MEMS Resonators look to displace Quartz resonators

Quartz crystals have remarkable mechanical and piezoelectric properties that have made them the timing components of choice for the past sixty years, but they also have some significant drawbacks. These include their sensitivity to heat, shock, vibration, and relatively large size. Crystals also have higher failure rates than silicon integrated circuits. And unlike silicon components, crystals become more expensive and perform worse when they are squeezed into smaller packages. These drawbacks are becoming more apparent and constraining in modern products. Many decades of research into various piezoelectrics, ceramics, and LC circuits, has uncovered no material or technology that is able to match the quartz crystal's exceptional temperature stability and phase noise performance. This has now changed with advances in MEMS design and manufacturing techniques.

MEMS resonators: decades of development
The microelectromechanical resonator has long had the most potential to replace quartz crystals. In work as early as 1965 [IEEE Applied Physics Letters, v.7 pp.84-86, 1965], H.C. Nathanson published papers describing micromachined resonators made with fine metal wire and films.

Unfortunately, the promise of cheap, high quality, fully integrated resonators ran into harsh realities. Early researchers made many advances but also discovered many difficult technical issues. These included silicon's large frequency temperature coefficient, aging from material fatigue, and drift from packaging contamination. This drift was one of the most intractable problems because the resonant elements were so small that a single atomic layer of contaminant could dramatically shift a MEMS resonator's frequency beyond acceptable limits.

The technology also had cost problems. Much of the packaging MEMS devices traditionally used was similar to that used for quartz crystals. Since this was a major portion of the cost of the finished quartz components, it was hard to gain a significant price advantage over the older more mature technology. In short, early MEMS resonators lacked the performance but had the cost burdens of quartz.

This has changed with the development of new fabrication and packaging technologies. MEMS oscillators are now a technical reality, have good performance, are very small, and are extremely cost-effective. The key to this breakthrough was discovered by scientists at the Bosch Research and Technology Center, Palo Alto, CA, from which SiTime Corp., Sunnyvale, CA, licensed its base technology. This technology, now further developed, allows SiTime to manufacture a massmarket MEMS oscillator that competes directly with quartz products.

While it took several innovations to make MEMS oscillators commercially practical, the key developments were in the MEMS First and EpiSeal packaging technologies.

MEMS First and EpiSeal
The MEMS First process builds MEMS components with standard CMOS foundry tools, avoiding costly custom processes and materials while leveraging the high-volume manufacturing and packaging infrastructure used to make standard CMOS circuits. This provides vital economic leverage by repurposing the tremendous investments made by the CMOS industry. One vital part of this, the EpiSeal process, encapsulates the MEMS resonators in an ultra-clean hermetic vacuum environment. This ensures that the resonators will stay stable over a lifetime of service.

Figure 1 illustrates a series of MEMS fabrication cross sections. The process begins (Fig. 2a) with etching 0.4 µm wide trenches down to the oxide insulation layer of 10 µm Silicon On Insulator (SOI) wafers to form resonator and electrode structures. In operation, these resonators will vibrate horizontally to the surface of the wafer.

The trenches are covered (Fig. 2b) with thin layers of oxide, silicon and polysilicon. Small vents are etched in the polysilicon layer through which some of the oxide is removed to release the resonator.

The wafers are placed (Fig. 2c) in an epitaxial reactor at over 1,000ºC to burn off contaminants, seal the vents shut, and grow thick silicon and polysilicon caps. The high temperature also anneals the resonators, smoothes the resonator surfaces, and leaves the resonators permanently sealed within extremely clean vacuum cavities. The thick polysilicon caps are mechanically strong and withstand over a 100 atmospheres pressure during subsequent plastic molding.

Vias are formed (Fig. 2d) through the cap silicon to form electric contacts to the resonator's drive and sense electrodes. The wafers are finished with simple metal traces and bondpads for multichip or system-on-chip packaged oscillators, or CMOS circuitry for integrated oscillators. Standard CMOS circuits may be built on the same die with the buried resonators, providing that care is taken not to place transistors in the polysilicon caps above the cavities.

The MEMS resonators are small and supply small signals. It is therefore necessary to keep the drive and sense circuitry close to the resonator. To do this, SiTime packages a CMOS driver chip with the resonator. The combination of resonator and driver is an oscillator. Figure 3 illustrates the packaged oscillator configuration.

After standard dicing, the MEMS resonators and the CMOS driver ICs are molded into standard plastic packages. SiTime chose QFN-type plastic injection molded packaging for high reliability, low lead inductance, and good thermal performance. This package style also enjoys flexible pad layout and low cost. The first SiTime products are packaged in 2.0x2.5, 2.5x3.2, 3.2x5.0 and 5.0x7.0 mm form factors with 0.85 mm height. These oscillators are direct replacements for quartz crystal based oscillators, fitting onto standard printed circuit board pad layouts.

It is especially noteworthy that this final packaging is done with inexpensive plastic molding. Quartz crystals are vacuum packaged in metal or ceramic enclosures. These are expensive, often dominating the final product manufacturing costs, and are not shared by the rest of the electronics industry.

SiTime's resonators are intrinsically stable. They are made from annealed silicon and silicon dioxide, nearly ideal materials to resist drift. This is compounded with the extreme cleanliness of the resonator chamber that prevents surface contamination and resultant de-tuning of the MEMS structure. Finally, the high-temperature annealing process eliminates any flaws, stresses and other mechanical defects that could affect stability.

Measurements at SiTime show resonator frequency drift of less than 0.05 ppm (parts per million) over two weeks at elevated temperature. Earlier investigations conducted at Stanford University demonstrated resonator frequency drift of less than one part per million over one year. These were measured without pre-annealing or pre-aging of the resonators. This is more stable than similarly treated quartz.

Quartz drifts primarily because it cannot be annealed at temperatures over 573ºC without undergoing a lattice change that renders it useless. This lower temperature tolerance requires that quartz crystals must be stabilized using more time-consuming, more costly, and less effective techniques (See the sidebar: "The Art of Stabilizing Quartz Crystals"). Quartz's lower maximum anneal temperature also makes it incompatible with the surface reformation that SiTime performs on its silicon resonators. Consequently, aging related frequency drift in quartz crystals is partially caused by changes in the mechanical properties of the crystal itself.

The ultra-clean EpiSeal encapsulated resonators have been tested for more than 300 temperature cycles from -50 to +80C with no discernable frequency shift or thermal hysteresis. Precision laboratory tests show the MEMS resonators have intrinsic hysteresis of less than +/- 0.05 ppm. In the case of quartz crystals, thermal hysteresis is caused by vacuum cavity contamination, support stress, and various poorly understood intrinsic effects. Common quartz AT-cut crystals in small packages typically show 0.1 to 0.5 ppm hysteresis.

Direct replacement of quartz and more
The SiT8002 programmable oscillators and the SiT11xx fixed frequency oscillators have a frequency range of 1 to 125 MHz and frequency error specifications of +/- 50 and 100 ppm over temperature, power supply, and aging. The programmable and fixed frequency parts are interchangeable, so designers can develop their circuits using the programmable versions and manufacture with the lower-priced fixed frequency parts.

These product specifications enable a number of present applications and future capabilities will support value-added applications including:
Consumer and computational products - Notebook computers, digital cameras, gaming boxes, video recorders, portable media players, set top boxes, high definition televisions, and printers all consume quartz products. For example, PC motherboards require numerous quartz crystals, quartz oscillators, VCXOs, and CMOS PLL chips.

Automotive applications - MEMS First silicon resonators are superior to quartz crystals in automotive applications due to their inherently better tolerance to extreme temperatures and vibrations. Since EpiSeal resonators are annealed at over 1000°C, normal operating temperatures have virtually no effect on them. Shock and vibration insensitivity are also improved because they have vastly higher fundamental resonant mode frequencies than mounted AT-cut quartz crystals. This makes them ideal candidates for use in engine and drive train control systems, telematics, security systems, and tire pressure monitoring.

Wireless applications - One of the early application targets for this technology is in ultracompact wireless nodes which integrate one or more resonators. Radio nodes in the 315, 433, 868, and 915 MHz range may all benefit from the < 20 ps RMS jitter and +/- 50 ppm performance of the first generation oscillators and achieve node space savings of greater than 50%. One or more MEMS First resonators may be integrated on a single die for wireless applications that require a 32.768 kHz or equivalent oscillator for a lowpower wake up and real-time clock functions, as well as a high-frequency oscillator for transmit, receive, and processing functions.

MEMS First resonators are compatible with standard CMOS processes and in the future may be integrated with PLLs, digital logic, and analog circuitry on the same die. Alternately, the tiny chips containing raw resonators can be attached directly to other ICs using standard bump-bond techniques and packaged as a single device. A single-chip technology that integrates both the MEMS resonator and drive electronics is currently under development.

A nod to Moore's Law
MEMS resonators are much smaller than common quartz crystals. Standard silicon fabrication techniques readily produce parts with sub-micron features and nanometer precision. A completed MEMS resonator is typically a few tenths of a millimeter across. In comparison, quartz crystals are typically a few millimeters across; about a hundred times larger area. Smaller parts mean smaller final packages - among SiTime's first products are the world's smallest programmable oscillators.

MEMS resonators improve in performance as they are made with finer geometries. As newer CMOS technologies scale down in size, MEMS resonators built in the same factories can be scaled as well, and they will have improved performance. SiTime's current resonators have electrode spacing of 0.4µm, which sets a limit on the amount of signal presented to the CMOS oscillator circuits. In future generations, finer geometries will reduce this electrode gap. This will increase the sense signals and improve the signal to noise ratios, giving the oscillators better phase noise and better jitter specifications. This is not true for quartz crystals. As quartz crystals are shrunk they perform worse, with lower Q, worse phase noise, more stress sensitivity, worse activity dips, etc. With electronics we expect devices to get cheaper as they get smaller. Packing more transistors onto the same silicon area decreases so as we drive to smaller products, quartz will perform worse and cost more, while MEMS will perform better and cost less.

The improved performance at smaller geometries will enable development of compact MEMS oscillators with higher operating frequencies and lower phase noise that will benefit applications such as cell phones, which rely on high-accuracy frequency sources. Researchers are confident that as MEMS-based timing devices mature, they will be able to meet even the stringent standards for GSM and CDMA cell phone TCXOs - most likely in 2008.

SiTime's unique MEMS technology has several advantages over quartz crystal technology. It allows smaller components, fit and functional replacements, reduced costs, as well as more advanced products in the future. Expanding capabilities will expand the addressable applications while decreasing cost will force a technological sea-change in the timing market.