lunes, 26 de julio de 2010


Micromechanical switches have the capability to offer substantial advantages over their conventional semiconductor counterparts, such as low power consumption, very high isolation, low insertion loss, linear performance and low cost. Expected applications include such areas as high-isolation switching networks, phase shifters and reconfigurable filters, for use in phased-array antennas or filter-path selection in multi-band communications systems at high frequencies (10-100GHz) [1-3]. This paper demonstrates a low-voltage, capacitive shunt microswitch that has a lifetime of greater than five billion switching cycles.

The physical implementation of the capacitive shunt switch consists of a grounded metallic membrane supported a few microns over a passivated coplanar waveguide (CPW) transmission line, Figure 1. During operation of the switch, an RF (radio-frequency) signal and DC bias voltage are superimposed and applied to the signal line input port. In the switch 'on' state, the DC bias is zero, the membrane remains up and the capacitance of the switch is low (of the order of 1-100fF), allowing the RF signal to pass to the output port unimpeded. Increasing the DC bias voltage causes the membrane to pull down to the intermetal

dielectric due to the electrostatic force between the membrane and transmission line. This increases the capacitance to 1-10pF, shorting the RF signal to ground, and turning the switch 'off'. Removal of the DC bias allows the bridge to return to its original position because of the mechanical restoring forces exerted by the membrane anchors. A simplified circuit diagram for the shunt switch is shown in Figure 2.

For conventional switch designs, the bias voltage required to turn off the switch is often relatively high (30-100V). This high voltage necessitates the use of up-converter circuitry and can be difficult to achieve in certain portable wireless communications applications. More importantly, the high electric fields set up across the thin dielectric while the switch is in the 'off' state can cause electrons to be injected into or removed from the dielectric layer. This is known as dielectric charging, and the resultant parasitic charge buildup leads to a shift in the pull-down voltage of the switch equal to

if the charge is assumed to be located at the dielectric-air interface and have surface density Qp, and Cd is the dielectric capacitance per unit area [4-7]. This shift in the pull-in voltage is necessary to compensate for the reduced electric field in the airgap due to the charge at the dielectric surface. It is clear that if the voltage shift exceeds the applied voltage, the switch will fail to actuate and remain 'on'. Alternatively, the switch can fail in the 'off' state due to the attractive force between the charge and metal membrane.

Dielectric charging is strongly dependant on bias voltage, and it has been suggested by Goldsmith et al that a tenfold increase in lifetime results from reducing the required bias voltage by 6-7V [7].

Since mechanical failures are now relatively rare in MEMS, charging has become the dominant failure mechanism of these devices. Almost all capacitive microswitches suffer to some degree from a dielectric charging problem, and it is important that the effect be minimised if such switches are to become commercially successful.
This work concentrates on reducing the dielectric charging problem by reducing the voltage needed to actuate the switch and by using a high-quality silicon oxide as the dielectric layer.

Meander-type tethers are used to suspend the central switch membrane, Figure 1, resulting in a low pull-down voltage (~15V) whilst maintaining reliability. A stress gradient of approximately 200MPa/μm exists in the aluminium film, but this is negated by the placement of the supporting tethers on the diagonal at the membrane corners. The result is a membrane that exhibits a maximum convex bow of 0.5-1.0μm over its width of 200μm. This symmetrical bowing reduces up-state capacitance and hence insertion loss, and so is advantageous rather than being a problem. Perforations in the membrane allow for easy removal of sacrificial layer and enable faster switching speeds by reducing damping.
By using silicon oxide as a dielectric layer, the problems due to dielectric charging are further reduced, although a drawback associated with this is the slight loss of isolation in the off state, due to the relatively low dielectric constant of silicon oxide [2].

A device schematic is illustrated in Figure 3. The switch is surface micromachined using the low-temperature (<350oc)

Including corrections for electrostatic fringing fields and the voltage offset due to dielectric charging of the intermetal dielectric, the pull-down voltage for the devices may be approximated by

where k is the effective spring constant of the structure, fc is a correction for electrostatic fringing fields, A is the membrane area, εo is the permittivity of free space, and go the initial airgap spacing. VQ is the voltage shift due to trapped charge in the dielectric layer.

Using finite element simulations [9], the spring constant for the 200 x 200μm2 device shown in Figure 1 has been evaluated as 2.35Nm-1, and the fringing field correction factor can be as high as 1.2 for an airgap of 5μm. The density of the trapped charge at the oxide surface has been evaluated to be -5.5 x 1010e/cm2. More information on this model is in [10].Pull-down voltages for a number of devices of varying airgap height are illustrated in Figure 4; measured and modelled results are in good agreement. For comparison, results for switches measuring 100 x 100μm2 are also illustrated, although the high pull-down voltages and low isolation exhibited by these devices means they may be of limited interest.

The RF performance of a typical switch is shown in Figure 5. The device membrane measures 200 x 200μm2 and has an airgap of approximately 5μm. Dielectric thickness is 100nm. Six closely spaced switches have been measured and the results averages. Switching voltage is (14.8 +/- 0.8)V, and switching speed is estimated to be approximately 15μs [2]. RF analysis takes place on a Cascade Microtech probe station using a HP8722D network analyser, and measurements have been de-embedded by subtracting the attenuation of a through line of identical dimensions to the device line from all measurements. At 30GHz, average insertion loss and isolation are -0.2dB and -19.0dB, respectively.

For the designs used here, PECVD silicon oxide is the intermetal dielectric, instead of the more commonly used silicon nitride. The oxide has a lower dielectric constant than the more commonly used silicon nitride, (εr = 5.0 instead of 7.0), and because isolation depends on the dielectric permittivity, some reduction in isolation is expected. However, silicon oxide is a much better quality dielectric and does not suffer from the same level of dielectric charging [2, 10].

This was followed by a 'hold-down test', in which the switch was held in the down state at a bias voltage of 16V for a period of 18 hours, without any adverse effects. The switch returned to the up-state immediately the bias was removed, with a change of less than 1.5% in the down-state capacitance over the length of the test. More recently, a 200 x 200μm2 switch (pull-down voltage 12V) was actuated over 5.5 billion times without failure at a bias voltage of 16V and frequency of 5kHz. These tests have been stopped due to time constraints on equipment, and it is expected that substantial increases in this lifetime can be made as time permits.

A long-lifetime, low-voltage micromechanical shunt switch has been demonstrated. The capacitive structure is completely CMOS compatible and has a pull-in voltage of approximately 15V. Radio-frequency performance is promising; insertion loss is -0.2dB and isolation is -19.0dB at 30GHz.
Jorge Polentino

No hay comentarios:

Publicar un comentario