on research originally started at UC Berkeley, the patented CMEMS
integration technique for fabricating a MEMS oscillator atop an already
finished CMOS chip was conceived by Silicon Clocks founder Emmanuel
Quevy—now MEMS Engineering Director at Silicon Labs—and his team. Over
the last three years, Quevy and his MEMS engineering team collaborated
with Silicon Labs’ CMOS teams to integratedSilicon Clock's CMEMS
technology with Silicon Labs’ existing timing product lines.
fully qualifying its monolithic fabrication technique, the first CMEMS
family—the Silicon Labs 50x—was designed from the ground up to achieve
smaller size, higher reliability, better aging characteristics and lower
cost than the competition.
"We took the necessary time during
the development of the SiLabs-50x to make sure that our first CMEMS
oscillator family was not only smaller, more stable and more reliable
than quartz, but also that it was scalable and less expensive than both
quartz and the two-die MEMS oscillators from our competitors," Quevy
Because the MEMS elements of a CMEMS chip are located on
top of a standard die—instead of on the bottom or sandwiched between
layers—any CMOS foundry can create and test them using ordinary
fabrication and wafer probe techniques.
materials with opposite temperature coefficients into the construction
of its MEMS resonator, CMEMS lightens the load on its frequency-locked
loop circuitry too, resulting in frequency stability that it claims is
superior to both quartz and competing MEMS oscillators. And by fully
characterizing its CMEMS oscillators, Silicon Labs can now guarantee an
operating life 10-times longer than either quartz or competitors MEMS
oscillators, all-the-while maintaining superior frequency stability
despite extreme temperatures, solder shift, load pulling, vibration,
shock and power-supply variations.
The Si50x CMEMS family is
designed for frequencies between 32-kHz and 100-MHz with frequency
stability options as low as plus or minus 20 parts per million over the
entire industrial temperature range of minus 4 to plus 85 degrees
Celsius. Less expensive models with loosened stability and temperature
range options are also available. Programmable options support multiple
low-power modes, low-period jitter modes, programmable rise-fall times
and polarity-configurable output-enable lines.
customize their own parts using field-programmable "blanks" or can use
online tools to order custom-made samples within two weeks. Packages as
small as 2-by-2.5 millimeters are are available as are larger packages
that are pin-compatible with existing quartz and MEMS oscillators
facilitating easy performance comparisons and retrofits. Models are
available with one-, two- or four-frequency outputs as well as a fully
programmable model that allows frequency fine-tuning measured in parts
SiTime mentioned above is selling oscillators that can be embedded in the same package as the microcontroller. I would say that's the way to go.
I wonder how SiLabs compares to the competition? Personally, I tend to favor MEMS platforms based on SOI that are DRIE:d into the structures. Mostly due to the simplicity and the use of monocrystalline Si. SiLabs deposit a pretty thick layer of poly-SiGe (SiGe can be deposited at pretty low temperatures which is critical if CMOS-wafers are the substrate that is deposited) that is formed into the resonator structure.
Guess in the end it's a question about the yield of the processes.
The 2mm x 2.5mm size is nice. It's disappointing how large a lot of the current oscillators are compared to the tiny chips around them. Sometimes the oscillator is bigger than the MCU.
My next question is, when can I get these integrated with the MCU on the same dice?