MEMS for Reconfigurable Wide-Band RF ICs

Transcription

1 MEMS for Reconfigurable Wide-Band RF ICs A.M. Ionescu Swiss Federal Institute of Technology ausanne (EPF), Electronics aboratory (EG), EB-Ecublens, CH-1015 ausanne, Switzerland, Abstract This paper reports on the state-of-the-art RF MEMS for reconfigurable RF ICs. Design, fabrication and performance features are reported for various RF MEMS passive devices including contact and capacitive switches, SG-MOSFET, capacitors and inductors suitable for wide-band applications. General insights concerning programmable RF ICs architectures are presented. Finally, aging and some specific packaging solutions for RF MEMS are discussed. 1. Introduction The end of the 20 th century experienced major changes in the radio-frequency (RF) technology, [1]-[2]. The exponentially growing interests in mobile communications and networking of information have brought new challenges, with particular emphasis on more affordable and integrable technologies able to provide superior RF functionalities per unit of volume. Immediate impact has been observed on the research and development of silicon-based RF integrated circuits (ICs) with special emphasis on sub-micron silicon CMOS, SOI and SiGe heterojunction bipolar transistors (HBT s). New technologies have also emerged, stimulated by the potentially enormous markets. One of these is the micro-electro-mechanical-systems (MEMS) technology. Ironically, first developments in RF MEMS originated from the requirements in airborne systems, but nowadays it appears that the telecom mass market is the main driver of this field. Even though first MEMS devices were demonstrated in the late 1960 s [3], major developments for applications have had to wait for more than 30 years because of non-adapted technologies. Generally, MEMS can be considered as the merger of the IC world with the micron-scale mechanical world. Conventional IC fabrication techniques stand along with more exotic chemical and mechanical processes, in order to provide the MEMS structures. MEMS devices are extremely attractive for RF ICs (especially wireless) because of their major gains in terms of: (i) device and system miniaturization (resulting in lower costs), (ii) integration (same planar technology to manufacture integrated circuits), (iii) power savings (low power operated devices), (iv) higher performances (such as high-q for passive devices) and (v) new integrated functionalities (like programmable interconnects and programmable/tunable passive devices, resulting in programmable/tunable RF IC architectures). Generally, high performance RF MEMS have been developed mostly for narrow band applications. In our view, wideband performance requirements for MEMS, as necessitated by airborne applications, could positively impact the overall mobile communication applications in the near future. If one accepts that future mobile handsets should have multi-functions (be used as a classical mobile phone in two or three allowed bands and/or also as a AN terminal and/or as a computer network terminal) and, moreover, offer access to different service providers emitting in different RF/microwave bands, the various frequencies associated with these broadcasts will require the use of wideband internal functions at the front end level of the transceiver. This wide-band constraint will then, naturally extend to the MEMS device level. In this paper, typical MEMS solutions for RF switches, high-q capacitors and inductors with and without programmable and/or tunable characteristics together with their performances and use in basic RF IC blocks that benefit new functionalities, are reported. 2. MEMS passive devices 2.1 RF MEMS switches The basic building block of a RF MEMS circuit is the RF MEMS switch. It plays a similar role to that of a MOS-type switch in a standard RF system. In principle, there are two main switch categories: (a) contact and (b) capacitive (contact-less), as shown in Fig. 1. The capacitive switch architecture is very similar to a variable MEMS capacitor, with two stable states and a capacitance variation of orders of magnitudes between the two states. A contact switch behaves as a switch from DC to high frequency. In contrast, a capacitive switch is useful only at high frequency and has improved performance in terms of contact resistance degradation (reliability), sticking and power handling. MEMS switches have several clear advantages over their solidstate counterparts (FETs and PIN diodes): much lower series resistances, low power operation and negligible inter-modulation distortion [1]-[2]. In recent years, silicon micro-machining has been especially used to manufacture low-loss RF switches from 100MHz up to 50GHz, with around 1-10µs speed and DC operation in excess of 10V [4]-[7]. The C ON /C OFF capacitance ratios are typically around , with the maximal capacitance of the order of a few pf. Their dimensions are of the order of few 100µmx100µm with air gaps of a few of µm. These dimensions,

2 combined with the use of metals (Al and Cu) or polysilicon for the membranes, results in a switch mass of around 10-8 kg. The electrostatic actuation of a capacitive switch exploits the analytical formulation of the pull-in voltage: V PI 8K 3 = g (1) 0 27ε Ww 0 where K is the effective spring constant, W is the CPW conductor line width, w is the switch active membrane width and g 0 is the nominal gap height. In order to reduce V PI for sub-10v operation the following solutions are under investigation: the reduction of the spring constant K by a smart design of the switch arms and the reduction of the air gap g 0 up to a feasible limit, without paying penalty in terms of the switch return loss. The surface of the membrane should stay less than around 10 5 µm 2, which adds an extra constraint for a low-voltage design. Enhanced high frequency metrics essentially relate to the switch S-parameters [4], [6]: the insertion loss in on state (better then 0.2dB), the isolation (1/ S 21 ) in the off state (better than 20dB) and the return loss (1/ S 11 ) in the both states, are the main parameters. The equivalent circuit of a capacitive switch is depicted in Fig. 2, [8]. From the technological point of view, integration with CMOS active devices is presently achievable: this involves the use of materials (metals: Al, Cu, AlCu, and polysilicon) and of micromachining techniques, which are fully CMOS compatible (in- or post-process). Cantilever beam OFF ow loss substrate ow loss substrate Metal membrane Pull-down electrode g 0 (a) OFF ON ower electrode: Signal (CPW) Contact electrode Thin dielectric ON (b) Fig. 1 Schematics of a: (a) contact RF MEMS switch and (b) capacitive (contact-less) RF MEMS switch. 2.2 RF MEMS tunable/programmable capacitors The tunable MEMS capacitor is inspired by the capacitive switch architecture. The tuning range of a MEMS capacitor is severely limited by the snap-down event (non-equilibrium) which occurs at a distance of g 0 /3 (air gap dimension). This event limits the capacitor tuning range to a ratio of 1.5:1. In order to increase the tuning range, one proposed solution was to use a threeplate capacitor which increase the theoretical ratio to 2:1 [9]. However, the practical ratios are much more reduced (no tuning range in excess of 50% has been reported with these architectures). A new capacitor architecture, which exploits the use of two de-coupled lateral electrostatic tuning plates E and E ) has been recently proposed [10], as shown in Fig. 3. The condition for a quasi-infinite tuning is simply: d 2 /3>d 1. High-quality factors, Q, in excess of have been successfully demonstrated for present MEMS tunable capacitors. Based on the combination of both fixed and tunable MEMS capacitors, connected by capacitive switches (Fig. 4), the command of which is performed by binary signals, 2 n -bit-programmable capacitor banks with some continuos tuning features, can be imagined and realized. These capacitor banks are expected to provide new functionalities at RF circuit levels. One key step in the fabrication of a MEMS tunable capacitor is the membrane releasing by the wet etching of a sacrificial layer. Etch-holes with dimension of around 2x2µm 2 are usually used in order to facilitate this process. At EPF, we are developing a new membrane releasing process [11], based on the SF 6 dry etching of a- Si. Typical released Al-membranes using the EPF s dry process are reported in Figs. 5 a and b. d 2d1 E E E Fig. 3 Cross section of the RF MEMS capacitor [10] with increased tuning range: electrodes E and E are used for actuation purposes. Z 0 C Z 0 C 2C 2 2 C 2 n C Rs Fig. 4 Schematic of programmable and tunable capacitor bank that uses tunable and fixed value RF MEMS capacitors. Fig. 2 Equivalent-circuit model of the capacitive switch.

3 (a) formulation of a metal-metal switch architecture is no longer valid (higher V PI values are obtained). Based on electro-mechanical simulations and a new adapted analytical model [12], we have investigated for switching and tuning applications the SG-MOSFET architecture with metal-over-gate, thin oxide (<50nm) and low substrate doping. The overall switching capacitance characteristics reported in Figs. 6 and 7 suggest that: (I) excellent ratios of C on /C off for RF switches (>100) and interesting tuning ranges (>30%) with low voltage operation are easily achievable for thin oxides (<15nm) and (II) less than 5V switching operation is conditioned by low values of the spring constant <500N/m, which can be provided by a special hinge design, in case of A<10 4 µm 2. Hinge n+ W Gate (Al) b n+ (b) Fig. 4 Released AlSi-membranes for SG-MOSFET using the dry etching of sacrificial a-si. 2.3 The suspended-gate (SG) MOSFET The suspended-gate (SG) MOSFET combines in a single cell a MOSFET and a MEMS switch with a metal-over-gate architecture. A typical SG-MOSFET design [12]-[13] is given in Fig. 5a that shows also its 3D numerically simulated actuation under a uniformly distribute force of 1000 Pa. The cross section and the principle of the SG-MOSFET are depicted in Figs. 5 b and c. When the gate voltage, V g, is increased, the intrinsic gate-voltage, V gint, which drives the MOS channel formation, is tuned according to a capacitor divider, eq. (2), and the membrane moves continuously downwards as long as the equilibrium is maintained between electrostatic and elastic forces, eq. (3): Vg Vg int = (2) 1 + C / C kx ε gc int gap A(V V ) 2 1 air g gint = (3) 2 2 (t gap0 x) where x is the vertical gate displacement and C gcint, C gap are the intrinsic gate-to-channel capacitance of the underneath MOSFET and the air-gap capacitance, respectively. When V g equals the pull-in voltage, V PI, unstable equilibrium is reached and the switch (suspended membrane) moves from the off to the on state (t gap =0). It is worth noting that the use of a MOS device underneath a suspended membrane could positively impact its stabilization [14] and the V pi Al suspended (movable) gate source n+ x t ox t g0 (a) drain p-si n+ SOI Spring V g V gint V d (b) (c) Fig. 5 SG-MOSFET: a) architecture, b) cross section and c) equivalent circuit. Cgc (F/m 2 ) 10-2 tox(nm) = Switch OFF: Cgc = OW Fig. 6 SG-MOS extrinsic gate-to-channel capacitance, C gc, vs. gate voltage, V g, as predicted by the analytical modeling: C on /C off >100 for t ox <20nm. V s Switch ON: Cgc =HIGH k(n/m)=100 A=Wx(µmxµm)=100x100 tgap0(µm)= Vg (V)

4 C gc ( F/m 2 ) ~C ox k(n/m)= 500 t ox(nm)=10, t gap0(µm)=0.19 A=Wx(µmxµm)=100x Vg (V) Fig. 7 SG-MOS capacitance, C gc, vs. gate voltage, V g, for a thin oxide t ox =10nm and k as a parameter. 2.4 High-Q RF MEMS Inductors Telecom applications require MEMS inductance values in the range of 1-50nH with high quality factors (Q>10). Nowadays, high-q MEMS inductors can be achieved based on two main architectures: (i) the horizontal-plane suspended inductor architecture [15] (Fig. 8), which combines back-side silicon substrate dry etching and uses low-k dielectrics and (ii) the vertical plane or out-of-plane inductor architecture, similar to [16] (Fig. 9), which can be fabricated by the sacrificial etching of an insulator layer in embedded low-k/cu structures. Generally, the high-q is available for a narrow band (see Fig. 10) and more research efforts have to be dedicated in order to extend these performances over wide band (at least a frequency decade: 1-10GHz). In order to achieve high-q programmable/tunable inductor banks for multi-function at circuit level, the combination of high-q MEMS inductors with highperformance contact switches (relays) is needed, as suggested by Fig. 11. It is worth noting that this kind of architecture needs more innovative design efforts in order to preserve the high-q of the individual inductors and eventually integrate it in 3D IC architectures [17]. More accurate modeling and adapted electrical characterization techniques at high frequency, at device and interconnect levels, are expected to positively feedback the specific design. Some extra difficulties concerning the most adapted technology for such a bank of inductors originate in the preference for copper for high-q inductors and of more mature process using Al for the micro-relays. M2 via M1 Insulator ow loss substrate Fig. 9 Schematics of an out-of-plane inductor using twometal levels (M1 and M2) and one vias. The surface micro-machining of a core sacrificial layer provides the suspended structure. ow-k 2 M2 Via ow-k1 Via Via Underpass (M1) Si3N4 SiO2 Si-p Etched cavity Si-p M1 M2 a) GND pad Fig. 10 Simulation based prediction of the Q-factor of a horizontal plane suspended inductor (the design is not wide-band optimized): it appears that high performance (i.e. Q>10-20) is available for much less than one decade of frequency N b) p+ Fig. 8 a) Cross section of a suspended spiral inductor by dry etching of the underneath substrate and b) spiral inductor design (including p+ guard-ring). Fig. 11 Schematic of a programmable/tunable bank of inductors: the tuning can be achieved by a ferromagnetic core displacement or by coupling techniques.

5 3. Aging and Reliability of RF MEMS Despite all major progress made over the last several years, RF MEMS is not yet a mature technology and aging and reliability characterization and modeling works are still in progress, at both the material and device levels. One should note that MEMS reliability has particularly to deal with the existence of moving micromechanical parts that are not existing in traditional microelectronics and raise new issues. For instance, present contact switches are limited to some cycles without failure, while the requirement from some communication applications stand as high as cycles. Hence, alternative architectures or/and innovation at the switch design level are strongly required. An important source of inspiration for MEMS reliability advanced characterization could be the development of modern interconnects that use similar materials and architectures. 4. RF circuit blocks benefiting from MEMS In the following paragraph, some basic RF circuit blocks that benefit from the use of MEMS passive devices are presented. I. Voltage-Controlled Oscillators (VCOs) VCOs are used in phase-locked loops (P s) and frequency synthesizers to provide precise frequencies. The oscillator frequency must be stable against several various factors like: temperature, aging, noise, etc. The most important factor that dictates the oscillator stability is the Q-factor, the value of which has to be in excess of a threshold value. Another main RF requirement for oscillators is a low-phase noise [18], [19]. This is difficult to achieve with conventional ring oscillators and since the phase noise is inversely proportional to the quality factor, high-q MEMS inductors and capacitors are highly desirable for VCOs. Moreover, using tunable passive MEMS components provides VCOs with frequency tuning capability, as shown in Fig. 12. Frequency tuning ranges of around 10% and phase noise of 98dBc/Hz and 126dBc/Hz at 100kHz and 600kHz offsets from the 1.9GHz carrier have been recently reported [18]. It should also be noted that a trade-off exists between the low power and the low phase noise. V DD C C II. Filters Other important bricks of RF ICs that can benefit form high-q MEMS passive components and their tuning features are the RF passive filters [20] (such as the low insertion loss ladder passive filters, the schematic of which is shown in Fig. 13). Generally, banks of passive tunable MEMS devices are very useful to all types of programmable filters destined to mobile communication systems because of the new added functionality together with low power operation. The active RF MEMS filters [22] are not discussed here. Fig. 13 Schematic of a ladder filter based on tunable MEMS capacitors. III. Programmable Interconnects, Phase Shifters and TTDs Other attractive applications of the RF MEMS switches concern the programmable interconnects as a challenging alternative to CMOS crossbars. The approach proposed by Duewer et al [22] is to use contact-less switches and a programming scheme whereby arrays of switches are programmed with only X- and Y-controls, Fig. 14. The reported prototypes were fabricated with 200:1 capacitance ratios. The phase shifter application [1], can be addressed in a similar way: N different binary loops can be connected in series switches to provide 2 N possible electrical delays between input and output. The length of each loop is designed to make the 2 N delays equal to an integer multiple of the last significant delay plus a built-in offset delay, resulting in a discrete phase-shifter. Another fundamental brick of beam-forming networks, whether for passive or active antennas that can benefit from MEMS are the True Time Delay ines [23] (TTD). To allow the beam-forming in different spatial directions at different times, it is necessary to make the delay variable and, as previously explained, this can be done by using pieces of micro-strip transmission lines interconnected by RF switches. Variable TTD is more interesting than phase shifting lines because of its larger applicability and quasi-independent on the frequency. In Y-control R 0 I bias Fig. 12 Basic architecture of a tunable VCO that exploits high-q MEMS passive devices. X-control Capacitive switch Fig. 14 Cell of an array addressing scheme, [22]. Out

6 5. Packaging Issues Appropriate packaging for MEMS RF devices must comply with microscopic movable mechanical structures requirements in terms of local protection during the wafer processing, after the releasing step, and during dicing and chip handling. In many cases, collective wafer level capping by wafer bonding may be a suitable solution. Moreover, the RF package has to preserve the high frequency performances (high-q) of individual RF MEMS devices. Currently, both wafer-level chip size packaging and flip-chip solutions complies with all these requirements for RF MEMS, [24], [25]. It is also worth noting that, typically, an RF packaging has also to be evaluated in terms of: (i) series inductance to the RF input/output ports, (ii) common lead inductance between RF ground of the die and the package, (iii) coupling between package legs and (iv) dielectric loading as a result of the material covering the die. Thus, the package optimization task can be more difficult that is expected because all the mentioned parameters/effects could be frequency dependent. New packaging concepts such as system-on-a-package instead of systems-on-a-chip and single-packaging solutions have already emerged stimulated by the use of RF MEMS. 6. Conclusion MEMS seem predestined to become a hallmark 21stcentury manufacturing technology due to their unique synergy between various unrelated application fields and new functionalities. Even though not yet a mature technology, RF MEMS is expected to be the main driver of progress, innovation and standardisation for the MEMS technology, in general. Particularly, RF MEMS passive devices demonstrate unique characteristics for use in future RF ICs with reconfigurable architectures and enhanced wide-band performances. Acknowledgements: The author acknowledges: Dr. Pierre Nicole from Thales Airborne Systems, France, Prof. Michel Declercq, Dr. Catherine Dehollain and Vincent Pott from EG-EPF, Dr. Philippe Fluckiger, Dr. Cyrille Hibert and Raphael Fritschi from the Center of Microtechnology of the Swiss Federal Institute of Technology (EPF), ausanne, Switzerland for useful discussions and help in the preparation of parts of this work. Particular thanks are expressed to Dr. Kaustav Banerjee from Stanford University, for reading the manuscript and providing enlightening comments. This work has been partially supported by the EPF s internal research grant on Passive MEMS Devices for RF ICs and by the Swiss National Science Foundation via JRP N 7RUPJ References [1] E.R. Brown, IEEE Transactions on Microwave Theory and Techniques, vol. 46 (1998) p [2] C.T. Nguyen, IEEE Trans. On Microwave Theory and Techniques, vol 47 (1999) p [3] K.E. Petersen, IBM J. Res. Develop., vol. 23 (1971) p [4] C.. Goldsmith, Z. Yao, S. Eshelman, D. Deniston, IEEE Microwave and guided Wave etts., vol.8 (1998) p [5] J.J. Yao and M.F. Chang, 8 th Int. Conf. on Solid- State Sensors and Act., Eurosensors IX (1995) p [6] C. Goldsmith, J. Randall, S. Eshleman, T.H. in, IEEE MTT-S Digest (1996) p [7] S. Pacheco, C.T. Nguyen,. Katehi, Proc. IEEE MTT-S Int. Microwave Symp., Baltimore, Maryland (1998) p [8] J.B. Muldavin and G.M. Rebeiz, IEEE Trans. On Microwave Theory and Techniques, vol. 48 (2000) p [9] A. Dec and K. Suyama, IEEE Trans. On Microwave Theory and Techn., vol. 46 (1998) p [10] J. Zhou, C. iu, J. Schutt-Aine, J. Chen and S.-M. Kang, Proc. of IEDM 2000 (2000) p [11] R. Fritschi and C. Hibert, CMI-EPF Internal Research Report, [12] V. Pott, A.M. Ionescu, R. Fritschi, C. Hibert, Ph. Fluckiger, G.A. Racine, M. Declercq, Ph. Renaud, A. Rusu, D. Dobrescu,. Dobrescu, Proc. Of the International Annual Conference, CAS 2001, Sinaia, Romania (2001) in press. [13] A.M. Ionescu et al, submitted to IEDM [14] J. Seeger and S. Crary, Proc. of Transducers 97, Chicago (1997) p [15] C.H. Ahn and M.G. Allen, Trans. On Industrial Electronics, vol. 45 (1998) p [16] J.B. Yoon, B.K. Kim, E, Yoon, C.-K. Kim, IEEE Electron Device etts, vol. 20, 1999, p [17] K. Banerjee, S.J. Souri, P. Kapur, and K.C. Saraswat, Proceedings of the IEEE, Vol. 89 (2001) p [18] A. Dec and K. Suyama, IEEE Journal of Solid- State-Circuits, vol. 35 (2000) p [19] D. Young, J. Tham and B. Boser, Proc. Int. Conf. Solid-State Sensors and Actuators (1999) p. P4D41. [20] J. Ehmke; J. Brank; A. Malczewski; B. Pillans; S. Eshelman; J. Yao, C. Goldsmith, Proc. of Emerging Technologies Symposium: Broadband, Wireless Internet Access (2000) p. 4. [21] C.T. Nguyen, Proceedings of Bipolar/BiCMOS Circuits and Technology Meeting, (2000) p [22] B.E. Duewer; J.M. Wilson, D.A. Winick, P.D. Franzon, Proceedings 20th Anniversary Conference on Advanced Research in VSI (1999) p [23] N. Scott Barker, G. Rebeiz, IEEE Trans. On Microwave Theory and Techniques, vol. 46 (1998) [24] G. Weinberger, Solid-State Circuits Conference, 2000 Digest of Technical Papers, ISSCC (2000) p. 20. [25] H.D. Wu, K.F. Harsh, R.S. Irwin., W. Zhang, A. Mickelson, Y.C. ee, 1998 MTT-S Digest (1998) p. 127.