a force of .0059 nN/volt2 per comb-finger height (ym) - - PDF document

A High Force Low Area MEMS Thermal Actuator

ABSTRACT zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Michael J. Sinclair Microsoft Research Microsoft Corporation One Microsoft Way, Redmond, WA 98052 Phone: (425) 703-8343 Fax: (425) 936-7329 Email: sinclair@,microsoft.com INTRODUCTION This paper presents a new type of MEMS (micro-electro- mechanical systems) actuator consisting of an array of in- plane micro-fabricated thermal buckle-beam actuators. The technology used in MEMS actuators is typically magnetic, electrostatic or thermal. Magnetic actuators may require special materials in the fabrication process while electrostatic actuators typically require high voltages, large chip areas and produce very low forces. Thermal actuators have seen some use in MEMS applications, the most popular being the pseudo-bimorph that relies on differential expansion of a cold and hot arm to cause it to bend in-plane (parallel to the substrate). These thermal actuators typically generate on the

• rder of a few micro-Newtons each but can be combined for

larger forces by linking with small tendons. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

A disadvantage of

this type of actuator is that it moves in an arc where most desired movements are linear. Also, when combined in an array, the linking tendons consume much of the energy in bending them. Also, arrays of these can still occupy a fairly large chip area. The electro-thermal actuator described here resembles a chevron where an array of buckle-beams are packed close together and link two common anchored arms with a movable third arm. Arrays can be made within a single released micromachined layer and generate many mN of force. Additional actuators can be arrayed with no coupling penalty and occupy much less area that an equivalent pseudo-bimorph

• actuator. Preliminary tests indicate that a 450 x 120 pm array

consumes 240 mW of power, deflection up to 14 pm and can produce many milli-Newtons. A chip of actuator geometry variations and different applications has been fabricated and tested. Key Words: MEMS, micro-electro-mechanical systems, thermal actuator, buckle-beam Most batch fabricated micro-mechanical systems require on- chip movement of microstructures, either by outside forces (pressure, acceleration) or put into motion by on-chip

• actuators. The desired attributes of an intemal actuator are

small chip real estate, large deflection (>lOpm) and an electrical requirement compatible with today’s CMOS

• circuitry. MEMS actuators are typically used for either one-

time deployment of structures for automatic assembly, an in- use adjustment such as focusing or tweaking an optical parameter or constant periodic actuation as in the case of micro-optic scanners. Electrostatic actuators rely on the attractive forces between oppositely charged conductors in close proximity. Magnetic actuation uses the force of attraction or repulsion between a magnetic field produced by an electric current and a magnetic material or other

• electromagnet. These are typically relegated to laboratory

research, as they usually require exotic fabrications steps. Electro-thermal actuators rely on the joule heating and resulting small mechanical expansion of a conductor when a current is passed through it. One of the most popular actuators in the MEMS community is the electrostatic comb drive. This type of actuator can produce a force of .0059 nN/volt2 per comb-finger height (ym) zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

• 113. A

100 finger, 2 pm thick comb drive occupies a chip area of about 0.15 square mm and will produce an output force of around 3 pN with a 50 volt drive at negligible current. This yields an actuator force density of about 20 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

pN per square

• mm. Proportionately higher forces can be achieved with

higher aspect ratio structures. Advantages of the electrostatic actuator are small actuation energy and relatively high frequency response. Disadvantages are high drive voltage, large area and low output force. Conversely, employing the thermal actuator array proposed by Reid [2], one can achieve about 450 pN per square mm of MEMS chip area. The electrical power required is 3.87 mW per pN. These actuators depend on the differential thermal expansion of two polysilicon arms to produce a pseudo- bimorph that deflects in an arc. For an array, these devices may be coupled to a beam through bending yokes. These yokes however, consume much of the force output of their

0-7803-591 2-7/00/$10.0002000

I E

E E

127

2000 Inter Society Conference

• n Thermal Phenomena

Authorized licensed use limited to: IEEE Xplore. Downloaded on October 21, 2008 at 05:26 from IEEE Xplore. Restrictions apply.

actuators just in their bending. The coupling of two zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

• r more

actuators to a common beam produces a linear movement zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

• usually desired in MEMS systems. The actuator array

presented here consists of only one thermal expansion beam per actuator and can produce about zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

3700 pN

per square mm and 1.53 mW per pN. As structures become more complicated, especially in the case of free-space optical devices [3], the one-time deployment required for assembly becomes more important and reliant on high-force, low-area

• actuators. Many of the deployment actuators today are of the

comb drive type and typically occupy many times the area of the device they are deploying. DEVICE FABRICATION The tested actuators were fabricated using the Multi-User MEMS Processes (MUMPs) zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

[4]. MUMPs is a surface

micromachining process employing a substrate, an insulating nitride layer and three structural polysilicon layers separated by two sacrificial oxide layers as shown in Fig. 1, The second Figure 2. Single buckle-beam actuator. The applied voltage causes ohmic heating and expansion between the two fixed anchors, buckling the beam at the midpoint. (parallel to the substrate). The actuator displacement zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA d is given by

d = [ Z2 + 2 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

(I) I' - zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA 1 ~ o s ( a ) ~ - 1 sin(a) 1.Spm 2.Opm 0.5pm

Figure 1. Cross-section view of the MUMPs fabrication process showing three polysilicon layers with a single anchor point. and third polysilicon layers (Poly1 and Poly2) are both releasable to act as movable structures. A 0

. 5

pm gold layer can be pattem-deposited on the Poly2 layer for optical reflection or increased conductivity. The final step performed is an HF etch of the intervening sacrificial oxide layers and subsequent drying. As is the problem with many MEMS devices, stiction can and does occur between released layers. Movement by the high-force thermal actuators usually releases the offending structure. This is not always the case in stuck electrostatic actuators that have to be freed manually. ACTUATOR DESIGN The design is based on an in-plane buckle-beam actuator [SI, [6] as shown in Fig. 2. As a voltage is applied between the mechanical anchors, ohmic heating of the two half-beams causes them to expand and ultimately buckle. The resistivity

• f polysilicon allows the actuator to operate at voltages and

currents compatible with standard integrated circuitry (CMOS). The beam is normally designed with a pre-bend angle a so buckling will have an affinity to move in-plane where

1 is the single beam length (buckle-beam half-length) I' is the elongation of the beam due to thermal

a is the pre-bend angle of the beam.

expansion and the coefficient of thermal expansion used for polysilicon is 2 . 3 3 ~ 1 O-6f'C. The buckle-beam heating temperature was kept below 800 'C to prevent self-annealing which can cause irreversible damage. It should be noted that a buckle-beam can be fabricated out of either or both of the MUMPs polysilicon released layers giving a possible actuator thickness of 1.5, 2.0 or 3.5 pm. All that is required of the fabrication process is that it include at least one releasable layer with a positive temperature coefficient of expansion and capable of carrying a current for

• hmic heating.

Arrays of buckle-beam devices can be easily designed by arranging them in a pattern resembling a chevron as shown in

• Fig. 3. A center beam is added to stiffen the midpoint and

allow mechanical coupling of the individual beams as well as providing a method of transmitting the linear force to another

• device. There is no theoretical limit to the number of beams

added as long as the device and conductors can handle the current and heat, the beams can lose heat rapidly and there is no cross coupling of heat from one beam to another. Most of the actuator arrays explored in this paper consist of pairs of 218 pm half-beams in varying number and thickness. This length was chosen for reasons of published optimum configuration for other thermal actuators using the MUMPS process [2]. If more than one actuator array is connected (mechanically and electrically) to a single micro-structure, care must be taken to eliminate any common mode currents that arise when the actuators are excited differently, Fig. 4 shows an

128

2000 Inter Society Conference

• n Thermal Phenomena

Authorized licensed use limited to: IEEE Xplore. Downloaded on October 21, 2008 at 05:26 from IEEE Xplore. Restrictions apply.

FORCE AND DEFLECTION MEASUREMENT zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Force Work Coupling Beam Figure 3. An array of four buckle-beam actuators with the addition of a coupling beam. The output force is linear zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

• four

times that of a single actuator, Force

I zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

1

I

A CAD image for a typical actuator is shown in Fig. 5 with a buckle-beam cross-section shown in Fig. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

• 6. In this case, linear

capture bearings were added to prevent the center coupling beam from deflecting or buckling out-of-plane. As the pre-bias angle is usually very small, on the order of a fraction of a degree, it was thought that there would be a tendency for the beams to buckle away from the substrate. This was not the case in most observed tests, unless the center beam was constrained to move in the desired direction. In test cases

Electrical 2 x 2 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

x zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA 200 nm

Capture connection polysilicon beam.

\

Bending beam \Calibrated stops to for force measurment truncate bending bemm

Figure 5. CAD image of a typical chevron actuator array. Note the capture bearings used to prevent the buckle-beams from deflecting out-of-plane. 2x2 um polysilicon buckle beam Coupling beam Anchor zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

\ I

/

Figure 4. Alternative connection methods to minimize cross

• currents. If more than one actuator array is connected to a

microstructure, the common-mode current must be minimized

• r damage could result.

Dimple Nitride Substrate - Figure 6. Cross-section view of the actuator array of Fig. 5 . Note the dimple bearing used to reduce stiction. alternative method of electrical connection to the buckle-beam array that can help eliminate this problem. The current in Fig. 3 passes from one anchor to the other, placing the center- coipling beam at the half-resistanceholtagk point. The current shown in Fig. 4 is fed from both anchors toward the coupling beam, which is at ground or a common mode potential. This alternate connection could also be used to cause unequal currents to flow in both halves of the array, moving the center coupling beam a small distance either way and orthogonal to the primary displacement direction. This action is presently being explored in an application where the smaller displacement can be used to accurately position an external structure clamped by the larger displacement. where the pre-bias angle was zero, most actuators attempted to buckle out-of-plane. Also shown in Fig. 5 is the vernier used to measure the actuator’s deflection and has a resolution of about 0.5

• pm. Note also the force measuring bending-beam

and the calibrated backing structure. In operation, the actuator first moves, un-forced, by 2 pm where it then comes in contact with the force measurement beam. Using a micromanipulator probe, one can mechanically “short out” a section of the bending-beam in 4 pm increments, resulting in a stiffer or higher applied force on the actuator. Unfortunately, the beam wasn’t stiff enough as most actuators exceeded expectations in

• utput force and often broke the force bending-beam. Data

had to be gathered from actuators in which a number of buckle beams were physically removed to reduce the force to a

129

2000 Inter Society Conference

• n Thermal Phenomena

Authorized licensed use limited to: IEEE Xplore. Downloaded on October 21, 2008 at 05:26 from IEEE Xplore. Restrictions apply.

measurable quantity. The full array response can be predicted as the forces from each buckle-beam add linearly. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA A feature used and shown in Fig. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA 6 is a dimple bearing available in the MUMPs process. As was stated earlier, stiction can prevent structures from moving where they are in contact with the nitride layer. 0.5 pm dimples help reduce the surface area of released polysilicon structures that would normally have contact with the nitride layer. Fig. 7 is a photomicrograph of the actuator depicted in Fig. 5. Figure 7. Photomicrograph of a chevron actuator array. The probe at the bottom is used to shorten the force-measuring beam to 60 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

pm

The actuator is deflected about 7 pm. A probe is shown shortening the 130 pm force beam to 60 pm. By measuring the deflection of the bending-beam, the applied force can be

• computed. Note that the applied force is not the same as the

total force capability of the actuator. For small deflections of the force beam, the applied force is calculated [2] F zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA = E t d w 3 4 i3 where F = applied force E = Young’s modulus - 160 Gpa [7] t = beam thickness in pm w = beam width in pm

1 =beam length in pm

TEST RESULTS Variations in actuator geometry design were explored in an attempt to discover optimal actuator configuration. Tested values included pre-bend angle, actuator thickness, beam length, excitation power, and number of beams per array. Measurements were made by placing the released MUMPs die

• n a probe station. Microprobes were used to connect to the

chip by contacting the bonding pads designed in the devices. A finction generator and amplifier were used as the excitation source with an oscilloscope monitoring the terminal voltage. Observations were made through the probe station’s

• microscope. In operation, a probe was positioned against the

force measuring bending-beam at the required length for the applied force. A low frequency square wave excitation signal was then applied to the actuator through its connection pads and deflection noted by observing the vernier with a microscope. The actuators exhibited an output force proportional to the number of buckle-beams, actuator layer thickness and pre- bend angle. The measured unforced deflection was dependent

• n actuation voltage, beam length and pre-bend angle and not
• n actuator thickness or number of beam-pairs. The resistance

was predictably proportional to the beam length and inversely proportional to the beam thickness and number of beams per array. A graph of actuator deflection for various pre-bend angles is shown in Fig. 8. The data was taken from a number of 2 pm thick structures at two different applied forces by selecting force beam lengths of 130 and 60 pm. It can be seen (and predicted) that actuators with a small pre-bend angle zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

(<OS

degrees) exhibited little or no deflection and hence little output

• force. For very small pre-bend angles, the actuator sometimes

refused to move in-plane and, instead, buckled out-of-plane. Predictably, with larger pre-bend angles, the deflection was reduced but the available force increased. Fig. 8 also indicates that, for a 60 pm force beam, the.optima1 pre-bend is around 1

!

i zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

• 130

um force beam zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

• 60

um force

1

beam 1

2

3

Pre-bend

Angle (degrees) Figure 8. Deflection distance vs. pre-bend angle results for 2.0 pm thick polysilicon actuators. Two different force- measuring beam lengths were used. degree for maximum deflection, Fig. 9 is a graph of deflection for a series of 2 pm actuators with various pre-bend and excitation voltages. It indicates that the actuators exhibit a linear response when the excitation is above 2 volts. The slope

• f the curves indicates the higher the pre-bend angle, the lower

the deflection. Deflection response was measured in all actuators to be around 2KHz. This was measured by increasing the excitation frequency until the measured deflection fell to half the

2000 Inter Society Conference

130

• n Thermal Phenomena

Authorized licensed use limited to: IEEE Xplore. Downloaded on October 21, 2008 at 05:26 from IEEE Xplore. Restrictions apply.

I zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA I zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

APPLICATION

I

Deflection vs Actuation Voltage for Actuators with Different Pre-bend Angles

I

1 ’ 2 1

I

I zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

5

I C

I

• Actuation (volts) zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

I I

*pre-bend=.524 deg

• pre-bend=l.O5

deg +pre-bend=1.31

11

deg

• pre-bend=l.83

I zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

Figure 9. Actuator deflection vs. excitation voltage. All actuators appear to be linear for excitations greater than 2

• volts. The slope is pre-bend angle dependent.

maximum amplitude. Preliminary tests indicate that an order

• f magnitude increase in usable actuation frequency can be

realized by using a mechanically high-Q structure operating at resonance. CONCLUSIONS These actuators can be used almost anywhere MEMS linear motion is required. They produce high force though consume considerable power. The small deflection (-10 Fm) can be extended through leverage, gearing or clutcWfriction drive. As was stated previously, the force output exceeded expectations, as the force bending-beam was not designed strong enough. In

• rder to obtain a meaningful force measurement, all but two

buckle-beams were removed from a 2 p actuator with a pre- bend of 1 .OS degrees. The modified actuator was able to bend a 68 p

m force-measuring beam by 3 p.

This corresponds to a force of about 20 pN

• r 5

pN

per single beam (one half of a buckle-beam). This value is equivalent to the force obtained from a single pseudo bi-morph actuator described by Reid, et.

• al. [2]. Using the technique described here, a buckle-beam

actuator array consisting of 48 two pm thick beams would have an output force of around 240 pN and occupy an area of zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA 65,OO sq. pm. This same force would require an array of pseudo bi-morph devices occupying 533,000 sq. pm of chip real estate or 12,000,000 sq. pm for an electrostatic actuator

• array. Future work will consist of exploring other geometry

variations of these actuators such as longer beam lengths, beam-to-beam spacing and methods to avoid out-of-plane

• buckling. Applications will include deployment systems for
• ptical devices, rotary motion, mirror scanners and other

display devices. The photomicrograph in Fig. 1 is an example of a gear motor powered by two actuator arrays. in this application, the actuators are connected similarly to that shown in Fig. 4 in

• rder to reduce the common mode current when driving the

actuators from different sources. The actuators were excited in phase quadrature. This phase difference causes the linear drive gear to move in a circular motion, meshing about half of the time with the larger circular gear. Square wave signals such as the output from CMOS digital circuitry may also be used to actuate this motor. If the amplitude is adjusted properly, the drive gear will advance the circular gear by one tooth for each four-phase period. The result is similar to a four-phase stepper motor, though only one phase produces gear drive. The motor is capable of speeds around 1,600 rpm. Another application is a linear rack drive where the actuators are used to advance a beam through friction and can be used to deploy MEMS out-of-plane devices such as those used in free- space optical systems. Figure 10. Photomicrograph of a thermally actuated gear

• motor. The actuators are connected in a common mode

configuration and excited with two sources in phase quadrature that causes the drive gear to move in a circular pattern.

131

2000 Inter Society Conference

• n Thermal Phenomena

Authorized licensed use limited to: IEEE Xplore. Downloaded on October 21, 2008 at 05:26 from IEEE Xplore. Restrictions apply.

BIBLIOGRAPHY zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

[I] zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

W.

YE, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

• S. Mukherjee, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

N.

MacDonald, “Optimal Shape

Design of an Electrostatic Comb Drive in Micro-Electro-

Mechanical Systems,” zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA NSF zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA research report under grant ECS-

9321 508.

[2] J. Reid, V. Bright, J. Comtois, “Force Measurements of a Polysilicon Thermal Micro-Actuator, ”

• Proc. SPIE, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

vol.

2882,

• pp. 296-306, 1997.

131 M. Wu, “Micromachining for Optical and Optoelectronic Systems, ”

• Proc. IEEE, vol. 85, no. 1 1 , pp. 1833-1856, Nov.
• 1977. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

[4] D

.

Koester, R. Mahedevan, A. Shishkoff, and K. Marcus, “MUMPS Design Handbook,” rev. 4, Cronos Integrated Microsystems, 302 1 Comwallis Rd, Research Triangle Park, NC 27709, httr,://www.memsrus,com/cronos/MUMPs.pdf., May 1999. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

[5] L. Que, J. Park, and Y.

Gianchandani, “Bent-Beam Electro-Thermal Actuators for High Force Applications,” IEEE MEMS ’99 Conference, Orlando, FL, pp. 31-36, 1999. [6]

• R. Cragunand and L. Howell, “Linear Thennomechanical

Microactuators,” MEMS 1999, vol. I , Nashville, TN. pp. 181- 188, Nov., 1999. [7] “MEMS Material Database,” MEMS Clearinghouse, htt~:/!nicins.isi.cdu/mcms/matcrials/indclt.html.

132

2000 Inter Society Conference

• n Thermal Phenomena

Authorized licensed use limited to: IEEE Xplore. Downloaded on October 21, 2008 at 05:26 from IEEE Xplore. Restrictions apply.