ME358 Heat Transfer in Microdevices

Term Paper for ME358, Heat Transfer in Microdevices

Feasibility study for a novel temperature regulator with variable thermal resistance
Gordon Kariya, Hyeun-su Kim, and Masafumi Nakamura
Department of Mechanical Engineering, Stanford University

Abstract
A feasibility study for a novel passive temperature regulator with variable thermal resistance is presented to give a solution
to the problem of maintaining constant device temperature with a low power budget while being exposed to large variations in
ambient temperature. The approach for the passive variable resistance studied in this paper uses an array of bimorph cantilevers
which is actuated by ambient temperature. For this, bimorph cantilever mechanics is investigated intensively. When a bimorph
cantilever contacts the device surface, there can be large contact resistance unless enough contact pressure is applied. To avoid
this problem, the relationship between contact resistance and pressure is studied. With required resolution, several relationships
between design parameters are made. By combining design constraints, cantilever mechanics, and thermal resistance modeling,
the optimal design parameter values for several material combinations are iteratively determined.

1. Introduction designed with varying lengths to vary the amount of heat

Many electric devices need to be kept in a stable conduction.
temperature condition for their reliability. To achieve that,
usually heaters and coolers are used. However, there are 2. Thermal resistance of each cantilever
cases in which power consumption by the heater and cooler
This variable thermal resistance temperature regulator is
can be too much for the limited power budget of the device.
studied to keep a given micro device at its required
With a low power budget, if ambient temperature varies over
operating temperature. The device needs to be operated at
a large range, it is not easy to keep the device at a constant
65°C and it dissipates 40mW heat. The ambient temperature
temperature.
is assumed to vary from 0 to 60°C.
In the MEMS community, numerous heat sinking methods
are investigated because most devices dissipate heat and ΔT Tdevice  Tambient
Rth   (1)
should not be overheated. However, many of these methods q qdevice  qheater
are not applicable for this application not only because they
are solely oriented for heat removal and not developed for From the equation (1), qdevice is constant and Tdevice needs
use in combination with heaters. Peltier heater/ cooler pair is to be kept constant. If we use a heater, ambient temperature
a possible candidate; however, it consumes too much power variation can be compensated by heating power; however, a
and it also has the isolation problem. To achieve variable huge amount of power is needed. For this reason we explore
thermal resistance, several methods have been studied by passive thermal control as an alternative to active methods
others. One of them is the phase change method. Changing such as heating and refrigeration.
phase from austenite phase to martensite phase or vice versa
will change heat capacitance, and consequently the amount
of heat transfer changes. This method is again not applicable
for this device because phase change requires high
temperature while the maximum ambient temperature for this
device is about 333K.
We present a novel passive temperature to give a solution
to the problem of maintaining constant device temperature
with a low power budget while being exposed to large
variations in ambient temperature. This temperature regulator
will mount a device on it and keep the device at a given
temperature in a vacuum sealing. The device and the
regulator will be supported by very thin and tall columns to
give sufficient insulation at low ambient temperatures within
the package. To get variable resistance, we employed an
array of bimorph cantilevers. As ambient temperature rises,
bimorph actuators on the inner surface of the package start to
contact to the device, creating additional heat paths to carry
away heat from the device. The array of bimorphs is

1
 ME358 Heat Transfer in Microdevices
from cantilever itself are correlated. We can find a solution
with iterative loops for this inter-related design parameters
device ASIC in steady state in one temperature. Before this system goes
Supporting
Si to equilibrium condition at certain temperature, contact
columns
resistances keep varying and there must be some tolerance
in design. Calculation showed this transient heat transfer
before its equilibrium condition is not significant because
contact resistance before equilibrium condition is
exponentially big and there must be very small heat transfer
before its equilibrium condition.

4. Cantilever Mechanics
For the cantilever beam equations, the variables are as in
Figure 2. The equations for the force and deflection of the
cantilevers are derived mostly from bi-metallic beam
calculations by Chu[1], et al and Timoshenko[2]. The force
exerted by a cantilever is given as equation (5).
Figure.1. Basic configurations of the array of bimorphs.

Because ambient temperature varies from 0 to 60°C, 60 b2

cantilevers are required for a 1°C temperature control
resolution. At 0°C, the initial thermal resistance when no
cantilevers are contacting is as equation (2). t2

Tdevice  Tambient , min 65  (0)

R0    1625 [ K / W ] E2, 2 t1
q device 40 10 3
(2) E1, 1
This is achieved with vacuum and support structures.
When ambient temperature is at k°C, thermal resistance of
the whole system is Rk and the thermal resistance of one b1
cantilever which is activated at k°C is r,k. Then, Rk can be
E1, 1, b1, t1: the Young’s modulus, thermal expansion
expressed as (3), coefficient, width, and thickness
1 1 respectively of the bottom half of the
Rk   cantilever
1 1k
1 1 (3)
  E2, 2, b2, t2: the Young’s modulus, thermal expansion
R0 i 1 ri Rk 1 rk coefficient, width, and thickness
Where R represents total resistance and r represents respectively of the upper half of the
individual cantilever resistance. Consequently, from the cantilever
following equation (4), the required thermal resistance of
each cantilever is calculated. Figure 2. Variables for beam theory
1
rk 
1

1 (4) 3( EI )* d eff
Rk Rk 1 Feq  (5)
L3
Each cantilever's required thermal resistance is a constraint where deff is the deflection needed to “uncurl” the
for each cantilever’s geometry. cantilever from a curved state, L is the length, and (EI)* is
the flexural rigidity of the composite cantilever. This force
represents the amount of force necessary to “uncurl” the
3. Design Parameters and constraints
cantilever by an amount deff.
The basic constraints for this design study are mostly from
The flexural rigidity of the composite beam is given by
the required cantilever resistance to balance out ambient
equation (6).
temperature changing. The resolution we want to get also
plays a role as a constraint. Contact resistance is a function of  b t3  t  
2
 b t3  t 
temperature, force, materials selection, roughness, and ( EI )*  E1  1 1  b1t1  y  1    E2  2 2  b2t 2  t1  2  y 
 12  2    12  2 
geometry. Force is again correlated with geometry and  
temperature. This means contact resistance and resistance (6)

2
 ME358 Heat Transfer in Microdevices
where y is the location of the neutral axis of the
1
composite beam, given by equation (7).


E b t 2  E2b2  t1  t 2   t12
2
  4rc F  2

a   w 
y 1 11
2 E1b1t1  E2b2t 2  (12)

(7)
 E * 
The deflection needed to “uncurl” the cantilever is given
by equation (8).  
L2 Therefore the apparent contact area is:
d eff   offset (8)
2 a
Aa  rc w sin 1 (13)
where k is the curvature of the beam due to the change in rc
temperature, and offset is the initial distance between the
cantilever and the device.
SHAPE \* MERGEFORMAT
The curvature k of the beam due to the change in
temperature is given by equation (9).
6b1b2 E1 E2t1t 2  t1  t 2   2  1  T

b E t   b E t 
2 2
1 11 2
2 2
2 2  2b1b2 E1 E2t1t2  2t12  3t1t 2  2t 22 
2a
Figure 3. The schematic of contact region
(9)
where T is the change in temperature.
In practice, however, two bodies will actually touch only
at a few discrete spots. Thus, the thermal contact
5. Contact resistance conductance is dependant upon the characteristics of the
This section considers the resistance to the flow of heat surfaces, the mechanical pressure between them, and
between two bodies in contact in a vacuum. We mainly cited whether there is any conducting fluid in the interstices of
literature by M.G. Cooper, B.B. Mikic and M. M. the interface.
Yovanovich [3-5]. Their theory is also cited in several papers Although conducting fluid in the interstices of the
[6-8]. interface is negligible in our application, we need to take
For this application, the deformation must not be the account of surface roughness, so the real contact area is
plastic but elastic because the operation has to be repeatable. different from apparent contact area. The ratio of the real
According to the literature [3], when the parameter   3, the and apparent area is shown as:
deformation will be predominantly elastic.  is given by Ar P
equation (10). This condition gave the value of tan as 0.04,  a 1 (14)
Aa H
where tan is the mean of the absolute slope of a surface
profile. Because it is denoted that an average value of tan This expression also indicates that elastic deformation
for bead blasted surfaces is 0.1, it is reasonable to assume cannot exist if the contact pressure is larger than the
that tan is 0.04 for silicon wafer surface. hardness. This would limit the maximum value of applied
force. Therefore the force cannot be arbitrarily large, despite
H
  (10) the fact that larger force would create the smaller contact
(E*) tan  resistance, as discussed later.
The thermal contact conductance is defined follows: Mikic proposed the thermal conductance in elastic
deformation as:
Q
h
Aa (11)
he 
1 Z Q 1  2 Pe   
T
Where, we have approximated the apparent contact area as
2 2 1  Pe 
1.5
 (15)

line contact; e.g., contact between cylindrical bodies. Using And following simple expression approximates the plot of
Hertz theory, the width of the apparent contact area is given he :
as:
k tan 
P 2
he  1.55 ) 0.94 [W/Km2]
(
 E * tan 
(16)
Because we would like to obtain rcontact [K/W], he has to be
multiplied by apparent area. Hence,

3
 ME358 Heat Transfer in Microdevices
1
rcontact  [K/W] (17) width / thickness aspect ratio
he Aa
2.5

Finally, rcontact is expressed as function of force. 2

width / thickness
1
  F 
0.94

  2   1.5
 k tan   Aa  
rcontact  1.55 Aa 
  E tan  
   
   
  1
1
  F 
0.94

  2  
   1
  
  
4 Fr  2
   0.5 
   w  E   
  2 rw sin 1   
 
   r    1

 4 Fr  2  
       
 k tan   
 
  1
  wE  0 
 1.55 2 r w sin  280
 290 300 310 320 330
  E tan    r 
  
      th
 
 
 i cantilever, denoted by Tactivation, i
 
   
   
   


  Figure


4 Width/thickness aspect ratio
   

(18)
As can be seen in Figure 4, the ratio between width and
It can be seen that rcontact is a very weak function of Aa. thickness approaches 1:10 for some cantilevers. These
Thus, this can be approximated by the following relation: cantilevers have widths of 2-3µm, thicknesses of 22µm and
rcontact  F 1 (19) lengths of ~600µm, so it is a cause for concern from a
stability and fabrication standpoint. In second stage,
Since force is a function of temperature, the force terms in contact resistance will also be included in the design of the
rcontact can be replaced by temperature and rcontact is expressed cantilevers, and the cantilever width is expected to be larger
solely as a function of temperature. in order to compensate for the added resistance.

Second stage – consideration of contact resistance:

6. Design consideration approaches In order to design the cantilever, the contact resistance
First stage: must also be considered. For the lengths obtained from the
first stage, the width must be recalculated to compensate for
The first stage of this design study was to determine the
this additional resistance. Calculations have been done
dimensions of the cantilevers while assuming contact
iteratively for both temperature and width, while satisfying
resistance to be negligible. The cantilever was assumed to be
resistance requirements.
isothermal.
For example, starting with an initial guess for the edge
The length of each cantilever was found by calculating the temperature of the cantilever, the average temperature of the
length required to have the cantilever just touch the device at cantilever is calculated with equation (22). Using this
the cantilever’s activation temperature. This was done by average temperature, the width is iteratively solved for
finding L such that: using the secant method until the total resistance of the
3( EI )* d eff cantilever converges to the resistance requirement. This
Feq  0 (20) yields a value for the contact resistance, from which the
L3 new guess for the edge temperature is calculated with
Note that the calculation of the cantilever lengths is equation (23). These iterations over edge temperature and
independent resistance, and based solely on the mechanical width are repeated until equilibrium is reached to a certain
aspects of curvature and offset. For this reason the lengths tolerance.
calculated in this manner are also applicable when contact
resistance is taken into consideration in second stage of this qdevice
design study. The width of each cantilever was calculated rcontact
from the following equation (21).
length rcantileve
width  (21)
Rcantilever  k1t1  k 2t 2  Tambient r Tedge Tdevice

Figure 5. Thermal circuit of a cantilever

4
 ME358 Heat Transfer in Microdevices
Tedge  Tambient
Tavg  (22) 9
2 8 contact resistance

Resistance [1E4K/W]
Tedge  Tdevice  rcontact q device (23) 7
6 cantilever resistance
5
Upon taking account of contact resistance, the edge 4
temperature was found not to change considerably, 3
approximately 61ºC. The reason for this is considered to be 2
because contact resistance as well as heat from each 1
cantilever is small. 0
From this calculation, we obtained the new values of 1 11 21 31 41 51
cantilever width. Although the width becomes slightly larger ith cantilever
as expected, it is still similar to that found in the first stage,
as shown in Figure 6. There are two ways to solve this
problem. One is to decrease the thickness of the cantilever, Figure 7. ith cantilever at the temperature
so that the width has to be larger to maintain same Figure 7 shows the values of cantilever and contact
conductance. Another solution would be to employ materials resistances at equilibrium condition for each cantilever. In
with lower conductivity. Calculations using silicon nitride this graph, the changes in contact resistance with
has shown it to be a good candidate for this purpose because temperature change are not considered. For example,
it has more than 10 times lower conductivity than silicon
while the CTE is still comparable with that of silicon.
Therefore to maintain the thermal conductance, the cross- r1  83200  rcantilever ,1  rcontact ,1  78100  5100 [ K / W ]
sectional area (and hence, width) for heat path needs to be (24)
larger.
However, because rcontact is function of temperature, the
equation can be expressed as:
ri (T )  rcantilever ,i  rcontact ,i (T )
(25)
Therefore, total resistance at T=Ti can be expressed as:
1 1 1
With contact resistance   (26)
Without contact resistance Ri (T ) Ri 1 (T ) ri (T )
Even though contact resistance is a function of
temperature, the relative effect of contact resistance on the
total resistance for longer cantilevers would become smaller
as the temperature goes higher. As can be seen in Figure 7,
the percentage of contact resistance for longer cantilevers is
already small. Therefore, we could assume that the
resistance of ith cantilever is approximated as being
constant.

7. Discussion/Alternative solution.
Figure 6. ith cantilever width in equilibrium condition
As the ambient temperature rises, the cantilevers in the
array should contact the device to assist in conducting heat
away, and as the ambient temperature cools, the cantilevers
should detach from the device since they are no longer
necessary to cool the device. However, one potential
limitation of this bimorph concept is that while the
cantilevers will attach themselves to the device as the
ambient temperature rises, they will tend to remain attached
to the device when the ambient temperature cools. Once a
cantilever attaches to a device, heat flow from the device to
the ambient will produce a temperature gradient within the
cantilever, raising the average temperature of the cantilever
above its activation temperature. Thus, even when the
ambient temperature drops below the activation temperature
of the cantilever, the average temperature of the cantilever
5
 ME358 Heat Transfer in Microdevices
will still remain above the activation temperature, and the 7. F. H. Milanez and M. B. H. Mantelli, Theoretical and
cantilever will remain attached to the device. This simple two experimental studies of a bi-metallic heat switch for
layer bimorph is not a good solution because of this problem. space applications, Int. J. Heat Mass Transfer 46, 24,
To avoid this problem, we propose new design to separate 4573-4586 (2003)
the two functions of heat conduction and bimorph actuator. A 8. E. Gmelin, M. A. Palmer, M. Reuther, and R. Villar,
four layer bimorph cantilever with an insulating layer is one Thermal boundary resistance of mechanical contacts
candidate. The next step that we need to explore is finite between solids at sub-ambient temperatures, J. Phys.
element modeling for this structure to see if the insulating D: Appl. Phys. 32, 19-43 (1999)
layer works properly to thermally separate the bimorph and
conducting parts.
Conductive layer
Insulating layer
Silicon
Aluminum

Figure.8. Variables array on the test structure chip.

8. Conclusion
The design of the temperature regulator with variable
thermal resistance was studied. We calculated the dimensions
of each cantilever with/without consideration of contact
resistance. Because contact resistance in equilibrium state at
activation temperature is small, these different considerations
didn't show considerable difference in dimension. Obtained
dimensions for each stage represent unrealistically high
aspect ratio. Therefore alternative approaches, such as use of
thinner cantilever or low conductivity material were
suggested. Also, as a solution for the detachment of the
cantilever, a 4 layer cantilever which consists of bimorph,
insulator, and conductor was proposed. In order to achieve
the feasible design, further analysis on materials, dimension
and structure is required.

9. Reference.
1. Chu W-H, Mehregany M, and Mullen R 1993
Analysis of tip deflection and force of a bimetallic
cantilever microactuator J. Micromech. Microeng. 3
pp4-7
2. Timoshenko S. 1930 Strength of Materials Part I
Elementary Theory and Problems Lancaster Press, Inc.
Lancaster, PA pp242 - 246
3. M.G. Cooper, B.B. Mikic, and M. M. Yovanovich,
Thermal contact conductance, Int. J. Heat Mass
Transfer 12, 279-300 (1969)
4. B. B. Mikic, Thermal constriction resistance due to
non-uniform surface conditions; contact resistance at
non-uniform interface pressure, Int. J. Heat Mass
Transfer 13, 1497-1500 (1970)
5. B. B. Mikic, Thermal contact conductance; theoretical
considerations, Int. J. Heat Mass Transfer 17, 205-214
(1974)
6. K. Nishino, S. Yamashita, and K. Torii, Thremal contact
conductance under low applied load in a vacuum
environment, Experimental Thermal and Fluid Science
10, 258-271 (1995)

6
 ME358 Heat Transfer in Microdevices

Appendix 5. Flip wafer over and

An example of fabrication process flow mask topside

1. Start with SOI

wafer and mask
backside
6. Etch through device
layer, BOX, and part
of handle using
anisotropic etch
2. Etch through
handle using
anisotropic
etch in STS 7. Remove photoresist
Etch and
release the
cantilevers
3. Etch through
BOX
8. Cantilever will bend
upwards as ambient
temperature
increases due to
aluminum뭩 greater
4. Sputter CTE than silicon
Aluminum onto
exposed
backside