Portland, Ore. -- The last holdout against the microminiaturization of electronics--the quartz crystal--may be set to fall. Wielding oscillators based on microelectromechanical-system technology, two startups aim to break quartz crystals' monopoly on the mechanical time references used in virtually all electronic devices today. If the two are successful, larger competitors with established MEMS programs could be hot on their heels.
But while the rivals envision a multibillion-dollar market for MEMS oscillators, the parts' price tag could confine them, at least at the outset, to niche markets.
Quartz oscillators provide electronic circuits with a resonant mechanical clock by virtue of a vibrating piezoelectric crystal. MEMS alternatives promise smaller, higher-precision time bases that are unaffected by moisture, temperature or mechanical vibrations. "The oscillator is an often forgotten market, but it is large, it is growing and it has had very little substantive innovations in the last few decades," said Venkat Bahl, vice president of marketing at Discera Inc. (San Jose, Calif.). "Our MEMS oscillators are going to be significantly smaller, significantly higher-performance and, in the long run, significantly lower-cost, the holy grail being integrating MEMS oscillators onto CMOS chips."
For now, Discera is readying a pin-for-pin replacement for the popular Epson 8002 quartz crystal. The MOS1 series "can be selected in any frequency from 1 MHz to 125 MHz," said Wan-Thai Hsu, chief technology officer at the startup.
Analysts said both Discera and its only current competitor, Robert Bosch spin-off SiTime Corp. (Sunnyvale, Calif.), will aim to lower their parts' cost below that of quartz crystals by using the same techniques that have been applied for conventional CMOS devices: mass production and miniaturization. The ultimate vision is integration of the oscillator function directly onto CMOS chips that hold other circuitry, though for now technical obstacles stand in the way of that goal.
"MEMS will begin to slowly replace crystals in cell phones beginning in about 2008, but with no substantial volumes until after 2010," said Stephen Cullen, a contributing analyst at In-Stat. "The size is right, but the cost is too high for widespread adoption."
MEMS oscillators can be tuned for high-precision applications, where their features may justify the price premium. And indeed, "both Discera and SiTime seem to be targeting the high end of the market: low-jitter parts operating above 100 MHz. That is where a 40-cent MEMS part makes sense compared with a 15-cent crystal oscillator," said analyst Stephan Ohr at Gartner Dataquest.
Ohr pegged the market for precision timing generators at "about a $350 million slice of the total market for standalone oscillators, PLLs [phase-locked loops], timing generators and delay lines," and predicted it would grow to account for "a larger proportion of a total $1.2 billion market by 2010."
The total potential market for MEMS oscillators in all the places where crystal oscillators are used today exceeds $3 billion, according to the MEMS chip vendors. Dean Freeman, a semiconductor-processing expert at Gartner Dataquest, agreed that "the market is good-sized when you consider all the potential applications."
The question, Freeman said, is "whether Discera and SiTime can build MEMS parts as inexpensively as quartz. MEMS makes their parts more reproducible than quartz, as a result of the semiconductor-manufacturing process, but will they be able to build them as cost-effectively as quartz? If not, then they will have to convince the user that they are adding value."
Discera acknowledges that its initial parts are aimed at customers seeking added value in the form of higher precision or extreme durability, but the company also contends that by creating its MEMS devices on CMOS processing lines, it can eventually drive the cost down to compete with quartz on price.
"Unlike quartz crystals, which must have an individual hermetic seal added to each ceramic package, MEMS can go in plastic packages. Our packaging cost is less than half that for quartz," said Discera's Hsu. "Unlike quartz crystals, our resonator is batch fabricated, potentially giving it lower cost. And the third component of an oscillator--the integrated circuit--is pretty much the same for both the quartz crystal and MEMS."
If Discera and SiTime are successful in the quartz crystal replacement markets, it's likely that other companies with MEMS programs will rush to ready competitive offerings. "There are half a dozen companies working on MEMS timing products right now," said Cullen of In-Stat, though "Discera and SiTime are the two that seem to be closest to commercial success. Discera has been on a longer and slower ramp, but I think they are at about the same place now as SiTime."
Discera's MEMS technology was home-grown from underpinnings laid by Hsu at the University of Michigan, Ann Arbor. SiTime licensed its "deep trench" approach from MEMS giant Robert Bosch.
NXP Semiconductors also has a mature MEMS technology, developed by Philips before that company spun out its semiconductor unit as NXP, but thus far the company hasn't entered the oscillator market. "A dark horse in this race could be NXP, who I believe has the technology and, in its new incarnation, could decide to run with it," said Cullen.
When current drives a quartz crystal, it behaves like a circuit composed of a resistor, inductor and capacitor (RLC), giving it a precise resonant frequency--synchronized to the expansion and contraction of the piezoelectric material--when placed in a feedback loop. For high-precision applications, the hermetically sealed package also includes a temperature-controlled "oven" and a mounting with shock absorbers to prevent perturbation by external mechanical vibrations.
MEMS oscillators likewise behave like resonant RLC circuits, but they integrate all the functions of millimeter-sized quartz crystal into a micron-sized "tuning fork" that is naturally immune to temperature and vibration.
"Quartz crystals have been with us for so long because you can't get a high-quality, high-stability time base with electronics alone--you need a mechanical frequency reference. The whole radio-frequency MEMS area is based on trying to construct very small mechanical frequency references on a CMOS chip," said Hsu.
Though Discera's MEMS is cast into a CMOS chip, neither its nor SiTime's MEMS chips hold circuitry. Today, adding MEMS to a CMOS chip with circuitry would risk turning the circuitry to soup by virtue of the high temperatures needed to sculpt MEMS structures from silicon.
Consequently, like quartz crystals, MEMS resonators today require a separate CMOS chip, containing a feedback path to maintain oscillation, a PLL to stabilize frequency and other conditioning circuitry to optimize performance. In the case of quartz crystals, the CMOS chip is wired to the crystal inside the package. By contrast, the MEMS resonator, as a planar CMOS chip itself, can be directly bonded to the chip holding the circuitry.
Without the conditioning circuitry on the CMOS chip, a micron-sized MEMS resonator is not as accurate as its large quartz counterpart. When the CMOS circuitry is added to the mix, however, MEMS oscillators can be tuned more precisely than their quartz counterparts, Hsu maintained.
According to Hsu, understanding how to sculpt the MEMS mechanical resonator on a CMOS chip and then bond it to the separate circuitry chip was the easy part. The hard part--involving more than a decade's worth of work--has been building optimized devices that can compete with mature quartz crystal technologies from manufacturing giants like Epson.
"We started 10 years ago [in the lab] with very low-frequency MEMS resonators--in the kilohertz range--but since then we have learned a lot about MEMS structures, taking us first into the megahertz range and today all the way up to 1.5 GHz in the lab," said Hsu.
By 2001, when Discera was founded, Hsu had validated research chips that proved the MEMS oscillator concept. Over the next few years, his team created MEMS tuning forks of various sizes and shapes--including bars, combs, beams, disks, rings and "wine glasses." (The MOS1 uses a bar resonator structure, but the company says it has other architectures in its patent portfolio that will offer higher performance, lower phase noise and ultrahigh stability.)
By 2003, Discera's designs had progressed to a manufacturable process that could be transferred to a foundry. The company claims that all four of the CMOS foundries it has evaluated thus far have been able to build MEMS resonators using its process. (Dalsa Semiconductor, in Bromont, Québec, will build the first devices.)
In the last three years, Discera has concentrated on crafting manufacturable devices. "Our resonator design is a trade-off between the highest-quality research chips and a manufacturable solution," Hsu said. "For example, most MEMS designs require a 3-volt bias, which is high for today's circuitry, but over the last three years we have changed our design to reduce our MEMS bias voltage plus increase its power-handling capability."
Discera and SiTime both say they are sampling units now to potential customers. Both promise to deliver volume production of their MEMS oscillators as soon as customers qualify them late this year or early in 2007.
Discera's customers will require three to six months to qualify the MOS1 for specific applications. In the meantime, the company says it is refining its volume-production materials set and is characterizing its chips for extreme conditions, quantifying their ruggedness to attract customers.
"Our resonators are much more rugged than quartz crystal oscillators [and] keep on working reliably even under extreme temperatures," Hsu said. "One of our engineers put our resonator in liquid nitrogen, for example, and it still made a very stable frequency reference. Another customer hit our oscillator with 30,000 g's of shock, and it kept on working."
One military application now evaluating Discera's MEMS oscillator chip has been paying $65 for a quartz crystal oscillator that can withstand very harsh environments. For such applications, switching to a MEMS oscillator priced at less than $1 could be a no-brainer.
See related image