Today, microelectromechanical systems (MEMS)-based oscillators are merely nudging the traditional crystal oscillator market. Significant design wins will occur soon if the MEMS companies are right, but commercial quantities of the first products only recently became available. Still, a number of factors have converged that could make MEMS-based oscillators successful.
Virtually any product that includes a microprocessor or microcontroller needs a clock, and this covers the large range of high-volume personal electronics devices from cell phones to PDAs and PCs. It's not as though other types of MEMS-based devices aren't already being used.
According to Karen Lightman, managing director of MEMS Industry Group, “…we are now seeing dozens of MEMS devices in a single mobile phone: MEMS-enabled RF, silicon microphones, and camera stabilization.”1 In the past, MEMS-based clock sources simply have proven too difficult to make commercially.
Although a MEMS-based oscillator appears to be straightforward, the development work required to achieve performance comparable to quartz crystal oscillators has been significant. Quartz crystal units—the resonators—are well understood and benefit from decades of manufacturing refinement. They provide a low-cost, very accurate, low-drift timing reference and are used by the billions. According to SiTime, one of the companies making MEMS-based oscillators, at least 10 billion quartz crystal units are manufactured each year.2
MEMS-based devices present a different set of technical pros and cons because of their extremely small size and silicon material. An advantage of silicon is that high-temperature annealing can be used to stress-relieve resonators and minimize aging effects. In contrast, it takes a relatively long time and is expensive to stress-relieve quartz crystal units because the annealing temperature must be low.
On the negative side, contamination has been a major factor stopping MEMS-based oscillator commercialization. Because of the extremely small size, frequency accuracy can be affected by any contaminates within the resonator package. As a result, breakthroughs in the manufacturing process were needed to achieve the required cleanliness and hermetic seal around the resonator. Coincidentally, quartz crystal manufacturers have found they too need to provide near-perfect packaging for the latest millimeter-size quartz devices for the same reason.
Quartz Is a Tough Act to Follow
Overcoming technical challenges is an obvious requirement, but MEMS-based oscillators also need to be commercially successful against an extremely well-established product. Initial MEMS-based oscillators must perform at least as well as their quartz counterparts and at a similar or lower cost. There have to be advantages to using the new devices beyond just size, given that quartz oscillators already are available in a package as small as 2.0 mm x 2.5 mm x 0.85 mm.
Figure 1a represents the equivalent electrical circuit of a quartz crystal. C0 is the sum of the capacitance of the electrodes deposited on the thin crystal plate and any stray capacitance of the enclosure. C1, L, and R describe the electrical analog to the crystal's mechanical vibrations and are called the motional parameters. The ratio of C0 to C1 is proportional to the electrical energy stored in the capacitor formed by the electrodes divided by the mechanical energy stored elastically in the crystal.
Figure 1b presents a graph of reactance vs. frequency. A quartz resonator typically is operated between the series resonance frequency (fS) and the antiresonance frequency (fA). The slope of the curve above fS is inversely proportional to C1. This means that a very small value of C1 will result in a steep slope and little available tuning range between fS and fA. Conversely, a small C1 can be desirable because it improves the crystal's frequency stability.
For an AT-type crystal, one cut at angles relative to the crystallographic axes that provide very low temperature sensitivity, typical circuit values are C1 = 0.01 pF, L = 0.1 H, and R = 5 Ω. fS is defined as
For these values
For the circuit in Figure 1a, the quality factor Q = 1/(2π fS C1R). This value is called the unloaded Q because it applies to the crystal unit itself without any influence from the additional oscillator circuitry. Q is proportional to the ratio of the reactance to the resistance or equivalently the ratio of the energy stored to the energy lost per cycle. For a series RCL circuit
or with the example values
This is a very high value, typical of an unloaded resonator.
The loaded Q is lower often because of the amplifier's high output impedance, especially at very high frequencies. A series resonant crystal unit needs to be driven from a low impedance source to maintain a high loaded Q.
The Driscoll type of oscillator is one way to achieve this. It places the crystal unit in the emitter circuit of a cascode-connected transistor amplifier. With sufficient bias, the emitter output impedance can be very low.3
Loaded Q clearly affects short-term frequency stability. Not so obvious is the connection between loaded Q and phase noise. D. B. Leeson's 1966 paper, “A Simple Model of Feedback Oscillator Noise Spectrum,” Proceedings of the IEEE, includes an equation with loaded Q multiplied by frequency offset in the denominator of an expression for noise power. It follows that phase noise will be greater in an oscillator with a low value of loaded Q at small frequency offsets. So, in addition to ensuring better short-term stability, a high loaded Q value also reduces close-in phase noise.
MEMS-Based Resonators
Against this background, how do MEMS-based resonators and oscillators compare? Actually, MEMS technology is being used not just to make silicon resonators, but also to manufacture ever-smaller quartz crystal units. As the demand for very small crystal oscillators has grown, it hasn't been possible simply to scale down existing designs. Instead, Epson Toyocom has developed a QMEMS process that combines quartz and MEMS technologies.
For smaller quartz crystal units, it is necessary to significantly bevel the edges so that the uniform central portion defines the electrical and piezoelectric behavior. This precision microfabrication is done using MEMS techniques. The QMEMS FC-12M Tuning Fork Crystal Unit provides a nominal 32-kHz to 78-kHz oscillation frequency with ??30-ppm to ??50-ppm tolerance in a 2-mm x 1.2-mm x 0.6-mm package.
Figures 2a through 2d show the results of a detailed finite element analysis (FEA) simulation of crystal beveling. The desired mode of vibration is a thickness shear (TS) mode; but without beveling, flexure and thickness twist modes also are present that reduce the size of the TS frequency peak. With increasing amounts of beveling, the other modes are significantly suppressed, resulting in a strong TS mode.4
SiTime, a company making only silicon resonators and MEMS-based oscillators, was formed in 2004 to commercialize a series of inventions that solved the problems previously limiting the use of MEMS resonators. The MEMS First™ manufacturing process includes the deep reactive ion etch (DRIE) process licensed from Robert Bosch, GmbH. Using this method in a standard CMOS fab, resonators are formed that measure 300-??m across and have a Q of 80,000. The resonator is sealed, combined with a CMOS driver IC, and packaged to produce a complete oscillator.
This type of resonator isn't electrically connected to the oscillator circuitry in the same way as a crystal. Instead, the oscillator electrostatically excites and senses the resonator vibration so the loaded Q should remain high. On the other hand, because silicon's Young's Modulus changes with temperature, a silicon resonator's frequency has about a -20-ppm/??C temperature coefficient (tempco). SiTime uses an electronic frequency compensation method invented by A. Partridge, one of the company's cofounders.
According to Markus Lutz, executive vice president, CTO, and another cofounder, “…[the] first products already have achieved six-sigma process capability to provide ??50-ppm frequency tolerance including initial offset and long-term drift over the full industrial temperature range from
-40??C to + 85??C.” Dr. Lutz invented the quadratic-shaped resonator as well as EpiPoly encapsulation, key to low-cost MEMS plastic packaging.
The separately packaged resonator is sold as part SiT0100, has a die size of 0.8 mm x 0.6 mm x 0.15 mm, oscillates at 5.1 MHz with Q of 80,000 and 0.15-ppm/25 years typical aging, and exhibits -115-dBc/Hz phase noise at 10-kHz offset. The SiT0200, a 20-MHz version, is under development. Complete oscillators covering the 1-MHz to 125-MHz range combine the resonator with a low-noise Boser oscillator circuit, resonator driver, nonvolatile configuration memory, and a PLL to derive the required output frequency.
Piyush Sevalia, the company's vice president of marketing, claimed, “SiTime's MEMS resonators have significant size and cost efficiencies compared to quartz resonators. In addition, they are more robust and can be packaged with standard IC packaging technologies.”
Another company solely engaged in MEMS-based resonators and oscillators is Discera. This business was founded in 2001 by Dr. Clark T.-C. Nguyen, presently a professor in the Department of Electrical Engineering and Computer Sciences at the University of California at Berkeley and a director of the Berkeley Sensor & Actuator Center.
The company draws upon several resonator designs depending on the application. For example, a folded-beam comb-drive 32.768-kHz resonator is suitable for a real-time clock or watch. At the opposite extreme, disk or wineglass mode resonators have demonstrated high Q at RF frequencies.
Discera's MOS1 Oscillator family covers the 1-MHz to 125-MHz range and, like SiTime devices, consists of a separately packaged resonator bonded to the oscillator chip and then encapsulated in plastic. Figure 3 shows the MEMS resonator protected by a cap—the square piece on the very top of the stack—that was attached by glass frit wafer bonding under high vacuum. The resonator and cap are mounted on the oscillator ASIC, the bottom die. Using the company's PureSilicon™ technology, there seems to be no problem in meeting the required long-term and temperature-related stabilities routinely achieved by crystal oscillators.
To put the frequency tolerance in perspective, Michael Semos, a product marketing engineer at Epson Electronics America, Timing Products Business Unit, said, “The available frequency tolerances are based on market requirements, and the majority of applications we serve still requires from ??50 ppm to ??100 ppm.”
Wan-Thai Hsu, chief technology officer for Discera, said, “The guaranteed tolerances on production parts are based on very early conservative estimates from initial device testing. However, we have been seeing an average of ??10-ppm frequency deviation for the current production parts.” Specific frequencies are derived from the basic resonator frequency via a PLL.
It's obvious that the company's plans are not limited to this series of oscillators. In a recent paper detailing research on micromechanical filter arrays, the comment was made, “While the initial product offerings will consist of frequency and clock sources, the eventual goal of MEMS frequency control research is the complete integration of all the components necessary in a radio architecture, including clock sources, reference oscillators, mixers, and filters.”5
In another paper, Mr. Hsu clarified some of the differences between achievable quartz oscillator performance and the current state of MEMS devices. “At this moment, given the fact that phase noise performance is traded for digital temperature compensation of MEMS oscillators, vibrating MEMS oscillators are only good for clock and timing applications although that is a huge market. To achieve the specifications for RF applications, one of the future research focuses should be on making temperature compensation without sacrificing the phase noise performance.”6
Silicon Clocks is taking yet another approach to MEMS-based oscillators. One of the major reasons that MEMS resonators cannot be integrated on the same die as CMOS oscillator circuitry is the high temperature required for MEMS processing. Silicon Clocks has developed a lower-temperature silicon germanium (SiGe) process that allows a MEMS resonator to be formed directly above the oscillator circuitry.
According to Andrew McCraith, business development manager and one of the company's founders, “Instead of trying to replace a single quartz crystal or oscillator, Silicon Clocks is developing timing solutions using multi-MEMS oscillators. These products can consolidate multiple resonators and oscillators into a single component with an accompanying reduction in size, cost, and power consumption.”
Mr. McCraith also addressed the high phase noise of competing MEMS oscillators and the advantages of his company's approach: “MEMS devices intrinsically are very sensitive to temperature. The existing compensation solutions require a large, noise-generating, power-hungry fractional-N PLL. As a result, there are limits to the performance that can be achieved at a viable size and cost.
“The challenge is exacerbated in oscillators with separate MEMS resonator and oscillator IC die,” he continued. “One of the advantages of the Silicon Clocks' fully integrated, single-chip solution is reduction in frequency variations by orders of magnitude and elimination of the fractional-N PLL. This approach enables tolerances comparable to or better than those associated with quartz crystal oscillators.”
The company is working closely with selected customers, but MEMS oscillator data sheets have not been publicly released at this time. However, the J-Series™ family of precision clock synthesizers provides a very low phase-jitter output at frequencies between 100 MHz and 675 MHz.
The device includes a SmartPLL™ that multiplies a 25-MHz to 45-MHz fundamental crystal oscillator frequency to more than 20 of the most common reference frequencies used in high-speed serial data interfaces. Integrated phase jitter is less than 1-ps rms and pk-pk period jitter typically less than 25 ps.
By way of comparison, the SiT8002 and SiT1 Oscillators are described in the SiTime 2007 product catalog as having approximately 20-ps rms jitter. A 100-MHz SiT1270 is shown in a scope photo to have a typical 67-ps pk-pk jitter. The company obviously recognizes the importance of low phase noise as the catalog continues, “The next-generation SiT8102 has approximately 3-ps rms period jitter. This jitter performance is suitable for clocking high-speed serial buses….”
Non-MEMS Oscillators
Oscillators typically have a resonator of some kind, and it can be an electrical circuit rather than a crystal or a MEMS device. Low-cost IC versions of free-running oscillators are readily available, but they tend to have a large frequency tolerance precisely because they lack a crystal or MEMS resonator.
Mobius Microsystems is trying to change that. The company's technology uses standard CMOS processes, which means that no additional costs are incurred. There is no resonator die to be bonded or separate process steps required after the basic CMOS circuitry has been formed.
Starting from a very high-frequency LC oscillator, termed a CMOS Harmonic Oscillator (CHO™), the required output frequency is produced by programming the required division ratio. Compared to frequency multiplication, common in PLL-based oscillators and responsible for increased phase noise, frequency division reduces phase noise. The company claims that, at higher offset frequencies, the CHO can provide up to 20-dB lower phase noise than PLL-based designs.
It's clear that a free-running oscillator could have these advantages, but how does it also maintain a tight frequency tolerance? According to Mobius Microsystems, “The foundation of this patented CHO design is a harmonic LC resonator circuit which…is designed to maintain its frequency accuracy in an open-loop configuration. The key innovation that Mobius brings to market is a complex analog control circuit that compensates for process, voltage, and temperature variations to stabilize the oscillating frequency….”
Tunc Cenger, director of marketing, commented, “To be commercially successful, an oscillator needs to have less than 500-ppm frequency error to comply with the USB 2.0 high-speed (HS) interface standard or 350 ppm for S-ATA Gen 2 and 300 ppm for PCI-Express. These three interfaces represent very significant markets that we address with our CHO technology. Nevertheless, it should be made clear that the tolerances we quote for our oscillators do not imply a fundamental technology limitation.”
The MM8511 Spread Spectrum Clock Generator is based on the CHO technology. The output frequency is factory programmed to a value between 10 MHz and 66 MHz, but the user can select either down-spread or center-spread operation with a range of from zero to -6% or zero to ??3%, respectively.
Summary
Would your application benefit from one of the new clock sources? If you need an accurate source but size is important, the oscillators from SiTime or Discera might be good choices. If the lowest cost is a factor and unit volume is high, Silicon Clocks' SiGe process is worth evaluating. These solutions can provide 50-ppm to 100-ppm frequency tolerance. Where a separately bonded resonator is used, package size is comparable to the smallest quartz oscillators, and the cost is similar if not lower.
If a higher frequency tolerance can be accepted, for example in a serial bus interface, Mobius Microsystems' CHO devices appear to have an advantage. Because they are built using standard CMOS processing, these oscillators should have the lowest cost. Because they do not multiply the fundamental resonator frequency with a PLL but instead divide it, phase noise also is very low.
Although products from only a few companies have been discussed in detail, the clock and oscillator market is huge, amounting to at least $3.2 billion in 2006, so it has attracted lots of attention.7
Nevertheless, it's early days for MEMS-based clock sources. Quartz crystal oscillators will continue to be the preferred solution for tight-tolerance timing references because there's no viable substitute. With the introduction of alternative technology oscillators that are smaller, cost competitive, and sufficiently accurate to meet 50-ppm to 300-ppm requirements, designers have a choice. No doubt, this market will be an exciting one to watch as MEMS-based resonators, free-running CMOS oscillators, and yet-to-be-introduced technologies attempt to unseat quartz from its entrenched position.
References
1. “MEMS Industry Challenges and Trends,” www.memsinvestorjournal.com/2008/06/mems-industry-challenges-and-trends.html#more
2. McDonald, J., “A New Paradigm in Time: Silicon MEMS Resonators vs. Quartz Crystals,” R&D, April, 2006.
3. Bartram, C., “Notes on the Driscoll VHF Overtone Crystal Oscillator and a New Low-Noise VHF Crystal Oscillator Topology,” http://www.christopherbartramrfdesign.com/blaenffos/oscillator/VLNO.pdf
4. Pao, S. Y., et al, “Beveling AT-cut Quartz Resonator Design by an Efficient Numerical Method,” www.txc.com.tw/download/tech_paper/2005-IUS-1(English).pdf
5. Clark, J. R., et al, “Parallel-Coupled Square-Resonator Micromechanical Filter Arrays,” Discera. 6. Hsu, W-T, “Vibrating RF MEMS for Timing and Frequency References,” Discera.
7. Bouchaud, J., and Knoblich, B., “MEMS, Microfluidics and Microsystems Executive Review,” Wicht Technologic Consulting, www.memsinvestorjournal.com/2006/10/mems_oscillator.html
| FOR MORE INFORMATION | Click below | |
| Discera | MOS1 Oscillator | Click here |
| Epson Toyocom | QMEMS FC-12M Tuning Fork Crystal Unit | Click here |
| Mobius Microsystems | MM8511 Spread-Spectrum Clock Generator | Click here |
| Silicon Clocks | J-Series Clock Synthesizer Family | Click here |
| SiTime | SiT8102 Oscillator | Click here |
.png?auto=format,compress&fit=crop&h=139&w=250&q=45)
