
within printed wiring boards that are cooled by various heat exchanger designs. The thermal analysis software will
automatically discretize the board based on the number of layers and the grid size specified for a planar surface to create a
three dimensional matrix of cubes. Heat input into the top and bottom surfaces are based on the grid size, component power
dissipation, as well as component placement and size.
Three kinds of heat exchange processes are used in this paper; A) fluid flowover , B) conduction with natural convection
within a vented enclosure, C) conduction within an unvented enclosure with radiative heat losses to the enclosing structure.
In heat exchange process ‘A’, the fluid (air) flow rate in kg/s is specified, as well as the space surrounding the flex circuit. A
personal computer would be a representative example for process ‘A’. In heat exchange process ‘B”, conduction within a
vented enclosure, there is no forced flow rate but the vent allows the internal atmosphere to be the same as the external
atmosphere. Natural convection is influenced by the space surrounding the flex circuit. An example could be a CD player.
Heat exchange process ‘C’, is similar to ‘B’ in the sense that conduction is a heat exchange mechanism used but is dissimilar
to ‘B’ in the sense that the unit is sealed or not vented so that there is no natural convection. Furthermore heat exchange
process ‘C’ also considers radiative heat transfer by specifying the emissivity of the flex circuit and the enclosure. An
example could be a cell phone.
3.0 Baseline Analysis
To illustrate the effects of the above three heat exchange mechanisms on a flex circuit, consider a single layer flex circuit 2
inches by 4 inches in size with a 160 pin CSP FPGA and two 28 pin CSP memory devices as shown in figure 1. The design
criteria for using a flex circuit is the lead compliance provided by the polyimide and thus the longer lifetime until fatigue
failures.
U1
U2
U3
Polyimide dielectric layer
Patterned conductor layer
Figure 1 - Simple 2 by 4 inch Flex Having three components dissipating 1.2 watts total.
Table 2 illustrates a baseline of case temperatures for the three components, in figure 1, mounted on the single sided flex
circuit and dissipating 1.2 watts total as allocated. Three heat exchange methods are identified along with the critical
parameter for that exchange method.
One half inch distance above
and below flex circuit to
enclosure
U1 (0.6 watts)
U2 (0.3 watts)
U3 (0.3 watts)
Table 2 - Case Temperature Baseline
Flowover
Vented Enclosure
Flow rate = 0.002 kg/s
External Temp = 20 C
40C boundary condition
Conduction on 2 sides
46
57
51
64
52
64
Not Vented and Radiative
Heat Transfer
Emissivity = .5
64
69
74
The results in Table 2 are point estimates and are not useful in and of themselves. However, a sensitivity analysis can provide
insight into the limitations of the technology. In the following three sections the sensitivity of case temperatures for the
components mounted on flex circuitry as in figure 1 is examined.
3.1 Flow-Over Heat Transfer
Consider, for example, the single sided flex circuit baseline of figure 1 with varying flow rates and varying power in U1. The
sensitivity results provided in figure 2 imply that at least 0.0005 kg/s of air must flow in order to keep temperature rises to a
manageable level for power dissipations up to 1 watt. In general, when a moving fluid is involved, a minimum flow rate is
needed to get under the ‘knee’ and after that point diminishing returns occur with further increases in flow rate. A similar
shaped curve will result for other configurations of flex geometry and component assignments.
U1 Temperature Rise versus Intake Air Flow Rate
Indexed By U1 Power Dissipation
Temperature Rise (C)
80
70
60
50
U1(dT) @ 0.6 W
U1(dT) @ 0.8 W
U1(dT) @ 1.0 W
40
30
20
10
0
0.0001
0.0006
0.0011
0.0016
0.0021
0.0026
Intake Air Flow Rate (kg/s)
Figure 2 - Temperature Rise of U1 for Varying Flow Rates and Power Dissipation
3.2 Vented Enclosures
Consider next the vented enclosure where heat exchange occurs by conduction to the enclosure as well as natural convection
through the vent. Numerous examples exist among consumer electronics. Figure 3 shows that the case temperature rise for
U1 is, of course, still dependent upon power dissipation but the temperature rise is not very sensitive to the ambient
temperature. The absolute temperature does increase with the ambient but not the temperature rise for the baseline spacing.
The one-half inch of space on both sides of the flex circuit is sufficient so that natural convection, in combination with
conduction, carries the heat. One may ask which mechanism is carrying away most of the heat? For figure 3, conduction
occurs along the whole perimeter of the flex circuit. In contrast, if two of the four sides are insulated from the enclosure, as
illustrated in the planar thermal model of figure 4, then thermal conduction only occurs from the two short sides. The results
shown in figure 5 give a few additional degrees of temperature rise to U1. By itself this design change seems acceptable.
However, when the two long sides of the flex are insulated and provide no heat transfer path then the heat from U1 has a
greater impact on U2 and U3. Figure 6 compares the case temperatures of U2 for the four sided versus two sided conduction
designs. With the long side of the flex insulated, U2 operates 10 degrees warmer. In general since polyimide is not inherently
a good thermal conductor, providing more conductive paths is desirable. To evaluate the impact of natural convection, figure
7 shows the sensitivity of U1 temperature rise to the space above and below the flex circuit. When the spacing falls to less
than 8 mm then the temperature rise increases dramatically. For the baseline design, it is concluded that natural convection is
removing more heat than conduction as long as the distances are large enough to stay below the ‘knee’. These curve shapes
are typical of the sensitivity obtained for other power dissipations, and other flex circuit geometries. Therefore with a vented
enclosure the combination of conduction and natural convection can manage the heat load provided sufficient convection
space and conduction perimeter is provided.
Temperature
Rise (C)
U1Temperature Rise Above Ambient
for Vented with Natural Convection
60
50
40
30
20
U1 @0.6 W
U1 @0.8 W
U1 @1.0 W
20
25
30
35
40
Ambient Temperature (C)
Figure 3 - Flex Circuit in a Vented Enclosure with Natural Convection
Figure 4 - Grid Layout With Two Insulated Sides
U1 Temperature Rise Above Ambient for Vented
with Natural Convection and Insulated
U1 Temperature Rise (C)
60
55
50
45
40
U1 @0.6 W
U1 @0.8 W
35
U1 @1.0 W
30
25
20
15
25
35
45
Ambient Temperature (C)
Figure 5 - Insulating Two long Sides for Vented with Natural Convection
U2 Case Temperature
(C)
U2 Case Temperature Versus Ambient
Temperature
100
U2 Case,
Insulated,
U1->0.6W
80
60
U2 Case,
U1->0.6W
40
20
0
20
25
30
35
40
Ambient Temperature (C)
Figure 6 - Insulating Two Sides of the Flex Creates Hotter Case Temperatures
Temperature Rise (C)
U1 Temperature Rise Vs Spacing Above and
Below Flex, Ambient = 20C
200
150
U1 @0.6W
U1 @0.8W
100
U1 @1.0W
50
0
2
4
6
8
10
12
14
Spacing (mm)
Figure 7 - Sensitivity of Temperature Rise to Space Allocated For Natural Convection.
3.3 Non Vented With Radiative Heat Transfer
In contrast, in a sealed or unvented environment, the internal fluid pressure does have a measurable effect on temperature
rise. A higher internal pressure implies greater density and a better heat transfer medium. Since it is difficult to maintain an
internal pressure other than atmospheric, due to leakage processes, it will be further assumed here that heat exchange process
‘C’ always has internal atmospheric pressure.
The emissivity is a significant factor for radiative cooling. The analysis tool requires board and enclosure emissivities. It is
assumed here that the inside of the enclosure is black anodized aluminum which has an emissivity of 0.82. Black plastic ICs
may assume an emissivity of 0.85 whereas 1 mil polyimide may have an emissivity of 0.67. Since the software tool just asks
for board emissivity, 0.75 will be used as an average considering the black plastic ICs and the polyimide.
Figure 8 provides the case temperatures for U1 for the non vented enclosure with radiative heat transfer and figure 9 takes the
same data and provides temperature rise above ambient. The rather interesting result is that figure 9 shows a decreasing
temperature rise as ambient temperature increases. Presumably this is due to the better efficiency of the radiator at higher
temperatures. According to the Stepfan Boltzman law the amount of heat radiated is proportional to the fourth power of the
absolute temperature of the flex-IC combination compared to the enclosure..
U1 Case Temperature
(C)
U1 Case Temperature for Unvented Enclosure
with Radiative Transfer
100
90
80
U1 @0.6W
70
U1 @0.8W
60
U1 @1.0W
50
40
20
25
30
35
40
Ambient Temperature (C)
Figure 8 - U1 Package Temperatures Increase with Ambient and with Power Dissipation
U1 Temperature Rise
Above Ambient (C)
U1 Temperature Rise Versus Ambient
Temperature
60
55
50
U1 @0.6W
45
U1 @0.8W
40
U1 @1.0W
35
30
20
25
30
35
40
Ambient Temperature (C)
Figure 9 - U1 Package Temperature Rise Above Ambient Decreases As Ambient Increases
All three heat transfer mechanisms are occurring under the ‘NV with radiation’ column header; conduction to the enclosure,
convection even though the air is still, and radiation. Another sensitivity that is explored is the distance between the flex
circuit and the enclosure. Figure 10 shows that as the distance separating the flex circuit from the enclosure decreases the
temperatures also decrease. Taken to the limit, U1’s case temperature would be the same as the 40 C enclosure temperature.
U1 Case Temperature
(C)
U1 Case Temperature Versus Flex Spacing To
40C Ambient
100
90
80
U1 @0.6 W
70
U1 @0.8 W
60
U1 @1 W
50
40
0
5
10
15
20
25
Spacing (mm)
Figure 10 - Sensitivity To Spacing Above and Below Flex Circuit
4.0 Thermal Vias and Copper Stubs
All the sensitivities explored thus far were for the simple single sided flex circuit in figure 1, baselined with the hypothetical
U1, U2, and U3. Several heat exchange mechanisms were evaluated. In reality for the higher pin count ICs multiple layers
are needed to make all the connections, so the baseline example is not realizable, but that doesn’t negate the thermal insights
for the various heat exchange mechanisms. A multilayer flex circuit permits making all the interconnects needed and also
provides more copper for heat transfer. Of course the tradeoff is that each successive layer makes the flex circuit less flexible
and the expected advantages of polyimide (lower CTE) soon diminish. Since it is desirable to maintain as few layers as
possible, the thermal challenge remains; remove the heat given the poor thermal conductivity of polyimide.
This section will first explore the impact of multiple layers of metallization and will only use the heat exchange mechanisms
of conduction within a vented enclosure with natural convection. The thicknesses used for the metallization and dielectric
are:
Metallization thickness - 0.0125 mm, 10% of the layer is patterned with copper
Dielectric thickness – 0.01778 mm, (polyimide E, or A, or Q are options in CALCE_PWA)
Table 3 - Comparison of Multilayer Flex Thermal Performance for Vented With Natural Convection Heat Exchange
Mechanism;
Conditions are Ambient =20C, 4 mm Spacing to Enclosure, Conduction from Two Sides only
U1 Case Temp @ 0.6W
U2 Case Temp @ 0.3W
U3 Case Temp @ 0.3W
Single metal layer
83
123
93
Two metal layers
70
86
71
Three metal layers
61
72
61
Table 3 shows that adding additional layers improves the thermal conductivity as expected. Selecting among the various
polyimide dielectrics does not affect thermal performance. The CALCE_PWA tool can provide thermal maps of the inner
layer temperatures as shown in figure 11.
Figure 11 - Flex Circuit Inner Layer Thermal Map for U1, U2, U3 Baseline
Since adding layers makes the circuit less flexible, the impact of adding thermal vias and copper conduits to an existing
interconnect layer to manage heat is an option. Assume that thermal vias, from a component like U1, are plumbed to a metal
trace that in turn connects to a heat sink as illustrated in figure 12. The amount of heat that a single metal trace can carry to
maintain a given temperature gradient is a function of the thermal conductivity, the cross sectional area and the path length to
the heat sink. Figure 13 shows the heat that can be carried away for various trace geometries. Short and wide traces are
preferred but wide traces aren’t an option around a congested interconnect area. Actually the thermal performance can be
better than illustrated in figure 13 since once past a congested area the traces can be made wider. Additionally, along the
length of the trace, heat can be transferred to the dielectric and it can radiate outward.
Figure 14 shows the thermal analysis grid wherein a short thermal conduit transfers heat from U1 to a nearby metal island
buried within the flex. The software tool can be used to evaluate the impact on the multilayer flex baseline in figure 12 when
thermal vias and thermal conduits are used to enhance the thermal conductivity. As anticipated, the routing for a high pin
count IC will constrain the options for implementing thermal conduits.
U1
U2
U3
Polyimide dielectric layers
Patterned conductor layers
Figure 12 - Multilayered Flex Circuit With Thermal Vias and Conduits
0.06
0.054
0.05
Watts ( Path_Length , 0.6⋅ mm)
0.04
Watts ( Path_Length , 1.2⋅ mm) 0.03
Watts ( Path_Length , 1.8⋅ mm)
0.02
0.01
−3
1.8×10
0
0.02
10⋅ mm
0.04
0.06
Path_Length
0.08
100⋅ mm
Figure 13 - Watts Removed By Various Width ½ oz copper Traces for dT=60C
Summary
Since thermal conduction is not good for polyimide, three types of heat exchange mechanisms were studied for removing
heat from a flex circuit. CALCE_PWA was the tool used to evaluate these mechanisms. For conduction enhanced with
moving air convection, a minimum flow rate can be identified. For a vented enclosure with natural convection a minimum
space surrounding the flex can be defined. For an enclosure without a vent, radiative heat transfer can augment conduction
heat transfer. The traditional approach of additional ground planes for thermal management, will give way to thermal vias
and thermal conduits implemented in the metalization layers of flex circuits.