Oscillator aims MEMS at the mainstream

That would eliminate the need to have more than one time-base chip per electronic device since they can all be built-into a single chip. "For example," Brown said, "you could pack a real-time clock at 32 kHz, plus a communications frequency oscillator operating at several hundreds of megahertz, plus you could add any number of other kilohertz or megahertz frequencies and all their PLLs on one monolithic piece of silicon. You can't do that with quartz."

Inside the parts
SiTime's oscillators are based on an internal MEMS resonator as an alternative to quartz-crystal resonators. Other companies have developed solid-state alternatives to quartz-crystals for RF and IF applications, such as surface acoustic wave (SAW) filters, ceramic filters and film bulk acoustic resonators (FBAR). SiTime, for its part, says its MEMS resonators not only apply all those applications but are less-expensive to add to systems, are more reliable and smaller, and don't need customization.

SiTime's initial product, the SiT8002, is pin-for-pin compatible with Epson's SG-8002 but also has lower jitter, lower phase noise, lower power consumption and a smaller package, the company said.

To fabricate a MEMS oscillator, SiTime begins with a silicon-on-insulator wafer, then adds just four mask levels to craft a mechanical "tuning fork" or "beam" that is attached to the substrate on one side and is severed elsewhere by a deep trench etch process. By opening a 400-nanometer gap for the 10-micron-tall beams, whose width is determined by the desired frequency, the beam can freely oscillate when driven electrostatically by an electrode on one side of the beam. A capacitive sensor is then located at the other side of the beam to detect the oscillation frequency and drive the phase-locked loop circuitry that conditions the signal to mimic a traditional quartz-crystal time base.

To ensure long-term reliability, SiTime grows an epitaxial silicon cap above a polysilicon layer across its entire wafer, securely encapsulating the MEMS oscillator and providing an atomically smooth surface over which normal CMOS circuitry could be grown. The capping operation — which is performed at 1,100°C to drive out such possible contaminants as moisture — isolates the structure from the environment so that it cannot affect performance now or in the future, ensuring the long-term stability that has traditionally eluded MEMS.

The other big pitfall that stumped previous MEMS resonator designers was poor temperature hysteresis and a temperature coefficient of 30 ppm/°C. SiTime claims to have solved these problems with electronic circuitry it adds to the CMOS portion of its two-chip solution.

"Across the surface of our wafers," said Brown, "we see about a 0.8 percent deviation in frequency. So we connect our resonator to a CMOS die, which pulls in the frequency to 5-ppm. The circuitry also performs temperature compensation and A/D conversion, and locks in the desired frequency with a phase-locked loop."

Brown also claimed that SiTime MEMS-First oscillators are "instantly on" with exactly the right frequency, whereas quartz crystals require a warmup period as the rest of the circuitry waits for the clock to settle. He further said that scaling down the MEMS resonator for higher frequencies will be easy.

"One of the remarkable things about MEMS-First technology," Brown said, "is that as its size is scaled down for higher frequencies, its performance actually goes up — which is the opposite of quartz, whose performance goes down for smaller sizes."

The company claims to have fabricated many geometries for its tuning-fork-like resonators using standard CMOS photolithographic techniques. Successful geometries include a four-beam structure attached only at the middle, a one-beam structure and a disk structure, all of which have different trade-offs regarding frequency and performance.