8/3/2019 Lisec 2004

1/4

Dielectric Material Impact on Capacitive RFMEMS ReliabilityT. Lisec,C .Huth and B. Wagner

Fraunhofer Institute for Silicon Technology (ISIT), Fraunhoferstr. 1,2 552 4 Itzehoe, Germany

Abstract - The influ ence of di fferen t types of dielectricson the switching behaviour and reliability of capacitiveRF MEMS switches fabricated by metal surface-micromachining is investigated. Sputtered AIN layers arecompared to PECVD Si,N4 and TalOs layers. It has beenfound that switches with sputtered AIN can he operatedwithout failure in a wide range of driving conditions.Besides the dielectric charging problem anotherdegradation has heen observed independent of thedielectric material. The effec t only occurs for operation inambient air and is probably caused by an electrochemicalreaction.

I. In troduct ionMEMS based switches have emerged as a seriousalternative to solid state GaAs or Si based devices inmicrowave applications. A lot of devices have beenalready presented hut no mass product is on the marketup to now. Packaging and reliability issues are themajor limiting factors [I].The packaging problem results from the fact that thestructural layers are usually made from highlyconductive metals (Au, Al) to meet low lossrequirements. Since the released structures easilydegrade at elevated temperatures standard wafer-levelpackaging requiring 300400C is not possible. Up tonow, conventional housing of single chips is thecommon way. However, this is expensive due to theproblems conceming the separation of wafers with freestanding MEMS into individual units and the handlingof them.The reliability of capacitive RF MEMS switches ismainly restricted by the charging of the dielectricmaterial in the active area. Up to now, the standarddielectrics is PECVD Si,N4 (PE nitride). It has adielectric constant of 6 i 7 and could he easily depositedat low temperatures. The main drawback is the strongcharging affinity. Nevertheless, switches with PEnitride have been tested for billions of cycles withoutfailure. It has been shown, that the total time themembrane is in the down position is the criticalparameter and not the num ber of cycles to failure [2].To minimise the charging at first the voltage applied tohold the membrane in the down position should be aslow as possible. A further improvement could heobtained with an alternating voltage. However, sinceinjection and removal mechanisms are not equal thecharging in PE nitride cannot be suppressed completelyPI.

As an alternative, PECVD S O , or LTO have beenused as dielectrics in capacitive RF switches. The oxideshows less charging and better breakdown stabilitycompared to PE nitride. However, the dielectricconstant k is much lower (typ. 2 4 ) which results in asmall capacitance ratio and consequently in aninsufficient performance at low GHz frequencies. Fromthis point of view materials with a high k seem to hemore attractive for RF MEMS. In [4] a capacitanceratio of 166 have been reached for a switch withsputtered BST (k =20). In the same paper for PECVDA1203 a ratio of 22 is reported (k =9-10). Thediscrepancy to the theoretical value of 99 is explainedby the roughness of the A120, ayer. Probably for thesame reason in [ 5 ] a ratio of only 13 has been obtainedfor devices with anodically deposited T a2 0s (k =2 5 ~ 3 0 ) .n both papers no results ahout the reliabilityof the switches is given.It is important that the dielectric layers can heproduced with high quality in advanced semiconductorequipmen t. Ta2Os is promising a s dielectrics forcapacitors in DRAM S while AIN is used for BA Wresonators. During the last years sputtering processeshave been developed by several suppliers ofsemiconductor equipment. In this paper AlN, Ta20s aswell as PE nitride layers have heen integrated in thecapacitive RF-MEMS switch process at FraunhoferISIT [6]. The reliability of the devices is compared.

U. Device descriptionFig. I presents a cross-sectional view of the surface-micromachined electrostatic switch and a photo of afinished device. The membrane consists of anevaporated Ni layer (about I pm thick), the spring-typeanchoring of thick electroplated Ni (about 16 pm). Thisprovides a high design flexibility allowing theimplementation of special structures for acompensation of thermally induced and intrinsicstresses. A more detailed description of design aspectscan he found in [6]. Signal line and ground area belowthe membrane are formed from an TaiPtlAu/Pt stack.Below the bridge the signal line is covered by one ofthe dielectric layers under investigation, AIN, Ta2OSorPE nitride. In case of AIN, the sputter depositionprocess was performed at Unaxis, Liechtenstein. Toavoid high control voltages on the signal line twoseparated actuation electrodes with an additional PEnitride isolation (dielectric 2) are embedded in theground area. The two Ni bars on top of the membrane

34th European Microwave Conference - Amsterdam, 2004 73

8/3/2019 Lisec 2004

2/4

should ensure a good flamess in the active area of theswitch. The lines outside the switching area and allother interconnections consist of 2 pm thickelectroplated A u. Sputtered T a2 05 layers were onlyintegrated in a process run with a previous switchdesign without actuation electrodes.

membrane I dielecuic2

Parameter SilN, ANlayer thickness (nm) 100 200

air gap (W 2.5 2.4

Si/signal line achlation electrode

Taros2002.6

Fig. 1. Cross-sectional iew ofthe switch and photo of a completelyprwessed device. I he membrane is 500 pm long and lypically 100pm wide.

Con ff)111. Test set up

All tests were performed on wafer level on a manualprobe station at laboratory conditions in ambient air orin a probe chamber at dry N2 atmosphere. Forcapacitance measurements and the application of DCvoltages below 40 VDc a HP 4284A LC R meter w asused. Voltage pulses above 60 Vx were provided by aKeithley SMLJ236. CON nd C,, as well as pull-in andbreakdown voltages were obtained using the LCRmeter only. The DC control voltage UC applied to thesignal line was ramped from 0 to 40 V in steps of 1 V .Simultaneously the capacitance values were measured.All other experiments were performed using theactuation electrodes for switch operation powered bythe SMU. Th e membrane was hold on ground to avoidan interference between LCR meter and SMU. Afterthe DC control voltage Uc was applied to the signalline the membrane was pulled down by a short 6 0 4 0VDc pulse on the actuation electrodes. Once in thedown position the membrane should be held there byUc for a predefined duration. For the tested switches aminimum value of IO VDc was required. After turningoff the control voltage the membrane normally shouldmove back to the initial up position. Cycling wasexecuted usually at 1 Hz . To suppress the charging ofthe actuation electrodes the DC pulses were alternatedin sign after each actuation. Three capacitance values

75 79 67

were measured during a cycle. COfflwas recorded afierturning on U, while the memb rane is still up. Afterswitching down the membrane with the DC actuationpulse C,. was obtain ed. Afte r Uc was removed Conwas measured. At this moment the membrane normallyshould be back in the up position.

Cm (PF)capacitance ratiobreakdown voltage (V)2.32 1.68 5.4531 21 61= 35 > 40 = 30

Charging behaviourFig. 2 illustrates the difference in the chargingbehaviour in air for identical switches with PE nitrideand AIN. In both cases the membrane was switcheddown altemating by a f 65 VD c pulse on the actuationelectrodes and hold with +20 VDc on the sign al line fo rabout 90 s. For the device with PE nitride thecapacitance decreases quickly after an initial peak andthe membrane releases spontaneously within a fewseconds although Uc is still present. Moreover, themembrane could be pulled down again without anyactuation pulse only by changing Uc in sign. For theswitch with AlN the membrane remains down until Ucis turned off. The capacitance has a very constantvalue. The switch can be operated repeatedly forvarious on-times under exactly the same switchmgconditions. The variation of C,, at the same Uc valuewith opposite sign is below 2 %.

74 34" European Microwave Conference - Amsterdam, 2004

8/3/2019 Lisec 2004

3/4

I I0 f ' i o 40 ' $0 ' so ' f l o oon l ime (ref) off

Fig. 2. Typical behaviour of he switch capacitance for identicaldeviceswith AlN and PE nit ide operated with a positive Dc controlvoltage.The behaviour always remains the same for bothmaterials. Even at 10 Voc the membrane of a switchwith PE nitride spontaneously releases after ahout 1min. This has never been ohselved for devices withAIN. At 40 Voc the memb rane remains down until thecontrol voltage is turned off.But also for AIN layers charging effects have beenmeasured. During the test illustrated by Fig. 3 Uc wasincreased from + I 5 to +30 Vo c in 1 V steps while theswitch was cycled with 1 Hz in air. At about 22 V C,,starts to oscillate. In cycles with positive actuationpulse the membr ane sticks temporary after U, is turnedoff and releases when Uc is applied again. At 25 V thetemporary sticking occur during every cycle,independent of voltage polarity. Although this is acharge induced effect it should not be related only tothe AIN in the active area of the switch since all thetime Corn emains low. Partially the temporary stickingmay result from the charging of the PE nitride on top ofthe actuation electrodes. At 29 V Cam ncreases as welldu e to a permanent sticking of the membrane.

CL 1.2

.@ 0. 8t0.4T-0.0 l ime@eo)

Fig. 3.at different positive U, YIIIUIS.Note, that the tested device correctly switches againafter Uc is reduced to 15 V. This happens reproduciblefor switches with AIN hut n ever could be achieved fordevices with PE nitride.The smaller charging effects in AIN probably resultfrom a much higher structural quality. PE nitrideusually has a weakly packaged amorphous structurewith a high hydrogen content. The trap density istypically very high wluch is known to be the main

C.. and Con values for a switch with AIN cycled with I Hz

reason for charging [3]. In contrast, the sputtered AINlayers are characterised by densely packed columnargrains with an excellent orientation. The AIN layers aremuch smoother than the PE nitride layers. It isassumed, that the defect density in such AIN is prettylow. Together with the large energy gap of 6,2 eV thiscould explain the reduced charging affinity.Unfortunately, detailed charging experiments andcycling tests could not he perform ed for T azOs due tothe lack of switches with actuation electrodes.However, the impression is that the charging effect inTalOS is as strong as in PE nitride.Lifetime testsAlthough switches with PE nitride could not be helddown permanently by DC control voltages, cycling at 1Hz was possible. In general, the lifetime is higher forlower Uc, for altemating Uc polarity and for highercycling frequencies. In most cases the switches stickafter some 103+104 ycles with positive or negative Ucof 10-20 Voc, as shown in Fig. 4. But for singledevices at IO Voc with alternating polarity, I O 6 cycleshave been reached w ithout failure. Ohviously, at somespecial conditions during cycling a kind of equilibriumis established between charging and dischargingprocesses in the P E nitride.

2.:

I0 5 10 15 20

1000 CyclesFig. 4.Hzwith +I 2 Vac. AAer 17.000cycles sticking occurs increasingly.Typical behaviour of a switch with PE nibide cycled at I

Fig. 5 presents a similar test for a switch with AINcycled at 1 Hz using +12 Voc. Instead of C,, and CO*the capacitance difference AC is plotted as a functionof the number of cycles. The initial increase of ACwithin the first 1000 cycles was observed for alldevices independently from the dielectrics. Probably,the effect can be explained by a rising smoothness ofthe contact surfaces repeatedly pressed together withhigh force. Nevertheless, for the device in Fig. 5 amean value of 1,546 pF with a standard deviation of0,009 pF could be calculated for the whole data set.The total down time of over IO5 econds is a factor 100higher than the maximum values for PE nitridereported in [2]. The switch was still working withoutfailure when the reliability test was stopped after600.000 cycles.But, at higher positive Uc values a strong decrease ofAC occurs already after a few thousand cycles (Fig. 6) .

34" European Microwave Conference - Amsterdam,2004 75

8/3/2019 Lisec 2004

4/4

g 1.541.501.481.4U

dry nitrogen

ambient airIPU 0.5 0 10 1 201000 Cycles

Fig.5 .(200ms downtime)witha control voltage of + I 2 Vrr.Note, there is no contradiction to Fig. 3 where stictionoccurs even at 22 V. That particular device wasdesigned to have a lower membrane stifmess andconsequently a lower restoring force. Other switcheswith a stiffer membrane could be operated easily at upto 40 Voc without sticking. For switches with PEnitride the same effect bas been observed if the devicesurvived a sufficiently large number of cycles withoutsticking.

Typical behaviour of a switch with AIN cycled at 1 Hz

I

I . , I0 10 20 30 401000Cycler

Fig.6 .at different positive U , values in ambient air.Behaviour ofAC for switches with AIN cycled with I Hz

This degradation can not be explained by charging. Allstressed switches were still working without sticking.Th e effect is suppressed if U, has a negative polarity.A possible reason for the problem could he thecombination of noble metals with Ni. In the presenceof a moisture film connecting the metals in the downstate of the switch an electrochemical reaction canoccur at the contact surfaces if a sufficiently highvoltage with the "right" polarity is applied.This assumption is confirmed by the fact, that ACremains stable also for positive Uc if the switch isoperated in dry N2. Fig. 7 compares two similardevices from the same wafer cycled with 1Hz and anU, of +30 V,, at iden tica l con ditio ns in air and in drynitrogen. Although the graph shows only the firstcycles the device in N2 has been tested for about150000 cycles without degradation.Additional tests with different frequencies and downtimes per cycle show that the degradation in airdepends on the total down time of the membrane butnot on the number of cycles.

Fig. .with AIN tested under identical conditionsin air and in dryNI.AC versus the number of cycles for two similar switches

V. CONCLUSIONSPE nitride and sputtered AIN have been tested asdielectrics for capacitive RF MEMS switches.Although some devices with PE nitride at particularconditions were operated for lo 6cycles without failurethey could not be held reliable in the on-state due to thecharging of the dielectric. In contrast, switches withsputtered AlN can be operated repeatedly for anyrequired duration in a wide range o f conditions withoutcharge induced failures. A total down time of over 10'seconds could be achieved. This indicates thatsputtered AIN probably would allow to overcome thelcnown reliability problems of capacitive RF M E M Sswitches.A special degradation effect observed for both types ofswitches at higher control voltages in air is explainedby electrochemical reactions in the presence ofmoisture. The degradation disappear if the devices areoperated in dry nitrogen.

VI. ACKNOWLEDGMENTSThe authors would like to thank E. Kiigler fromUnaxis, Liechtenstein for his helpful suggestions andthe service with sputtered AIN layers. This work waspartially funded by the European Commission in theproject M ELOD ICT IST- 1999-10945.

VII. REFERENCES[I ] J. DeNatale, R. Mihailovich, "R F M E M S

reliability", Proc. IEEE Transducers 2003, Boston,[2 ] I. deWolf, "Reliability investigation of RF MEMS",

Proc. M EMSWAV E 2003, Toulouse[3] G . M. Rebeiz (ed.), "RF MEMS Theory, Designand Technology" , ohn Wiley & Sons, 2003[4] E. Berland et. al., "Dielectric Materials in MEMSswitches: a comparison between BST and A120,",

Proc. MEMS WAV E 2003, Toulouse[5] X. Rottenberg et al., "Neuartige kapazitive HF -MEMS-Schalter", Electronic Embedded Systeme04/03161 T. Lisec et. al., "Surface-micromachined capacitiveRF switches with high thermal stahility and lowdrift using Ni as structural material", Proc.MEM SWA VE 2004, Uppsala

pp. 943

76 34" European Microwave Conference - Amsterdam, 2004