MEMS resonators vs. crystal oscillators for IC timing circuits


by Sergis Mushell and Steve Ohr, Gartner

Since the early 1980s, companies have been trying to replace quartz with silicon MEMS-based oscillators as the frequency reference in clock and timing oscillators. Developments in semiconductor process technology, packaging, and the integration of circuitry have enabled some progress for MEMS resonators. The resonators are effectively time-base generators, or references, similar in operating principle to the mechanical tuning fork used to tune musical instruments. A separate electronic oscillator provides impetus that forces the crystal to vibrate at a precise frequency. Those vibrations are captured and output by a gain buffer, and a phase-locked loop (PLL) captures and distributes the reference signal generated by the resonator-oscillator combination.

SiTime’s 8003 low-power programmable oscillator.
Click here to enlarge image

One of the most highly touted features of MEMS resonators is the integration they promise for integrated circuit (IC) timing circuits. Fabricated in silicon with bulk etching processes, MEMS resonators can in principal be tied to oscillator circuits and PLLs on the same silicon substrate. This would allow the clock and timing generators–together with the resonator–to occupy a single low-profile semiconductor package. This package, moreover, would support high-volume assembly techniques. Even where MEMS resonators and oscillators are separate chips, they could occupy the same package–one that would be smaller and easier to handle than the metal cans that currently house crystal oscillators. Thus, silicon MEMS-oscillator combinations are promoted as replacements for crystal oscillators in computers, communications equipment and digital consumer devices (like set-top boxes).

The lure of integration

Revenue for IC timing devices, which provide clock and trigger signals for all sorts of electronic circuits, is expected to grow as clock frequencies increase and the sequencing of electronic events becomes ever more precise. The stability of these silicon timing circuits–which constitute an estimated $3 billion market–depends on outboard precision time-based generators, which today are largely quartz crystal.

Figure 1: Architecture of a typical clock generator. (VCO = voltage-controlled oscillator). Source: Gartner, Discera
Click here to enlarge image

A number of startup companies have proposed replacing outboard crystals with silicon-based devices, microelectromechanical systems (MEMS) resonators, which would enable coupling with IC timing circuits on a complementary metal-oxide semiconductor (CMOS) system-on-chip (SoC).

But despite significant venture funding, resonator manufacturers have not resolved key technical and manufacturing issues. Consequently, they will be slow to displace less-costly crystal-based timing sources in systems.

Barriers to adoption

The goal of single-chip integration (the “holy grail” for the semiconductor timing industry) would be to include the resonator, the oscillator, the PLL and a temperature compensation circuit (TCC) on a single silicon substrate. The current structure of silicon MEMS-based devices utilizes a stacked-die arrangement, housed in a multi-chip package.

Discera QFN package oscillator.
Click here to enlarge image

Single-chip integrations–and market acceptance of MEMS resonators–is impeded by four issues:

  • Manufacturing cost. The formation of a mechanical armature requires time-consuming bulk etching. In addition, the performance of the MEMS resonator is impacted by the presence of moisture in the MEMS cavity. Thus, specialized packaging and encapsulation are required.
  • Temperature stability. Quartz technology has been around for decades, and offers temperature and long-term stability, as well as high-frequency products (which are offered at a premium). First-generation MEMS resonators do not have the same temperature stability as quartz and require TCCs. This is not likely to occur (enough to generate significant revenue) within the next four or five years. So MEMS resonators are currently useful for lower-frequency timing references (kHz rather than MHz).
  • Low phase noise. MEMS resonators suffer from phase noise, which limits their utility in digital-communications applications. Higher market penetration could occur if these issues were resolved.
  • Power consumption. The high power consumption and initial frequency stability are major challenges for silicon MEMS in handheld products.

    Click here to enlarge image

    MEMS solutions currently have an initial frequency accuracy that can vary by as much as 100 parts-per-million (ppm) from their intended center frequency. Some manufacturers are claiming accuracy to within 25ppm, though this would require a costly hand testing and sorting process. Unlike quartz crystals, which can be scored with a laser to precisely trim the resonant frequency, MEMS cannot be fine-tuned. It is possible to utilize a “fractional-N” frequency synthesizer in conjunction with PLL to extract more precise timing signals from the MEMS resonator-oscillator combination, but this will raise the cost of the timing solution, as well as promote a very large die size for the integrated solution. MEMS resonators also will require digital temperature compensation circuitry, which contributes to die size. In addition, the compensation circuits make their own contributions to higher-phase noise and jitter; this makes MEMS timing devices less attractive to the communications market.

    Investments in MEMS resonator technology

    Since early 2003 there have been a slew of startups targeting the replacement of quartz as the sole frequency reference source available to the electronics industry. The quartz crystal is one of the last electronic components that have yet to surrender to silicon integration. The resonating quartz crystal provides the pulses for every electronic circuit, and is inside every electronic device and piece of equipment.

    Quartz has been used as a timing source since the early 1940s. No other material or technology has been able to replace the quartz crystal because of its exceptional temperature stability and low phase noise. Though companies such as Maxim and Micro Oscillator have offered “quartz-less” timing sources in the past, these products did not make any major impact on the marketplace, because of system requirements for precision and issues over their stability.

    Conventional beveling process (left) vs. Epson Toyocom’s QMEMS process (right). Photo: 2.0mm x 1.6mm AT-cut quartz crystal fabricated by QMEMS.
    Click here to enlarge image

    Among the newer startups, there are four main approaches to replacing quartz crystals. Since there is no potential for integrating quartz with silicon and quartz large packaging, startups believe that silicon resonators can solve these problems. The four main approaches are:

  • Silicon MEMS, with devices offered by SiTime, Discera, and Silicon Clocks;
  • A hybrid QMEMS approach combining quartz with micro machine technology, offered by Epson;
  • A self-perpetuating Mobius Loop offered by Multigig; and
  • Silicon RC oscillators offered by Silicon Laboratories and Mobius Microsystems. [Changed to distinguish properly between Multigig, a supplier of clock generators using "Mobius Loop" technology, and Mobius Microsystems, a manufacturer of CMOS oscillators said to compete with crystal based oscillators in size and performance. -- Ed.]

    Note that silicon-based MEMS (Si MEMS) are fabricated on silicon, while QMEMS are fabricated on quartz using a photolithography process. Regardless of their manufacturing technology, the value proposition offered by all these vendors is the same: integration with CMOS, ideally lower costs, and standardized IC packaging–which will enable not just the use of low-profile packaging, but also the use of standardized IC pick-and-place assembly equipment. Regardless of the technology approach, the underlying issues addressed are the same.

    The three startup companies using Si MEMS have all based their technology on research from universities and research labs. SiTime’s initial research and technology came from Bosch labs and Stanford University, and it has secured three rounds of funding ($12 million in December 2004, another $12M in March 2006, and $20M in May 2007). Discera’s research came from the University of Michigan; it raised $4M in July 2001, $12M in April 2004, and $18M in April 2007. Silicon Clocks’ research came from UC Berkeley; it secured initial $10M funding in 1Q05, and another $8M in 3Q07. Silicon Clocks also merged with SiSync, a clock synthesizer startup.

    Other technologies and vendors

    While MEMS technology is being targeted at the displacement of quartz, other companies including Silicon Labs, Mobius Microsystems, and Multigig, are proposing alternative timing circuits, such as free-running LC oscillators. Silicon area is a concern, though. We believe these will have limited success, at least in the near term. [Tone softened to reflect improvements vs. crystal-based forerunners, when packaging benefits are taken into consideration. -- Ed.]

    As startup MEMS vendors continue to sample parts, there have been few reports of success in any major applications that would drive high-volume production of MEMS resonators. One manufacturer claims to be sampling parts to a cellular handset manufacturer. There have also been signs of particular interest in this area from timing-circuit vendors, such as IDT, which supply the clock generators and clock buffers for PC, server and consumer markets. By incorporating the technology into its silicon timing solutions, IDT could become a large consumer of MEMS resonators; conversely, it could manufacture its own circuits, in competition with the startups–or perhaps make an acquisition. Either scenario could result in more development and deployment of silicon MEMS technology.

    Assuming vendors can deliver on their promises, the market for MEMS resonators could grow to $100 million in 2012. Wide adoption could start in late 2009 or early 2010 if certain technical problems are resolved. Improving the accuracy and stability of MEMS resonators would increase the number of applications and markets that could successfully use them.

    Sergis Mushell is a principal analyst with Gartner’s Technology and Service Provider Research group, with primary focus on the semiconductor, storage, and industrial markets.
    Steve Ohr is research analyst for analog and power management ICs. He tracks the market forces propelling the growth of analog semiconductor use.