You are on page 1of 12

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 16, NO.

5, OCTOBER 2007

1173

A Highly Reliable Lateral MEMS Switch Utilizing Undoped Polysilicon as Isolation Material
Wendian Shi, Norman C. Tien, Member, IEEE, and Zhihong Li, Member, IEEE
AbstractThe lateral actuated switch requires an isolation structure to provide mechanical coupling and electrical isolation between the actuator and the contacts. This isolation structure usually imposes extra difculty on the fabrication process. In previous reports, we demonstrated a thermal actuated lateral switch, where the nitride isolation structure was a weak point, leading to reliability problems. In this paper, we developed a modied switch utilizing undoped polysilicon as the isolation material. The undoped-polysilicon isolation structure requires only one extra step of sheltered implantation, and it provides robust mechanical connection. A 20-m-long undoped-polysilicon isolation structure has a current leakage of less than 2 nA under a 15-V operation voltage. The proposed switch works under a 12-V driving voltage with 60-mW input power. The time response is measured to be 130 s, and a maximum operation frequency of 4.5 kHz is reached. An ON-state insertion loss of 0.41 dB at 20 GHz and an OFF-state isolation of 20 dB at 20 GHz have been achieved on the normal low-resistivity silicon substrate. The undoped-polysilicon isolation method can be used in other surface-micromachined lateral switches as well. [2007-0022] Index TermsElectrical isolation, isolation structure, mechanical coupling, microrelay, radio frequency (RF) switch.

I. I NTRODUCTION HE LATERALLY actuated microelectromechanical systems (MEMS) switch has drawn more attention recently due to its in-plane design exibility. The lateral switches can be fabricated with the surface-micromachined polysilicon process [1][3], the bulk-micromachined silicon process [4], [5], the thick metal plating process [6], or other nonstandard processes [7]. Various actuation approaches are also investigated, including the electrothermal actuation [2][4], the electrostatic actuation [8], the piezoelectric actuation [9], and the electromagnetic actuation [10]. Among all these methods, the electrostatic actuation and electrothermal actuation are most attractive. The electrostatic actuation has the merits of low power dissipation and high driving frequency, but a relatively high actuation voltage is usually needed. In contrary, the thermal actuation offers the advantages of low driving voltage, high driving force, and therefore low contact resistance [11], but the

Manuscript received November 24, 2006; revised April 2, 2007. This work was supported by the National Nature Science Foundation of China under Grant 60528009. Subject Editor H. Zappe. W. Shi and Z. Li are with the Department of Microelectronics, Peking University, Beijing 100871, China (e-mail: zhhli@ime.pku.edu.cn). N. C. Tien is with the Department of Electrical Engineering and Computer Science, Case School of Engineering, Case Western Reserve University, Cleveland, OH 44106-7071 USA. Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/JMEMS.2007.901121

high power consumption is a major drawback compared with the electrostatic actuation [12], [13]. Therefore, the actuation method should be selected according to the application requirements of the switches. One major benet of the lateral switch is the ability to cofabricate the actuator, the contacts, the conductor paths, and the supporting structures all in one single lithographic step [14]. However, an extra isolation structure is necessary for the mechanical coupling and electrical isolation between the actuator and the contacts. It usually requires extra process steps or special materials for this isolation structure. Reference [15] reported an isolation method of creating insulative regions in conductive materials for the silicon-oninsulator instruments. An inverse approach of creating conductive regions in dielectric materials was developed for the molded structures [16]. Both methods required combining process steps including etching trenches, relling dielectric/ conductive materials, and etching back to form the isolation structure. To simplify the process, a maskless anisotropic etch was used to form a dielectric sidewall for the isolation in the thermal actuator [17], but this method could not avoid to form dielectric sidewalls between the contact head and the signal lines in lateral switches. Reference [10] reported a lateral switch where a SiO2 layer was used as the isolation structure. In this case, the switch with the SiO2 layer was released with an unusual epoxy sacricial layer and oxygen plasma etch. It is not suitable in other cases such as surfaced-micromachined switches because the epoxy layer signicantly limits the thermal budget of the following processes. Photoresist was also employed as the isolation material in a thermal actuated lateral switch [1], but no detailed performance of the photoresist isolation structure was shown. In previous reports, we demonstrated a thermal actuated lateral switch that has the advantages of low driving voltage, high RF performance, and simple fabrication process [2], [3], [11]. This switch adopted a piece of low-stress silicon nitride lm as the isolation structure between the polysilicon actuator and the contact head. The nitride isolation structure was a weak point, which might cause reliability and yield problems, because the nitride-polysilicon adhesion was not very strong. During the operations of the switch, it was found that the nitridepolysilicon adhesion was easy to break, particularly, when contact force was increased to achieve low contact resistance. Besides, the nitride-polysilicon interface might be attacked by the hydrouoric (HF) acid etching during structure release. The SU8 photoresist has also been tried as the isolation material, but the deformation that is caused by the stress mismatch was too large to be tolerated.

1057-7157/$25.00 2007 IEEE

1174

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 16, NO. 5, OCTOBER 2007

Fig. 1. Schematic overview of (a) the proposed switch and (b) the previously reported switch, and the cross-sectional comparison of (c) the undoped-polysilicon connection and (d) the nitride connection.

Fig. 2. Schematic illustration of the loaded stress during the ON-state of the switch in the (a) push-type design and (b) pull-type design.

This paper proposes a modied thermal actuated lateral switch that utilizes the undoped polysilicon as the isolation material. The undoped-polysilicon isolation structure has the advantages of a simple process, robust mechanical strength, and high reliability. Due to this robust structure, a stable pull-type actuation design and an optimized actuator can be employed to reduce the switching time and power consumption. The performances and lifetime of the switch are systematically investigated.

II. D ESIGN A. Isolation Principle Fig. 1 compares the schematic overview of the proposed microswitch [Fig. 1(a)] and the previously reported one [Fig. 1(b)] [3]. The basic structures of these two devices are similar. An electrothermal V-shaped actuator, which is made of doped polysilicon, is employed to provide the in-plane motion of the

switch. The doped-polysilicon contact head, signal lines, and their sidewalls are coated with a gold lm. Depending on the design of actuation direction, as shown in Fig. 2, the V-shaped actuator pushes or pulls the contact head and connects the RF signal lines via sidewall contacts to turn on the switch. The main improvement of the proposed microswitch is the isolation structure between the actuator and contact head, as compared in Fig. 1(c) and (d). As shown in Fig. 1(d), the previously reported switch used a nitride connection as the isolation structure. In our experiments, it was found that the nitride connection was easy to break in switches with a pull-type actuation design. Alternatively, a push-type actuation design was employed in that work. However, the nitride connection still caused some mechanical failures and decreased the switchs yield and reliability in long-term operation. The problem was mainly due to the poor nitride-polysilicon adhesion. Section II-B simulates the stress prole on the nitride connection. In the pull-type switch, the nitride-polysilicon adhesion sustains tensile stress, as shown in

SHI et al.: LATERAL MEMS SWITCH UTILIZING UNDOPED POLYSILICON AS ISOLATION MATERIAL

1175

Fig. 3.

Simulated stress prole on the nitride connection: (a) pull-type actuation design and (b) push-type actuation design.

Fig. 3(a), which causes the peeling-off of the nitride lm. In the push-type switch, the adhesion sustains a compressive stress, as shown in Fig. 3(b), but the stress concentration on the corner of the adhesion interface contributes to the failures of the nitride connection. To handle this problem, the undoped-polysilicon connection is developed in the proposed switch, as shown in Fig. 1(c). The undoped-polysilicon connection is deposited with the doped-polysilicon actuator and the contact head in the same step of the low-pressure chemical vapor deposition (LPCVD) process. Therefore, the homogeneous connection structure has robust mechanical strength. It could sustain a large contact force, which is benecial for reducing the contact resistance. Besides, the undoped-polysilicon connection provides a planar surface, which is important for some other applications where succeeding lithography steps are necessary. The electrical isolation performance of the undoped-polysilicon connection is also sufcient as investigated in our experiments. Due to the improvement of the isolation structure, the pulltype actuation design can be employed in the proposed microswitch. Comparing with the push-type design, the pull-type design has the advantage of good mechanical stability, as shown in Fig. 2. The arrows in the beams indicate the direction of the sustained stress during turning on the switch. In the push-type design, the cross joint sustains compressive stresses in both the x-axis direction and y-axis direction, so it is easy to buckle out of plane when the compressive stress is increased [18]. Contrarily, the cross joint of the pull-type design sustains tensile stresses in the y-axis direction, which makes the structure more stable. Therefore, a larger contact force could be reached without buckling. B. Isolation Design and Simulation As the actuator of the proposed switch is driven by the Joule heating, the isolation structure also has to serve as the thermal isolation between the actuator and the contact head. The undoped polysilicon has a low thermal conductivity of about 13.8 W/m K due to its phonon scattering effect at grain boundaries [19], [20], which is comparable with the silicon nitride (1530 W/m K).

Fig. 4. Mesh model used in both the temperature prole simulation and the contact force simulation. TABLE I PARAMETERS USED IN THE COVENTORWARE SIMULATION ANALYSIS

TABLE II THERMAL BOUNDARY CONDITION USED IN THE SIMULATIONS

As the microswitch is suspended only a few micrometers above the substrate, the heat conduction toward the substrate contributes to an important factor of determining the heat loss on the connection structure [21]. Therefore, the undopedpolysilicon structures with different isolation patterns are employed to enhance the thermal isolation. Their thermal isolation performances are simulated in CoventorWare 2005. The mesh model of the simulation is shown in Fig. 4. The meshing element is the 27-node hexahedral. The element

1176

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 16, NO. 5, OCTOBER 2007

Fig. 5. Isolation structure designs of the undoped-polysilicon connection: (a) type I, (b) type II, and (c) type III.

size is 1 m in the planar directions (x-axis and y-axis) and 0.2 m in the extruding direction (z-axis). The physical constants that are used in the analysis are listed in Table I, and the following analysis does not consider the dependence of material properties on the temperature. Table II summarizes the thermal boundary conditions in the simulation. As the vertical spacing between the switch and the substrate z is only 2 m, the conduction heat transfer coefcient U to the substrate can be approximated by [21] U = kair /z (1)

previous switch has a length of only 6 m, and the according temperature prole is shown in Fig. 7(b), which shows that the thermal isolation performance is not as good as the proposed three types of undoped-polysilicon connection. In summary, the robust undoped-polysilicon connection enables longer and more complex isolation structures than the nitride connection, and both factors contribute to the improved thermal isolation performance. C. Actuator Design Consideration There have been reports on the static performances of the V-shaped actuator such as the maximum motion, the contact force, and the lifetime [17], [22], [23]. There are also reports on the dynamic performances of the actuator, such as studying the electrothermal responses with the line-shape microstructure [24] and simulating the time responses of the thermal beams [21]. The design of the V-shaped actuator is a complex tradeoff among the displacement, the contact force, and the mechanical stability [17], [25], [26]. Due to the stable pull-type actuation design, the optimized actuator with longer beam length and fewer beam number can be employed in the proposed microswitch without leading to the out-of-plane buckling. The schematic view of the V-shaped actuator in our switch is shown in Fig. 8. For the V-shaped beams of different lengths, the longer beam requires a lower heating temperature to provide the same displacement [17], which means that a lower operation temperature is sufcient to turn on the switch. So, a 400-mlong actuator is employed here instead of the 200-m-long actuator in the previous report. The power consumption is also an important consideration in the thermal actuated switch. Therefore, the three-beam design is employed here instead of the previous six-beam design to reduce the power consumption. The nal actuator design has dimensions of 400-m length, 4-m width, 2-m thickness, and 7-m offset at the center. The spacing between the V-shaped beams is 9 m. The gap height between the actuator and the substrate is 2 m, and there is an initial 3-m gap between the contact head and the signal lines. Furthermore, the contact forces that are provided by the actuator under different driving voltages are simulated in

where kair is the air thermal conductivity. Using a kair of 0.03 W m1 K1 , U is calculated to be 15 000 W m2 K1 [21]. The actuation voltage of the switch is applied between the left side and the right side of the actuator beams. Three different types of isolation structure designs are considered, as shown in Fig. 5. During the ON-state of the switch, the simulated temperature prole on the actuator, the undopedpolysilicon connection, and the contact head are shown in Fig. 6, where a scaling factor of 5.0 is used to magnify the deformation of the V-shaped beams. The results show that a sufcient temperature drop along the undoped-polysilicon connection can be achieved in all the three designs. As marked in Fig. 6, the V-shaped beams temperatures in the type-II design (735 K) and the type-III design (733 K) are slightly lower than the beams temperature in the type-I design (763 K). The type-III design provides the best thermal isolation performance with a temperature of about 352 K in the contact head. These results are obtained from static thermal analysis. Fig. 7(a) shows the temperature prole of a switch with the nitride connection. The thickness of the nitride lm is 0.6 m, and its planar dimensions (50-m length and 8-m width) are the same as the undoped-polysilicon connection in the type-I design. Comparing Fig. 7(a) with Fig. 6(a), it shows that the nitride connection has better thermal isolation performance than the undoped-polysilicon connection. However, a 50-m-length nitride lm would greatly weaken the mechanical strength of the switch. In fact, the nitride connection that is used in our

SHI et al.: LATERAL MEMS SWITCH UTILIZING UNDOPED POLYSILICON AS ISOLATION MATERIAL

1177

Fig. 6. Simulated temperature prole on the microswitch with different connection structure designs, i.e., (a) type I, (b) type II, and (c) type III, when the substrate temperature is set to 300 K.

CoventorWare with the mesh model and boundary condition of Section II-B. As shown in Fig. 9, the contact head moves across the 3-m gap and reaches the signal lines at a driving voltage between 8 and 9 V, which matches with our experiment well. Under a 12-V driving voltage, the actuator could provide a contact force of about 310 N, which is sufcient to achieve a stable contact with the resistance in the range of 100300 m [27]. The time for heating up or cooling down the V-shaped actuator can be esteemed by the rst mode thermal time constant [24], [26], [28]. In our design, the actuators thermal time constant is calculated to be 67 s, which indicates a maximum cutoff frequency of 14.9 kHz. Meanwhile, the mechanical resonant frequency of the microswitch is simulated to be 376 kHz in the y-axis direction, which is sufciently higher than the thermal cutoff frequency of 14.9 kHz. The microswitch that is proposed in this report occupied an area of approximately 500 m 100 m. The RF signal lines are separated by 40 m, and the contact area is designed to be

10 m 2 m or 14 m 2 m. The closing gap between the contact head and the signal lines is 4 m. After sputtering a gold layer with 0.5-m thickness, the gap distance is reduced to be about 3 m. III. F ABRICATION The microswitch is fabricated with the polysilicon surfacemicromachined process, and a silicon wafer ( = 24 cm) with a 0.3-m-thick thermal oxide layer is used as the substrate. Fig. 10 shows the cross-sectional schematic view of the process. First, a 0.2-m-thick Si3 N4 lm is deposited on the substrate by the LPCVD process. The Si3 N4 lm and the thermal oxide lm together form an insulation layer to reduce the substrate parasitics at high frequencies, which is due to the lossy nature of the silicon substrate. Then, a sacricial layer of 2-m-thick LPCVD SiO2 lm is deposited and patterned to form the anchor position, as shown in Fig. 10(b). The sacricial SiO2 layer remains undoped, which is different from the usual doped SiO2

1178

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 16, NO. 5, OCTOBER 2007

Fig. 7. Temperature prole on the switches with different nitride connection structures: (a) 50- and (b) 6-m-long nitride connections.

Fig. 8. Schematic view of the three-beam V-shaped actuator in the proposed switch.

Fig. 9. Simulated contact force versus the input power. The corresponding actuation voltages are indicated at selected points.

sacricial layer, such as borosilicate glass or phosphosilicate glass, to prevent the dopant diffusing from the sacricial layer into the polysilicon structures. Second, a 2-m-thick undoped LPCVD polysilicon lm is deposited at 610 C, and a 2-m-thick photoresist (Shipley 6818) lm is coated, patterned, and baked as the shelter layer for the following implantation. As shown in Fig. 10(d), the polysilicon lm is implanted with the P+ at a dosage of 1 1016 cm2 and an energy of 80 keV, while the patterned photoresist layer keeps the isolation area undoped. The undoped area forms the undoped-polysilicon isolation structure in the followed sequences. After a 0.5-m-thick LPCVD oxide layer is deposited to avoid the self-diffusion effect, the thermal annealing (1050 C, 1 h) is carried out to drive and activate the dopant. The annealing also contributes to reduce the residual stress in the polysilicon lm. Then, the top oxide layer is removed by the buffered HF acid (BHF) etching, and the polysilicon lm is patterned with inductively coupled plasma etch process, as shown in Fig. 10(e). In this step, the undoped area forms the isolation structure, while the doped area forms the actuator, the contact head, and the signal lines. Third, the method of partial release combined with liftoff process is employed to form the sidewall metal contacts and the metal on signal lines [3]. As shown in Fig. 10(f), a partial release step is performed by dipping the wafer into 6 : 1 BHF while exposing only the small region between the contact head and the RF signal lines. Approximately 1.0 m of the sacricial oxide in the gap area is removed to ensure the separation of sputtered gold on the contact sidewalls between the contact head and the signal lines. Then, the liftoff process was carried

SHI et al.: LATERAL MEMS SWITCH UTILIZING UNDOPED POLYSILICON AS ISOLATION MATERIAL

1179

Fig. 12. (a) Schematic view and (b) SEM picture of the on-chip testing structure for measuring the electrical isolation performance of the undopedpolysilicon connection.

Fig. 10. Cross-sectional illustration of the fabrication process sequence. (a) Deposit the Si3 N4 insulation layer. (b) Deposit and pattern the undoped SiO2 sacricial layer. (c) Deposit the undoped polysilicon and form the photoresist shelter. (d) Implant P+ into the polysilicon with the photoresist shelter. (e) Pattern the polysilicon structure after thermal annealing. (f) Partially release the sacricial layer between the contacts. (g) Sputter and lift off the gold lm to form the sidewall contacts. (h) Release the whole microswitch.

the thin actuator beams. Fig. 11(a) shows the scanning electron microscope (SEM) picture of the whole fabricated microswitch with the transmission line, and Fig. 11(b) shows a close-up view. The whole fabrication sequence is completed by standard MEMS processes with only four masks including the liftoff. The undoped-polysilicon connection is realized by one step of sheltered implantation without extra process. IV. T EST AND D ISCUSSION A. Undoped-Polysilicon Isolation Structure The undoped-polysilicon connection has good mechanical strength and could sustain a large contact force, either in the pull-type actuation design or in the push-type actuation design. In our experiments, the undoped-polysilicon connection would not break even when the V-shaped beams were burnt by Joule heating under an 18-V driving voltage, which corresponds to a contact force of higher than 500 N, as shown in Fig. 9. The electrical isolation performance of the undopedpolysilicon connection has been investigated with the on-chip testing structures, as shown in Fig. 12. The testing structures are 8-m-wide 2-m-thick 200-m-long doped-polysilicon bridges with the undoped-polysilicon connection at the center. The structures IV curves are measured with an HP series 4156B parameter analyzer. Fig. 13 shows that the undopedpolysilicon connection provides good electrical isolation in the switchs operating range of 15 V. The 10-, 20-, and 40-mlong connections provide the isolation with a current leakage of no more than 3 nA under 15-V voltage, which is sufcient for a thermal actuator. The 20- and 40-m-long connections current leakage remains no more than 5 nA when the voltage rises to 40 V. The 5-m-long connection, as shown in Fig. 14, is insufcient for the isolation because the thermal annealing process causes the lateral diffusion of the implanted P+ dopant. Fig. 15 illustrates the IV curves of the 40-m-long connection under different heating temperatures. The results show that the current leakage of the connection increases with the rise of

Fig. 11. SEM pictures of (a) the fabricated microswitch with transmission line and (b) a close-up view of the switch.

out. A 0.5-m-thick gold lm is sputtered and lifted off to leave gold only on the contact sidewalls and signal routing lines. Finally, the device is fully released in the concentrated HF acid for 15 min, as shown in Fig. 10(h). The sublimation drying after the HF release is employed to reduce the surface stiction of

1180

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 16, NO. 5, OCTOBER 2007

Fig. 13. IV curves of the undoped-polysilicon connections with three different lengths: 10, 20, and 40 m.

Fig. 15. IV curves of the undoped-polysilicon connections under different temperatures.

diagnose the residual stress. As shown in Fig. 17, the deection of the indicator is small. This indicates that the residual stress of the doped polysilicon is also ignorable, and no initial displacement of the contact head is observed in the released microswitch. B. DC Testing In the dc testing of the microswitch, a driving voltage of about 8.09.0 V is required to achieve a 3-m in-plane displacement. The switches start to be turned on under a driving voltage of about 11.0 V. In our experiments, a driving voltage of 12 V is used to obtain a stable metal contact with low contact resistance. The driving voltage of 12 V is higher than the previously reported 3 V. It is mainly due to the low doping concentration (7 1018 cm3 ) of the doped-polysilicon actuator, and a lower operation voltage can be achieved with a higher doping level. The microswitch provides a small contact resistance of 0.42 at a driving input of 12 V/5 mA, corresponding to the power consumption of 60 mW. A control group using a sixbeam actuator requires 115-mW power consumption. The input power is reduced effectively by reducing the number of actuator beams. The RF signal line has a current handling capacity of 50 mA. It is measured by increasing the carrying current through the signal path while monitoring the frequency response of the switch. The switch fails to deliver signals properly after the carrying current was increased from 50 to 55 mA. The failure is due to the damage of the metal contact, which is caused by the gold welding [3]. A 100-Hz 50%-duty square-wave signal with a peak voltage of 12 V is used to measure the frequency responses of the switch. While the square-wave signal is applied in the actuator as the excitation source, an oscilloscope is monitoring the voltage output at one terminal of the signal path, while the other terminal is supplied with a 1-V dc voltage. Fig. 18 plots the rising-edge response time [Fig. 18(a)] and the falling-edge response time [Fig. 18(b)]. The microswitch needs, on average, 114 s to switch on and less than 13 s to switch off. The rising edge is much longer than the falling edge because the actuator needs to travel a distance of about 3 m before contact

Fig. 14. Comparing the IV curves of the on-chip testing structure with and without 5-m-long undoped-polysilicon connection.

the temperature. Therefore, a switch design with low actuation temperature is preferred when using the undoped-polysilicon connection. At 390-K temperature, the current leakage of 30 nA under 15 V voltage is still sufcient and the electrical isolation performance could be further improved by employing the optimized isolation structure design. It is well known that the residual stress plays important roles in MEMS structures. If the dopant causes any stress mismatch between the doped and undoped polysilicon, an out-of-plane deformation of the released structure may occur. Therefore, the released doped-polysilicon cantilevers, with and without the undoped-polysilicon connection, have been employed to investigate the possible stress mismatch. Compared to the cantilevers without undoped-polysilicon connection, as shown in Fig. 16, the cantilevers with undoped-polysilicon connection show no observable out-of-plane deformation. It seems that the stress mismatch can be ignored, and this is also veried in the fabricated microswitch. Besides, the residual stress of the doped polysilicon causes an in-plane displacement of the contact head after releasing. If the stress is too large, the contact head might connect the signal lines even when no voltage applies on the microswitch. The freestanding rotating indicator structure [29] is adopted to

SHI et al.: LATERAL MEMS SWITCH UTILIZING UNDOPED POLYSILICON AS ISOLATION MATERIAL

1181

Fig. 16. (a) Schematic view of cantilever used for testing the stress mismatch and (b) the SEM image of the cantilevers tips.

Fig. 17. Deection of the free-standing rotating indicator structure that is made by doped polysilicon.

for the 400-m-long design. However, two important factors must also be considered. First, the electrothermal properties of the polysilicon depend on the temperature, particularly, when considering the specic heat, which affects the thermal time response much [20]. To provide the same contact force of 300 N, as simulated by CoventorWare, the 400-m-long design needs an average temperature of 600 K on the V-shaped beams, while the 200-m-long design needs a much higher temperature of 1000 K. Second, the temperature distribution along the V-shaped beam is not uniform [21]. The parts of the V-shaped beam that are located at two sides near the anchors might response faster than the part at the center, and the 400-m-long design provides longer parts at two sides for deformation. Above all, the experiment results show that the switching time is obviously reduced by utilizing the 400-mlong actuator. C. RF Performance For RF applications, the lateral switch is more favorable in microstrip circuits. For measurement convenience, however, the coplanar waveguide (CPW) transmission line is utilized to characterize the RF performances of the switch in this paper. Due to the lateral dimension of the switch, it is difcult to employ a standard 50- CPW line. Therefore, a nonstandard CPW transmission line is utilized for the switch circuit. The transmission line layout of the proposed switch is shown in Fig. 11(b), where the RF performances are not optimized. The insertion loss that is induced by the transmission line is eliminated by utilizing a de-embeded structure [3].

with the signal lines, while opening the microswitch is only separating the contact head and the signal lines by removing the input power [3]. The maximum operating frequency of the microswitch is 4.5 kHz. It is measured by monitoring the output signal when increasing the exciting signal frequency until the microswitch fails to deliver signal properly. Comparing with the previous report, the switching time is reduced much, from about 290 to 130 s, while the maximum operating frequency is increased from 2.5 to 4.5 kHz. The improvement is mainly due to utilization of the 400-mlong beam actuator instead of the 200-m-long design. At a temperature of 300 K, the calculated thermal time constant of 43 s for the 200-m-long design is smaller than the 67 s

1182

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 16, NO. 5, OCTOBER 2007

Fig. 19. ON-state insertion loss of the microswitch at different driving voltages.

Fig. 20. Fig. 18. Measured frequency responses of the microswitch as (a) the risingedge response time and (b) the falling-edge response time. CH1 is the input signal of the excitation source. CH2 is the output signal of the switch.

OFF -state

isolation of the microswitch.

The on-wafer RF performance of the microswitch was measured with an HP series network analyzer (Model 8510C) and cascade groundsignalground microprobes with 150-m pitch. A standard shortopenloadthrough calibration kit was used, and the two-port S-parameters were measured from 100 MHz up to 20.1 GHz. During the on-state of the microswitch, the insertion loss is extracted by subtracting the measured throughline loss (without the microswitch) from the two-port S-parameter measurement of the microswitch. As shown in Fig. 19, the microswitch has a low insertion loss of 0.41 dB up to 20 GHz. Insertion loss of the same microswitch at different driving voltages have also been investigated. The higher driving voltage provides a smaller RF insertion loss because of the corresponding higher contact force, as shown in Fig. 9. No reliability problem occurs due to increasing the contact force, which caused the nitridepolysilicon adhesion failure in the previous work. Fig. 20 shows the OFF-state isolation of the microswitch, and an isolation of about 20 dB at 20 GHz can be achieved. The substrate leakage of the microswitch is slightly high mainly due

to the normal low-resistivity silicon wafer. Besides, the thermal oxide insulation layer that is used in the proposed microswitch is only 0.3 m thick, whereas the oxide insulation layer that is used in the previous work was 1 m thick. Both the OFF-state isolation and ON-state insertion loss at high frequency can be further improved by employing the high-resistivity silicon wafer. D. Reliability Two dominant switch failure mechanisms have been observed in our experiments, including the contact degradation and the mechanical failure that is caused by the nitride connection. In our previously reported switch, the nitride-polysilicon adhesion could not sustain the tensile stress, which leads to the failure of nearly all the pull-type switches. In the push-type switches, the nitride connection could sustain a limited compressive stress, but its mechanical failures still contribute to an important factor of low yield and reliability problems in longterm operations. With the undoped-polysilicon connection, no mechanical failure of the proposed switch was observed among over 20 devices for reliability testing.

SHI et al.: LATERAL MEMS SWITCH UTILIZING UNDOPED POLYSILICON AS ISOLATION MATERIAL

1183

V. C ONCLUSION This paper has reported a modied thermal actuated lateral switch utilizing the undoped-polysilicon connection as the mechanical coupling and electrical isolation structure. The undoped-polysilicon isolation structure has the advantage of high reliability as its robust mechanical strength could sustain the large contact force necessary for a stable metal contact. Meanwhile, it provides promising electrical isolation performance. The proposed microswitch requires a driving voltage of 12 V and an input power of 60 mW. The switching time is measured to be 130 s, and a maximum operating frequency of 4.5 kHz is reached, which is nearly double of our previous report. An ON -state insertion loss of 0.41 dB and an OFF-state isolation of 20 dB at 20 GHz have been achieved on normal lowresistivity silicon substrate. Improved RF performances can be obtained by employing the high-resistivity silicon substrate. The proposed microswitch operates over one billion cycles without signicant degradation of the contact resistance, and the electrical isolation performance of the undoped-polysilicon connection shows no observable degradation. The simplicity of this four-mask fabrication process and the easy realization of the insulative mechanical coupling enable the further possibility of cascading different actuation approaches together into one device. The undoped isolation structure can also be used in other surface-micromachined polysilicon lateral switches.

Fig. 21. Change of the contact resistance versus switching cycles. TABLE III SUMMARY OF THE MICROSWITCH PERFORMANCE

In the proposed switch, the metal contact degradation plays the main role of the failure mechanism. A 1000-Hz 50%-duty square-wave signal with a peak voltage of 12 V is used as the actuation input in reliability tests. The results show that the device can operate over 1 109 cycles under cold switching. Fig. 21 depicts the change of contact resistance versus the number of operating cycles. No signicant degradation of the contact resistance was observed. The undoped-polysilicon connection could sustain a larger contact force, which leads to a more stable metal contact, contributing to the improvement of the switchs lifetime. However, the average lifetime of the devices is only about 0.6 106 in hot switching tests with a carrying current of about 10 mA on the signal path. The gold welding, process variation of sputtering, and surface roughness, which are in charge of the contact failure and affect the microswitch lifetime, have been investigated before [3], [30]. During the long-term operation, the undoped-polysilicon connection sustains a long-term thermal annealing due to the Joule heating in the actuator beams, as shown in Fig. 6. It might cause slow self-diffusion of the dopant from the dopedpolysilicon actuator into the undoped-polysilicon connection and affect the electrical isolation performance. The 20-mlong undoped-polysilicon connection is sufcient to avoid this effect as veried in our experiments. The connection structure shows no observable degradation of the electrical isolation performance after one billion cycles operation. The overall performances of the proposed microswitch are summarized in Table III.

ACKNOWLEDGMENT The authors would like to thank the staff at the National Key Laboratory of Nano/Micro Fabrication Technology, Peking University, for their help in the fabrication process. They would also like to thank Y. Tan, X. Wang, and X. Ji for their help in the simulation and for fruitful discussions. R EFERENCES
[1] E. J. J. Kruglick and K. S. J. Pister, Lateral MEMS microcontact considerations, J. Microelectromech. Syst., vol. 8, no. 3, pp. 24271, Sep. 1999. [2] Y. Wang, Z. Li, D. T. McCormick, and N. C. Tien, Low-voltage lateralcontact microrelays for RF applications, in Proc. 15th IEEE Int. Conf. Micro-Electro-Mechanical Syst., Jan. 2002, pp. 645648. [3] Y. Wang, Z. Li, D. T. McCormick, and N. C. Tien, A low-voltage lateral MEMS switch with high RF performance, J. Microelectromech. Syst., vol. 13, no. 6, pp. 902911, Dec. 2004. [4] R. W. Moseley, E. M. Yeatman, A. S. Holmes, R. R. A. Syms, A. P. Finlay, and P. Boniface, Laterally actuated, low voltage, 3-port RF MEMS switch, in Proc. 19th IEEE Int. Conf. Micro-Electro-Mech. Syst., Jan. 2006, pp. 878881. [5] Z. Li, D. Zhang, T. Li, W. Wang, and G. Wu, Bulk micromachined relay with lateral contact, J. Micromech. Microeng., vol. 10, no. 3, pp. 329333, 2000. [6] R. Wood, R. Mahadevan, V. Dudley, A. Cowen, E. Hill, and K. Markus, MEMS microrelays, Mechatronics, vol. 8, no. 5, pp. 535 547, Aug. 1998. [7] J. Simon, S. Saffer, F. Sherman, and C.-J. Kim, Lateral polysilicon microrelays with a mercury microdrop contact, IEEE Trans. Ind. Electron., vol. 45, no. 6, pp. 854860, Dec. 1998. [8] A. Q. Liu, M. Tang, A. Agarwal, and A. Alphones, Low-loss lateral micromachined switches for high frequency applications, J. Micromech. Microeng., vol. 15, no. 1, pp. 157167, Jan. 2005.

1184

JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 16, NO. 5, OCTOBER 2007

[9] Y. Shi and S. G. Kim, A lateral, self-cleaning, direct contact MEMS switch, in Proc. 18th IEEE Int. Conf. Micro-Electro-Mech. Syst., Jan. 2005, pp. 195198. [10] R. L. Borwick, III, P. A. Stupar, and J. DeNatale, A hybrid approach to low-voltage MEMS switches, in Proc. 12th IEEE Int. Conf. Solid-State Sensors, Actuators Microsyst. (Transducers), Jun. 2003, pp. 859862. [11] Y. Wang, Z. Li, D. T. McCormick, and N. C. Tien, A micromachined RF microrelay with electrothermal actuation, Sens. Actuators A, Phys., vol. 103, no. 1, pp. 231236, Jan. 2003. [12] R. Chan, R. Lesnick, D. Becher, and M. Feng, Low-actuation voltage RF MEMS shunt switch with cold switching lifetime of seven billion cycles, J. Microelectromech. Syst., vol. 12, no. 5, pp. 713719, Oct. 2003. [13] S. D. Lee, B. C. Jun, S. D. Kim, H. C. Park, J. K. Rhee, and K. Mizuno, An RFMEMS switch with low-actuation voltage and high reliability, J. Microelectromech. Syst., vol. 15, no. 6, pp. 16051611, Dec. 2006. [14] G. M. Rebeiz, RF MEMS: Theory, Design and Technology. Hoboken, NJ: Wiley, 2003, pp. 148149. [15] T. J. Brosnihan, J. M. Bustillo, A. P. Pisano, and R. T. Howe, Embedded interconnect and electrical isolation for high-aspect ratio, SOI inertial instruments, in Proc. 9th IEEE Int. Conf. Solid-State Sensors Actuators (Transducers), Jun. 1999, pp. 10021005. [16] L. Muller, J. M. Heck, R. T. Howe, and A. P. Pisano, Electrical isolation process for molded, high-aspect-ratio polysilicon microstructures, in Proc. 13th IEEE Int. Conf. Micro-Electro-Mech. Syst., Jan. 2000, pp. 590595. [17] L. Que, J.-S. Park, and Y. B. Gianchandani, Bent-beam electrothermal actuatorsPart I: Single beam and cascaded devices, J. Microelectromech. Syst., vol. 10, no. 2, pp. 247254, Jun. 2001. [18] M. Chiao and L. Lin, Self-buckling of micromachined beams under resistive heating, J. Microelectromech. Syst., vol. 9, no. 1, pp. 146151, Mar. 2000. [19] A. D. McConnell, S. Uma, and K. E. Goodson, Thermal conductivity of doped polysilicon layers, J. Microelectromech. Syst., vol. 10, no. 3, pp. 360369, Sep. 2001. [20] S. Uma, A. D. McConnell, M. Asheghi, K. Kurabayashi, and K. E. Goodson, Temperature dependent thermal conductivity of undoped polycrystalline silicon, Int. J. Thermophys., vol. 22, pp. 605616, 2001. [21] R. Hickey, D. Sameoto, T. Hubbard, and M. Kujath, Time and frequency response of two-arm micromachined thermal actuators, J. Micromech. Microeng., vol. 13, no. 1, pp. 4046, 2003. [22] Y. B. Gianchandani and K. Naja, Bent-beam strain sensors, J. Microelectromech. Syst., vol. 5, no. 1, pp. 5258, Mar. 1996. [23] L. L. Chu, L. Que, A. D. Oliver, and Y. B. Gianchandani, Lifetime studies of electrothermal bent-beam actuators in single-crystal silicon and polysilicon, J. Microelectromech. Syst., vol. 15, no. 3, pp. 498506, Jun. 2006. [24] L. Lin and M. Chiao, Electrothermal response of lineshape microstructures, Sens. Actuators A, Phys., vol. 55, no. 1, pp. 3541, Jul. 1996. [25] E. T. Enikov, S. S. Kedar, and K. V. Lazarov, Analytical model for analysis and design of v-shaped thermal microactuators, J. Microelectromech. Syst., vol. 14, no. 4, pp. 788798, Aug. 2005. [26] J. M. Maloney, D. S. Schreiber, and D. L. DoVoe, Large-force electrothermal linear micromotors, J. Micromech. Microeng., vol. 14, no. 2, pp. 226234, Feb. 2004. [27] D. Hyman and M. Mehregany, Contact physics of gold microcontacts for MEMS switches, IEEE Trans. Compon. Packag. Technol., vol. 22, no. 3, pp. 357364, Sep. 1999. [28] D. S. Schreiber, W. J. Cheng, J. M. Maloney, and D. L. DoVoe, Surface micromachined electrothermal V-beam micromotors, in Proc. ASME Int. Mech. Eng. Congr. Expo., 2001, pp. 17.

[29] B. P. V. Dieenhuizen, J. F. L. Goosen, P. J. French, and R. F. Wolffenbuttel, Comparison of techniques for measuring both compressive and tensile stress in thin lm, Sens. Actuators A, Phys., vol. 37/38, pp. 756765, 1993. [30] N. E. McGruer, G. G. Adams, L. Chen, Z. J. Guo, and Y. Du, Mechanical, thermal, and material inuences on ohmic-contact-type MEMS switch operation, in Proc. 19th IEEE Int. Conf. Micro-Electro-Mech. Syst., Jan. 2006, pp. 230233.

Wendian Shi was born in Zhejiang Province, China, in 1983. He received the B.S. degree from Peking University, Beijing, China, in 2004. He is currently working toward the M.S. degree at the Department of Microelectronics, Peking University. His research interests include design and fabrication of microelectromechanical systems (MEMS), in particular, RF MEMS and Bio MEMS.

Norman C. Tien (S87M89) received the B.S. degree from the University of California, Berkeley, the M.S. degree from the University of Illinois, Urbana Champaign, and the Ph.D. degree from the University of California, San Diego. He was a Professor and the Chair of the Department of Electrical and Computer Engineering, University of California, Davis, and held a joint appointment as Professor of electrical engineering and computer science at the University of California, Berkeley. He also served as a Codirector of the Berkeley Sensor and Actuator Center. In 2006, he joined the Department of Electrical Engineering and Computer Science, Case School of Engineering, Case Western Reserve University, Cleveland, OH, where he is currently the Nord Professor of Engineering and the Dean of the Case School of Engineering. His research interests are micro- and nanotechnology, in particular, applications in wireless communications, biomedical systems, and environmental monitoring. Dr. Tien is the Ohio Eminent Scholar in Condensed Matter Physics. He was the recipient of a National Science Foundation CARRER Award.

Zhihong Li (M02) received the B.S. degree and the Ph.D. degree, majoring in VLSI technology and reliability, from Peking University, Beijing, China, in 1992 and 1997, respectively. He then joined the MEMS Group, Department of Microelectronics, Peking University. From 2000 to 2004, he was a Postdoctoral Researcher at Cornell University, Ithaca, NY, and the University of California, Davis. He is currently a Professor and the Director of the MEMS Center, Department of Microelectronics, Peking University. His research interests include design and fabrication of microelectromechanical systems (MEMS), in particular, RF MEMS and Bio MEMS.

You might also like