PORTLAND, Ore. – Requirements for ultra-low jitter and high signal integrity required in PCI-Express interfaces normally require a temperature-controlled quartz crystal to drive a precision clock generator.
MEMS maker Discera Inc. claims to have overcome the requirement for a quartz crystal with a MEMS solution that reduces size and cost of up to third-generation devices. Discera (San Jose, Calif.) claimed it MEMS PCIe clock generator is smaller and less expensive than standard quartz crystal solutions, and “based on our competitive analysis, it is the only MEMS solution with dual outputs,” said Discera CEO Bruce Diamond.
[Get a 10% discount on ARM TechCon 2012 conference passes by using promo code EDIT. Click here to learn about the show and register.]
The DSC557-03 is said to deliver 100-MHz differential output clocks (XO) with phase jitter as low as 0.15 picoseconds and 10 parts-per-million stability up to temperatures as high as 125 degrees C.
The MEMS device is designed to eliminate the external quartz crystal along with capacitors and tuning resistors, saving space in communications, storage and networking devices, Discera said, adding that it draws 25 percent less power and provides 65 percent lower jitter than quartz crystals.
The DSC557-03 is available now in 16 pin drop-in replacements for popular quartz crystals, or a 14-pin QFP that uses 77 percent less board space. Engineering prototypes are programmable with production performance. Volume delivery of production parts is available two weeks after receipt of orders.
The opinions of commentators above seem to indicate MEMS timing devices are currently not a replacements for OCXO' or TCXO's and/or other non-MEMS timing devices that appear to have cost & performance advantages. It would be nice to see poster's like @psevalia contribute a comparative study of this...
Have to agree. MEMS timing devices are a solution looking for a problem. MEMS technology is better served for accelerometers, sensors and other miniaturized devices where the competing solutions converting mechanical to electrical signals are not viable or too expensive (in smartphones, tablets, automobiles etc).
Agreed that the frac-n approach is very problematic. Integer-N approaches rely upon a tunable reference oscillator which is stabilized across temperature and process. That works in the lab at low volume for short periods of time, but it is a million miles from capitalizing upon mature and well understood semiconductor processes. Quartz has nothing to fear, as these integer-n approaches are just science projects.
There is the problem of a fractional-N PLL to bring the frequency within tolerance. This adds jitter and power. Quartz still reigns when it comes to high-performance frequency generation. Solutions getting rid of the frac-N PLL will eventually beat out companies like Discera and SiTime. And there are a few companies already out there using CMOS based or other forms of tuned resonators. The quartz solution gets rid of the need for two cascaded PLLs needed when the output frequency is the GHz range. A single PLL coupled to an external quartz xtal or VCXO will have much better performance with less power.
MEMS oscillators are a bizarre market. It is an interesting technology that offers little value (with the exception of shock and vibe) to the customer. While MEMs does offer a solid advantage in microphones and motion-sensing, overall oscillator performance (including the new breed of piezo-actuated devices) only compete with incumbent technologies when David Copperfield math is performed. The price advantage seems completely artificial when one considers the complete picture.
Hi Colin - SiTime (www.sitime.com) introduced MEMS-based clock generators (with no external MEMS or quartz) in 2009 and have been shipping in volume since 2010. The three devices (SiT9103, SiT9104, SiT9105) offer single-ended, differential or a combination of single-ended and differential clocks respectively and can support many interface protocols including PCI. The single-ended frequencies go up to 220 MHz and differential frequencies go up to 800 MHz. Differential standards support LVPECL, LVDS, CML and HCSL.
Please see links below for more information.
Exec VP, Marketing, SiTime Corp.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.