Frequency Stability Control of MEMS Oscillators


Oscillators, also known as clocks, frequency references, or timing references, are an integral part of any electronic system. They are used in thousands of applications including computers, cameras, televisions, watches, music players, phones, printers, cars, and radios. For decades, quartz crystal oscillators have dominated the timing reference market because of their excellent stability. In other words, the output frequency of a quartz crystal oscillator can be made constant to within less than 0.000001% over a very large range of temperature, humidity, atmospheric pressure, supply voltage, etc.


Silicon micromechanical resonators, the focus of this research, have become an attractive alternative to quartz crystals for use in oscillators in consumer applications that do not require high stability. Silicon resonators are potentially CMOS compatible, cheaper, smaller, and consume less power than quartz. However, up to today, quartz oscillators have outperformed silicon resonators in terms of stability and phase noise. Our group hopes to remove these limitations and produce silicon frequency references that are suitable for high end applications such as encrypted radio communication.


Our group has developed an encapsulation process for our silicon resonators that seems to have solved the long term stability problems associated with humidity, pressure, and aging effects. However, our resonators still show significant frequency drift with temperature.

Several methods of frequency-temperature compensation have been demonstrated by our group to negate this effect. My research aims to demonstrate a new method which shows potential for ultra-high stability. The basic concept involves using two thermally coupled resonators of different compositions of Si and SiO2 to generate two frequencies with different temperature dependencies. The idea of using composite beam structures to alter the temperature dependence of our resonators was first developed by our group as a method of passive temperature compensation, but is being used here to enable temperature measurement.

By mixing these frequency signals together we generate a difference frequency that can be used as an accurate sensor of the resonators’ temperature.

Once the temperature of the resonator is known, a control signal can be applied to a heater to keep this temperature constant. The heater can be fabricated inside the encapsulated cavity alongside the resonator, thereby forming a micro-oven. The resulting small size and large thermal isolation of the micro-oven allow the heater to achieve excellent power efficiency. By maintaining the resonator at a constant temperature inside the micro-oven we thereby guarantee a stable output frequency regardless of the external ambient temperature. The control scheme resembles that of a phase locked loop.

Because the temperature sensor signal is a frequency rather than a voltage or current, measurement of this signal should be insensitive to process, temperature, and voltage variations. As long as the system remains linear, these variations can affect amplitude or phase but cannot affect frequency. In addition, notice that all electronic components including the phase detector and the heater are “inside the loop.” As a result, any variations in these components due to process, temperature, or voltage will be automatically corrected by the negative feedback. I hope to show that this is a significant advantage over the Q(T) method, which is limited by temperature variations in the control electronics.


This research is supported by an NDSEG fellowship and by funding from the DARPA HERMIT Program.