Micromachined RF side-steps tuning and switching dilemma

Micromachined RF side-steps tuning and switching dilemmaSteve Bush
Micromachining is coming to RF chips. The only important question remaining is when.
“It has been 20 years since a technology has come around that will give us such an improvement,” said Professor Elliot Brown, a micromachine specialist at the University of California, Los Angeles.  
  New horizons… Architectures that are unthinkable using off-chip components become viable with a MEMS approach. For instance, rather than implement a variable filter at the front of a radio receiver, the diagram above shows one mechanically resonant filter per reception band. Each performs all out-of-band rejection needed for the application and they are switched in and out using micromachined electrostatic relays. The whole thing could be built in the corner of a chip. The diagram below shows a phase-shifter, constructed simply from lengths of track and electrostatic micro-relays.  
Silicon semiconductors, so useful in digital and low-speed analogue circuits, are more limited in RF applications.
True, there are silicon processes that will work at 2GHz, but any tuning components or signal path switches have to be off-chip. And the tuning problem extends all the way down the RF spectrum.
Only semiconductors with high bulk impedances, such as expensive GaAs, can be used to make filters with any appreciable selectivity.
Making micro-electromechanical systems (MEMS)by micromachining looks set to side-step the switching and tuning problems by allowing the construction of low-loss relays, mechanically resonant filters, high-Q inductors and variable high-Q capacitors.
All of these have been constructed on-chip using modifications of standard semiconductor processes and are small enough to lay along-side other circuitry. The University of Michigan, for instance, has constructed a 92MHz bandpass filter with a Q of 8,000 which is only 13.1?m long.
Soon, claimed UCLA’s Brown, RF MEMS will be commonplace on chips. “You will start to think of them like inverters,” he said.
First will come on-chip RF switches in the form of electrostatically deflected relays.
The standard solution to implementing a compact gigahertz-range RF switch is to use a PIN diode where more than 1dBswitching loss is inevitable. A 3-6GHz PIN diode 8 x 8 crossbar switch has an insertion loss of around 21dB. Individual MEMS switches have already been constructed with losses of under 0.1dB and isolation over 35dB.
And they are fast, bounce-free switching in 4?s has been recorded, 20?s is more typical. Although 5V is the target, it takes 40V to operate a MEMS switch this quickly at the moment.
Other drawbacks are low manufacturing yield and limited life. Pellon said:”We need to improve the single switch yield to greater than 95 per cent and want a 30 billion operation life, enough for five years of continuous operation.” Vibration problems?
“To make a sensitive accelerometer you need a large mass and a very compliant spring. Resonator devices operate at VHF, and in future, UHF. Mass has to be small and stiffness high. These resonators are very much ‘anti-accelerometers’, less sensitive than crystals and surface acoustic wave devices,” said Professor Clarke Nguyen of the University of Michigan.
This said, variable capacitors made my micromachining are exactly masses on compliant springs and are therefore somewhat vulnerable to microphony under vibration. In addition, variable capacitors also suffer from Brownian motion noise, although this is not thought to be a big problem at this stage.  
MEMS frequency control components, for filters and oscillators, fall into two types. One uses micromachining to make conventional capacitors and inductors without the limited Q of traditional on-chip reactive components. This is being championed by the University of California, Berkeley – see photographs on this page. These components require special packaging, but operate at atmospheric pressure.
The other approach is to micromachine mechanically resonant structures that are capacitively coupled to the signal path and profoundly affect the signal. These are being made at the University of Michigan by professor Clarke Nguyen. So far his devices, which use combinations of inter-linked masses and cantilevers, have all operated at VHF frequencies.
Some suggest that this is the upper frequency limit of the technology, but he is strongly dismissive:”You could easily make 300MHz without sub-micron lithography. Gigahertz frequencies are reasonable.”
His work has shown that not only can filtering devices be made, but that they can be as time-stable as quartz if they are fabricated from mono-crystalline silicon.
There are several factors holding back the introduction of RF MEMS.
Technology stability is one. Micromachined devices are made using many types of wafer processing. What is needed is standard process, even if it does have some limitations, so that fabs can offer it for production and IP developers can start making libraries.
Michigan’s Nguyen favours a four-mask addition to the usual CMOSprocess as a basic MEMS process.
Designers are another problem, few of them think in terms of MEMS solutions yet.
Lastly, design tools are needed to develop micromachined products. These have to be able to handle the physics of the mass-spring structures used as well as the fluidics of surrounding atmospheres.
When will RF MEMS be qualified for production?
Professor Bernhard Boser of the University of California, Berkeley draws an analogy: “Micromachined accelerometers were five years from invention to qualification for use in airbags.” RF MEMShave been around for two years so far.

Leave a Reply

Your email address will not be published. Required fields are marked *