Sensors and Actuators A 91 (2001) 249±255

Analysis and testing of a silicon intrinsic-point heater

in a micropropulsion application
Robert L. Bayta,*, Kenneth S. Breuerb
a
United Technologies Research Center, E. Hartford, CT 0610, USA
b
Division of Engineering, Brown University, Providence, RI 02912, USA

Abstract

Ef®cient heating of a ¯uid is demonstrated using a novel heat exchanger in which bulk silicon forms both the heater structure and the
resistive heating elements. Current passed through the heater raises the temperature of the heater ®ns and this energy is transferred to a ¯uid
¯owing between adjacent ®ns. By exploiting the change in sign of the temperature coef®cient of resistivity of the heavily doped silicon, the
temperature of the system is stably maintained at the intrinsic point. A heat exchanger of this nature is integrated with a nozzle, resulting in
a microthruster with elevated chamber temperature, which greatly improves the speci®c impulse, or thrust per unit weight ¯ow of
propellant. A numerical model is presented to optimize the heater design. Benchtop tests demonstrate the inherent stability of the intrinsic
point heater design, while thrust tests demonstrate the improved fuel economy of the micropropulsion system. # 2001 Elsevier Science
B.V. All rights reserved.

Keywords: Microheater; Micropropulsion; MEMS; Deep reactive ion etching

1. Introduction
A wide variety of MEMS applications employ an inte-
grated heater, including microreactors (to catalyze reac-
tions), as demonstrated by Jensen et al. [1], and valves
(to actuate the poppet), as in Henning et al. [2], and
Bergstrem et al. [3]. These heaters typically are comprised
of a snaking thin-®lm, using either metal (e.g. [2]) or
polysilicon (e.g. [4]) as a resistor to generate heat over a
large surface area. In the case of both Bergstrem et al. and
Janson [4], the polysilicon thin ®lm was suspended in order
to minimize the conductive losses and to double the area
available for heat transfer by having both sides of the heater
in contact with the working ¯uid. These systems have been
demonstrated to work well, but are typically hampered by
the relatively low contact area between the ¯uid and the
heating element, which limits the system to relatively low
¯ow rates.
Heater effectiveness can be characterized as the ratio of
the observed temperature rise of the ¯uid to the maximum
possible temperature rise that could occur for a given wall
temperature.
Toutlet;fluid Tinlet;fluid
eˆ
Toutlet;wall Tinlet;fluid
*
Corresponding author.
Thus, a heater with very good thermal contact between the
¯uid and heating element would bring the ¯uid up to the wall
temperature and achieve an effectiveness of unity. Since the
dominant heat transfer mechanism is forced convection, heat
transfer to the ¯uid is governed by Newton's law of cooling.
q ˆ hA Tfluid Twall † (2)
where the area of fluid±heater interaction (A) is often the
practical limitation for achieving high effectiveness, h is the
convection coefficient, which is a function of the Reynolds
and Prandtl numbers, as well as the length scale character-
istic of the flow. By reducing the characteristic length scale,
which in this case is the width of the fluid passage in the
heater, h is increased because the gradients have increased,
and subsequently the effectiveness is increased. There is a
trade-off to be made since pressure drop also increases with
decreasing channel width, and an optimum system perfor-
mance can be obtained.
A second issue of common concern in gas ¯ow heaters is
stability. This arises because the gas viscosity increases with
temperature. Thus, if a local hotspot develops, the ¯uid
viscosity will increase, increasing the pressure drop across
the heat exchanger passage and thus reducing the mass ¯ow
(1) through the channel. This, in turn, reduces the convective
heat transfer and further raises the temperature in the
channel. In this manner, a thermal runaway can ensue
leading to device failure.

0924-4247/01/$ ± see front matter # 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 4 - 4 2 4 7 ( 0 1 ) 0 0 5 9 4 - 5
250 R.L. Bayt, K.S. Breuer / Sensors and Actuators A 91 (2001) 249±255

This paper presents a microfabricated solution for achiev-

ing a highly effective heater in which the heater elements are
formed from the structural material of the ¯uid system Ð
single crystal silicon. This architecture has two principal
advantages. First, the electrical and mechanical functions
are combined. Second, the thermo-electric properties of
silicon allows for an inherently stable operation at high
temperature without risk of thermal runaway.
In this paper, the heater is used to increase the chamber
temperature for ¯uid entering a micronozzle to create a
propulsion system for a microspacecraft. In such applica-
tions, maximizing the thrust per unit weight ¯ow of pro-
pellant (i.e. speci®c impulse, Isp) is highly desirable and is
enhanced by increasing the chamber temperature by pre-
heating the gas prior to expansion. We also present a simple
model to maximize the heater effectiveness, while minimiz-
ing the pressure drop through the ®n array. Varying the Fig. 2. The resistivity of doped silicon is peak when intrinsic and dopant
number of passages, their length, thickness and spacing charge carriers are equal. This plot is for a P-type 1019 atm/cc silicon wafer
accomplishes this goal. [5].

conduction band. When the number of intrinsic carriers

2. Heater operation
The overall heater design is shown in Fig. 1. Gas enters
the port in the top wafer and ¯ows through the heat exchan-
ger, which is de®ned by narrow parallel ®ns upstream of the
nozzle. The entire structure is fabricated from heavily doped
P-type silicon. An electric current ¯ows from top to bottom
through the device, and is focused through the ®ns, heating
them resistively.
The properties of silicon as a semiconductor are used to
maintain stable operation of the device at high temperature.
By fabricating the heater using heavily doped P-type silicon
wafers, the dopant holes will be the primary charge carriers
at low-temperatures. As power is dissipated and the tem-
perature of the device increases, the electrons bound in the
silicon valence bands become thermally excited to the
becomes dominant, the resistivity of the material is governed
by statistical mechanics, and decreases exponentially with
temperature.
This trend in resistance is shown in Fig. 2, and indicates
the intrinsic point for a silicon sample doped to 1019 atm/cc
[5]. The key feature of Fig. 2 is the change in the sign of the
temperature coef®cient of resistivity at the intrinsic point.
The heater is operated in a constant current mode. As the
current increases, the dissipated power, temperature, and
resistance, all rise in response. However, when the intrinsic
temperature is reached, any increase in dissipated power and
temperature results in a drop in resistance. Consequently, the
dissipated power falls and the device returns to operation at
the intrinsic point. Thus, a stable operating temperature is
maintained with the feedback provided by the resistive
properties of silicon.

Fig. 1. Schematic of device. The thruster is deep etched to provide maximum surface area of the heat exchanger and a high-aspect ratio nozzle.
 R.L. Bayt, K.S. Breuer / Sensors and Actuators A 91 (2001) 249±255 251

elements as shown in Fig. 3A. The heat is transferred to the

¯uid convectively and into the up and downstream portions
of the ®n through conduction.
In order to enhance the ®delity of the model, and take
advantage of the heater symmetry, the heat transfer analysis
is performed for the ®n segment shown in Fig. 3B. The
channel symmetry plane is the boundary one-half of the
channel width on each sides of the ®n. The conduction area
is the full cross-section, and the convection area is the wetted
area outlined. By taking into account the upper and lower
surfaces, a closer approximation to the heater geometry is
made.
By summing the energy inputs into each control volume,
the following system results.

d2 Tw
Fig. 3. Design simulations are based on a lumped-parameter model that kw A c dx hP dx Tw _ c dx ˆ 0
Tf † ‡ qA (3a)
dx2
considers the heat generation and transfer along a symmetrical channel
element. _ p dTf ˆ Ph Tw
mc Tf † dx (3b)
Heat ¯ows into the control volumes are positive. Eq. (3a)
3. Heater model is de®ned for the ®n, and Eq. (3b) is de®ned for the ¯uid. The
®rst term in Eq. (3a) is the conduction along the ®n. The
With a heater design selected, the geometry (shown in second term is the convective transfer to the ¯uid. The ®nal
Fig. 3) is to be optimized to yield the highest heater term is the volumetric heat dissipation. Eq. (4b) is the
effectiveness while maintaining the lowest pressure balance between the energy convected downstream, on
drop. Heat transfer in this instance is a convective transport the left-hand side of the equation, and the heat transfer to
problem and will be governed by the bulk motion of the the wall, on the right-hand side: kw is the ®n thermal
¯uid. conductivity, Ac the ®n cross-sectional area that governs
The heater design parameters can be divided into two conduction through the ®n, P the perimeter of the channel
groups: those that govern heat transfer and those that govern cross-section that governs the convective heat transfer, h the
power dissipation. Thus, ®n spacing or gap (wgap), the height convection coef®cient, q the volumetric heat generated, cp
of the channel (h0), the length of the channel (L), and the the coef®cient of speci®c heat. By dividing each equation
number of channels form the heater design space. The power through by the differential dx, canceling terms, and non-
dissipation is governed by the heater resistance, which in dimensionalizing the following system of governing equa-
turn is et by the ®n width, and the resistivity of the silicon. tions is established.
The operational parameters of the device are the mass
d2 Tw
_ the current supplied to the device (i), and the
¯ow rate (m), ‡ Bi Tw Tf † s ˆ 0 (4a)
dx2
gas supply temperature (Tf). With these parameters speci-
®ed, a thermal model is implemented that will determine for dTf
ˆ St Tw Tf † (4b)
the steady-state operating condition: the power that is dis- dx
sipated, the pressure drop across the ®ns, and the exit where
temperature, which will allow effectiveness to be computed.
A 1-D model is selected so that variations at each local hPh20 _ 20
qh hh0 P
Bi ˆ ; s ˆ ; St ˆ
cross-section can be neglected, and only variations along the kw Ac T i kw _ p
mc
length of the ®n are examined. As the ®n spacing grows,
there will be a larger thermal variation at each cross-section. This coupled system requires three boundary conditions.
But for the spacing examined in this analysis, the 1-D model The ®ns are assumed to be insulated at each end, and the
can be justi®ed by applying the classic problem of tempera- inlet ¯uid temperature is speci®ed. These are formulated
ture distribution in fully developed duct ¯ow (the Graetz as
problem), as derived in White [6]. dTw dTw
The model assumes that heat is being generated at a rate ˆ 0; ˆ 0; Ti ˆ 300 K (5)
dx xˆ0 dx xˆL
set by the electrical model, which is a constant current
supply, and variable heater resistance as depicted in As denoted by their symbols, the coef®cient of the wall-
Fig. 2. The dissipation is treated as a volumetric heating ¯uid temperature difference in the ®n equation is a form of
source within the ®n. The heat transfer is computed by the Biot number (Bi), and in the ¯uid equation the coef®cient
performing an energy balance on differential ®n and ¯uid of the wall-¯uid temperature difference is a form of the
252 R.L. Bayt, K.S. Breuer / Sensors and Actuators A 91 (2001) 249±255
Stanton number (St). The non-dimensional source term is
represented by s.
It is important to observe the relevance of each of the
parameters. The Biot number governs the ratio of the
conduction along the ®n to the convective heat transfer.
High Biot numbers combined with high ®n length-to-width
aspect ratios result in large temperature differences between
the entrance region and exit region of the ®ns. Because of the
high heat transfer in the entrance region, all of the heat
generated locally enters into the ¯uid, and heat generated
downstream is conducted along the ®n, so it can enter the
¯uid where gradients are highest. The Stanton number is
similar, as it governs the convection to advection ratio along
the ®n. Thus, when Biot number is low and Stanton number
is high, the ®ns are nearly isothermal. Finally, the source
term merely scales the volumetric heat dissipation, relative
to that conducted along the wall. When the ratio of Stanton
number to the source term is high, then the heater effec-
tiveness will be high. This is due to a majority of the
dissipated heat being convected into the ¯uid, which is
indicative of a high Stanton number ¯ow.
As mentioned previously, the heat transfer is governed by
the ¯uid mechanics occurring within the channel. Thus, the
convection parameter h is a function of Reynolds number as
well as the distance along the ®n the ¯uid has traveled. The
correlation for computing the heat transfer coef®cient was
developed by Stephan and is reported by Kakac et al. [7] for
developing ¯ow along parallel duct walls. For constant
temperature walls, the Nusselt number can be computed as
0:024 Pr=X  †1:14
NuT ˆ 7:55 ‡
1 ‡ 0:0358 Pr 0:81 =X ;0:64 †
where Pr is the Prandtl number of the fluid, and X  ˆ
x= Dfin =2††=RePr, Re Reynolds number based on fin spa-
cing, and x the distance along the fins.
4. Design studies
In order to minimize the pressure drop and maximize
heater effectiveness, a number of test cases were explored.
To best understand the effect of the governing non-dimen-
sional parameters, we consider four cases and determine
how they affect the heater performance. All cases are
operating at a mass ¯ow rate of 378 sccm, and have
Table 1
Heater design casesa
Case No. of fins L (mm) W (mm)
1 5 500 100
2 10 125 50
3 5 125 50
4 1 7000 50
a
Case 2 was selected for testing.
50 mm wide ®ns. This ¯ow rate was chosen to keep the
operating power below 10 W, and maintain as high a Rey-
nolds number through the nozzle as possible to minimize
viscous losses (550 at this condition, which has provided
reasonable performance [8]). The heater design and perfor-
mance parameters are summarized in Table 1.
Case 1 is a reasonable design. The pressure drop is 7% of
the total, and a very large temperature rise is achieved. In
case 2, the ratio of the Stanton number to source term has
decreased, and as expected, the effectiveness has dropped,
but only by 2%. However, the pressure drop has decreased
by a factor of 10. Thus, this is a better design due to the
higher injected pressure. For all cases, the dissipated power
is 7.7 W, which is set by the ¯ow rate and temperature rise.
The temperature pro®les along the ®ns are depicted in
Fig. 4a, and the silicon is almost isothermal.
Case 3 is the same as case 2, but with 5 ®ns. By halving
the number of ®ns, the mass ¯ow through a given channel is
doubled, and the characteristic length is doubled. This
reduces the temperature rise, as well as, increases the
pressure drop, as shown in Fig. 4b. This is the lowest Stanton
number to source term ratio, and the lowest effectiveness,
yet.
Finally, the case representative of a metalized snaking
resistor (a ¯at plate heater) is analyzed as a case with 1 ®n
(split between the sidewalls). Even with a chamber 7 mm
long, the heater effectiveness is only 90%. By using the
design with ®ns, a more compact arrangement can be
attained with a similar pressure drop.
The analysis of these design cases illustrates the dif®culty
of maintaining a temperature gradient in the bulk silicon at
(6)
this scale. Thus, the strategy has been to run the thruster chip
in an isothermal manner, and insulate the entire die within a
ceramic package to minimize losses. These calculations do
not account for conduction losses through the electrical or
gas lines, but will indicate the most effective heat transfer
from the standpoint of the ¯ow. Therefore, it is expected
that higher power dissipation will be required to reach a
given wall temperature, but the ¯uid will also reach that
temperature.
5. Heater/thruster fabrication
The heater/nozzle system is fabricated in silicon using
DRIE. A halo mask is used to simultaneously outline the
e DP (atm) Bi s St
99 0.21 1.1 0.5 6.5
97 0.02 1.8 1.5 13.0
85 0.04 1.8 2.5 6.6
90 0.01 0.9 1.5 0.15
 R.L. Bayt, K.S. Breuer / Sensors and Actuators A 91 (2001) 249±255 253

Fig. 4. Temperature profile along the channel: (a) 10 fins 125 mm in length; (b) 5 fins same length. Fluid temperature rise is less for (b), because more flow
must pass through a given channel.

large cavities as well as de®ne the small heater passages and

nozzle throat. By matching these widths, a constant loading
is maintained during the etch. In addition, a nested mask is
used which allows the through-wafer etch to proceed ahead
of the heater-®ns. This results in the heater ®ns being
connected by a 50 mm high `bus-bar', which maintains their
mechanical integrity and provides electrical functionality.
After etching, the cleared ¯ow channels are encapsulated by
fusion bonding silicon wafers to the upper and lowers
surfaces. This process is summarized in Fig. 5. Fig. 6 is a
S.E.M. of the 8.25:1 expansion ratio nozzle with a throat
width of 65 mm, a nozzle depth of 491 mm, and the heater
analyzed in case 2 of the previous section.
Fig. 6. S.E.M. of thruster with integrated heater after first bonding is
complete. The bus bar is holding the fins together.

Current is supplied to the heater through electrodes

clamped to the top and bottom of the die (see Fig. 1). The
large etched regions on either side of the ¯ow chamber
reduce parasitic current paths through the bulk silicon, and
direct a larger fraction of the total current through the ®ns,
which are in contact with the ¯ow. The working ¯uid (nitro-
gen) is delivered through Kovar tubing brazed to a stainless
steel plate and ®xed to the silicon via a glass frit. The plate is
clamped with an O-ring to a supply manifold. Further
fabrication and assembly details are found in Bayt [9].

6. Experimental result

Fig. 7 shows the temperature and power with time for a

Fig. 5. Fabrication process relies upon nested masking to release nozzle
chamber cavities, and maintain fin structure connected by a strip of silicon.
heater operated on the benchtop with no gas ¯ow. These
measurements are made with K-type thermocouples
attached directly to the silicon with thermally conductive
epoxy. The current is increased in a series of discrete steps,
as indicated by the dotted line. The temperature increases at
254 R.L. Bayt, K.S. Breuer / Sensors and Actuators A 91 (2001) 249±255
Fig. 7. Heater temperature variation with current. A 7008C intrinsic point
is observed, as predicted by the doping level.
about 258C/s immediately after power is applied, and then
asymptotes to a steady value during the next 300 s. This is
due to the increasing power dissipation with temperature. At
a current of 6.06 A, suf®cient power was dissipated to reach
the intrinsic temperature (7008C). For subsequent times,
the temperature is seen to rise brie¯y above the intrinsic
temperature before dropping back down due to the inherent
stability of the heater architecture. This behavior is observed
independently of the driving current, which was continually
increased.
Thrust tests were performed with the thruster depicted in
Fig. 6 to determine thrust and propellant ef®ciency. The
thrust stand was calibrated using known trial weights and
had an accuracy of 0.5 mN. The entire apparatus was
housed in a large vacuum chamber with a background
pressure of 50 mTorr.
Fig. 8 shows the speci®c impulse versus thrust for
this system operated over a range of heater powers. This
Fig. 8. Thrust per unit weight flow (Isp) is an indication of fuel economy,
and thrust indicates propulsive authority.
illustrates the fuel economy for a given control authority. For
®ne control, small thrusts are desired, but as shown here,
performance decreases for cold ¯ows. For the cold-¯ow
results (shown by square symbols), the thrust is reduced by
reducing the chamber pressure, and hence the mass ¯ow rate
and Reynolds number. Reynolds number is the governing
similarity parameter that scales the in¯uence of viscous
effects. For two values of the chamber pressure (38 and
44 psia), the pressure is held constant and the heater power is
increased by increasing the current through the ®ns. As the
temperature rises, the speci®c impulse rises (as expected)
and the thrust is observed to decrease. The thrust is com-
posed of contributions from the exit momentum (velocity)
and pressure.
Pe
T ˆ re Ue2 Ae ‡ Pe Ae ˆ gRTMe2 Ae ‡ Pe Ae ˆ mg†I
_ sp
RTe
(7)
where the flow density (r), velocity (U), area (A), pressure
(P) are defined at the exit plane (e). Note that this predicts
that the thrust should be independent of temperature (T).
Velocity is computed as the product of the local speed of
sound and Mach number, which under isentropic conditions
are both purely a function of the ratio of the local area to the
throat area of the nozzle. However, as temperature increases,
the Reynolds number decreases, and the exit Mach number
(Me) decreases due to the boundary layer blockage effects
described in Bayt et al. [10]. For this reason, the observed
thrust reduces with temperature, and even more importantly,
the Isp relative to the theoretical also decreases significantly.
This causes the deviation of the experimental data points in
Fig. 8 from a vertical line, which would be the ideal for a
constant pressure.
Fig. 9 illustrates the decrease in Isp relative to that
calculated for no viscous losses as a function of Reynolds
Fig. 9. The Isp efficiency variation with Reynolds number for heated cases
in comparison to unheated runs. This verifies that Reynolds number is the
relevant similarity parameter.
 R.L. Bayt, K.S. Breuer / Sensors and Actuators A 91 (2001) 249±255
Fig. 10. Variation of propulsive efficiency with Reynolds for the 44 psia
chamber pressure case. If the lead length is optimized for least power
losses, as well the conduction into the support, the least losses curve could
be achieved.
number. The consistency between hot and cold performance
data when normalized by Reynolds number indicates this
performance decrease is a viscous phenomenon and not a
limitation of the heat transfer in the ®n array. Though the Isp
in Fig. 8, is less than the theoretical value, it is 50% larger
than that achieved with the cold gas ¯ow. Thus, the space-
craft designer can trade electrical power for increased
propellant ef®ciency, which is desirable since power can
be replenished via the solar arrays, whereas propellant
cannot be re-supplied.
The overall propulsive ef®ciency (thrust for a given
propellant and electrical power) of this demonstration device
is quite low (18%) primarily due to parasitic electrical losses
in the electrical leads and test structure. The variation in this
ef®ciency is shown in Fig. 10, as a function of operating
condition or Reynolds number. Due to the constraints of the
test set-up, and limited optimization of the MEMS package,
considerable heat was lost to the test stand, and through the
leads. If the lead length were optimized, and the insulating
package improved, it is believed the curve labeled least
losses could be achieved.
7. Conclusions
The present design has two key features that make it
attractive for integrated micro-heat exchangers. First, the use
of bulk silicon as both the structural and electrical material
simpli®es the system architecture and allows for high ¯uid/
heater contact area. Second, the properties of silicon at the
255
intrinsic point are very attractive for stable operation, parti-
cularly for gases where thermal runaway can be problematic.
By increasing the chamber energy in a microthruster, the
mass ¯ow required for a given thrust level is greatly reduced.
This translates into increased satellite life for a given
propellant supply. Although the device ef®ciency is low
in this initial example, little attempt has been made to
minimize parasitic losses through the leads and test structure
and one expects the ef®ciency could be improved in future
designs.
Acknowledgements
The authors would like to thank Professor Martin Schmidt
of the MIT Microsystem Technology Lab, who ®rst sug-
gested the exploitation of the intrinsic properties of silicon
for the heater. His guidance was invaluable in the design and
fabrication phases. Additional fabrication support was pro-
vided by Dr. Arturo Ayon. Additional testing support was
provided by Dr. Vadim Khayms. This work was supported
by the Air Force Of®ce of Scienti®c Research.
References
[1] K. Jensen, R. Srinivasan, I.-M. Hsing, J. Ryley, M.P. Harold, M.A.
Schmidt, Micromachined chemical reactors for surface catalyzed
oxidation reactions, in: Proceedings of the 1996 Solid State Sensor
and Actuator Workshop, Hilton Head Isl., SC, 3±6 June 1996,
pp. 15±18.
[2] A.K. Henning, J. Fitch, E. Falsken, D. Hopkins, L. Lilly, R. Faeth, M.
Zdeblick, A thermopneumatically actuated microvalve for liquid
expansion and proportional control, in: Proceedings of the Interna-
tional Solid State Sensors and Actuators Conference, pp. 825±828.
[3] P.L. Bergstrom, J. Ji, Y.N. Liu, M. Kaviany, K.D. Wise, Thermally
driven phase-change microactuation, J. MEMS 4 (1995) 10±17.
[4] S.W. Janson, Batch-fabricated resistojets: initial results, in: Proceed-
ings of the International Electric Propulsion Conference, IEPC-
97-070, Cleveland, OH, USA, 1997.
[5] G.L. Pearson, J. Bardeen, Electrical properties of pure silicon and
silicon alloys containing boron and phosphorous, Phys. Rev. 75
(1949) 865±886.
[6] F.M. White, Viscous Fluid Flow, 2nd Edition, McGraw-Hill, New
York, 1991.
[7] R.K. Kakac, S. Shah, A.E. Bergles, Low Reynolds Number Heat
Exchangers, 1st Edition, Hemisphere, Washington, DC, 1983.
[8] R.L. Bayt, Analysis, fabrication and testing of a MEMS-based
micropropulsion system, Ph.D. thesis, MIT, 1999.
[9] R.L. Bayt, K.S. Breuer, A.A. Ayon, DRIE-fabricated nozzles for
generating supersonic flows in micropropulsion systems, in:
Proceedings of the 1998 Solid-State Sensor and Actuator Workshop,
Hilton Head Isl., SC, 8±11 June 1998, pp. 312±316.
[10] R.L. Bayt, K.S. Breuer, Fabrication and testing of micron-sized cold-
gas thrusters, in: Micropropulsion for Small Spacecraft, Progr.
Astronuat. Aeronaut. 187 (2000) 381±397.