Part 1 of this article described basic requirements for timing devices and listed classes of oscillators for different applications. Table 1 compares the performance of silicon MEMS and quartz oscillators, considering many of the parameters discussed in Part 1. In Part II, we go into more detail on several important concerns for high performance oscillators: temperature response, frequency control and addressing EMI reduction. We also cover practical considerations such as board design, product lead times and cost.
Table 1. Comparison of silicon MEMS and quartz oscillators.
Resonators expand and contract due to changes in temperature, affecting their resonating frequency and making temperature compensation critical to oscillator performance for demanding applications. Although quartz has a very low coefficient of thermal expansion, variation due to temperature change is still a large component in the frequency stability of quartz oscillators. Overall frequency stability of fixed-frequency quartz oscillators is ±20 to ±50 ppm without temperature compensation (See Figure 1).
Figure 1. Frequency stability for a SiTime ±25ppm rated MEMS XO and a ±25ppm rated quartz XO over a standard industrial temperature range.
The coefficient of thermal expansion of silicon is an order of magnitude higher than that of quartz, so temperature compensation is built into the oscillator circuitry of MEMS oscillators. MEMS oscillators incorporate a temperature-to-digital converter (TDC) to automatically correct for frequency variations of the oscillator due to temperature. Since all temperature compensation functions are integrated within the existing oscillator circuit, no additional components are needed.
As seen in Figure 1, SiTime's ±25ppm rated MEMS oscillators have better margin at both low and high temperature compared to ±25ppm rated quartz oscillators. MEMS TCXOs are available for applications demanding more precise frequency stability. Typically, MEMS TCXOs have more complex and higher performance compensating circuits, and also require higher order calibration algorithms as well as more extensive testing than MEMS XOs.
STABILITY FOR TELECOM AND NETWORKING APPLICATIONS
Telecom applications require an extremely stable local clock to keep time in case the reference clock is temporarily unavailable. The internal timing circuit needs to provide holdover for 24 to 48 hours when the system cannot communicate with a GPS satellite, for example.
Keeping the clock at a constant temperature is one way to achieve the precise frequency stability required for these applications. One way that a quartz-based TCXO accomplishes this is to add a local temperature environment called an ovenized compensation. The oven temperature is chosen at a flat point in the frequency versus temperature curve of the crystal, further improving frequency stability. This approach is very effective, enabling frequency stability on the order of 0.05 to ±0.5 ppm, but the heating chamber occupies valuable board space and consumes additional power. Another drawback is the time lag for the heater to reach steady state temperature after the device is powered on.
It is possible to address the need for ultra high precision frequency stability without a heating chamber. MEMS technology has evolved significantly, reaching OCXO-level performance without the added oven compensation. The oscillator is not physically heated, but the advanced architecture of the temperature compensation circuit, combined with more complex calibration algorithms and testing, achieves the same level of frequency stability as though the oscillator were held at constant temperature. Figure 2 shows the frequency stability of such an oscillator exposed to temperatures ranging from -40 to +85°C.
Figure 2. Silicon MEMS temperature-compensated (TCXO) stability. Source: SiTime bench characterization.
Voltage-controlled oscillators (VCXOs), in which output frequency changes with input voltage, allow fine-tuning of the frequency for improved clock synchronization in telecom, broadband and other applications. Control voltage is usually controlled by an analog circuit. MEMS oscillators have the additional option of digital voltage control, which makes it easier to control the input voltage.
Frequency can be controlled over a limited range of input voltage, and the amount by which the frequency can be pulled is limited by the frequency stability of the oscillator. Improved frequency stability provides greater absolute pull range (APR). Pull ranges are commonly up to ±200 ppm for quartz oscillators and can be as high as ±1600 ppm for MEMS oscillators.
Ideally the output frequency varies linearly with supply voltage, but oscillators do not exhibit exactly linear responses. Any deviation from linearity results in a change in KV, the slope of the frequency versus voltage curve. Since KV affects important PLL performance parameters such as bandwidth and phase margin, any variation in this parameter complicates PLL design. Linearity, expressed as percent deviation from a straight line, is typically around 5 to 10 % for quartz and less than 1% for MEMS oscillators.
Noise caused by electromagnetic interference (EMI) can be a problem with electronic applications. For example in mobile computing environments, due to the very dense circuit boards, radiated and conducted EMI can result in the system not passing environmental compliance tests (such as FCC Class A or Class B). If the PCB is not laid out optimally for EMI compliance, the harmonics that are produced by the oscillator devices can be a major source of undesirable EMI. Shielding reduces EMI, but this is expensive and not feasible for portable devices.
Filtering is another option, but it requires design and possibly board layout changes, delaying the product release. If EMI issues are discovered late in the design cycle, this can be very costly. One solution is to replace the existing oscillator with a programmable spread spectrum MEMS oscillator (SSXO). This approach reduces the radiated energy that the oscillator generates without expensive board redesign since the new oscillator is a drop-in replacement for the existing component. Many microprocessor, microcontroller and memory applications use SSXOs.
SSXOs work by modulating the clock output frequency slowly over time with a triangular waveform to reduce the peak spectral energy. The modulation frequency is typically around 30 kHz. The two options for spreading the energy across a wider range are center spread (±0.25 to ±2.0 % of the frequency) or down spread (-0.5 to -4.0 %) modes. Spreading out the energy spectrum significantly reduces peak amplitude, as seen in the spectrum plots of a standard MEMS oscillator and an SSXO with -2% down spread in Figure 3. This peak energy amplitude suppression applies not only to the fundamental frequency of the clock but also extends to the harmonics, effectively reducing EMI for all components of the oscillator output. Since the amount of EMI reduction is proportional to frequency, higher frequency harmonics experience even greater reduction than the fundamental frequency. However, due to the spreading of energy, SSXOs cannot be used in certain applications such as wireless and high data rate Serial I/O.
Figure 3. Noise reduction benefits of a spread spectrum clock.
PRACTICAL DESIGN CONSIDERATIONS
The performance of each component in a system is important, but so is the overall board layout. In a typical MEMS oscillator package, the resonator die is placed on top of the oscillator IC and connected with wire bonds. One important benefit of using MEMS oscillators is that they are available in many standard plastic package sizes, including 2.5 x 2.0, 3.2 x 2.5, 5.0 x 3.2 and 7.0 x 5.0 mm. This gives the designer flexibility in choosing whichever package will fit best into the board design.
If a MEMS oscillator is to replace quartz in an existing design, it is best to choose a package that matches the footprint of the quartz oscillator. If the goal is to shrink the size of the entire board during a redesign, or if design changes for other components limit the amount of space available for the oscillator, it may be possible to switch to a MEMS oscillator in a smaller package while maintaining or improving performance and features.
Some applications require ultra-thin packages. A quartz oscillator may be too tall to fit within the required form factor, and a standard MEMS oscillator may be as well. One option to accommodate the height restriction is to choose a MEMS oscillator design where the resonator is attached next to the oscillator IC rather than on top of it. This design fits into a package that is only 0.25 mm high, as compared to 0.75 mm for standard plastic packages.
PRODUCT LEAD TIME
While designers are mostly focused on the performance and the feature sets of components, it is important to also consider the availability of components to improve overall time to market for the final product. For example, the lead-time of oscillator components depends on oscillator category and desired functions.
Quartz crystal oscillators need to be precision cut for the specific frequency desired. It is not practical or cost-effective for manufacturers to store inventory for all possible frequencies, so they generally stock a limited number of standard frequency crystals. It may be possible to find a manufacturer that carries the frequencies you need, with phase noise, operating voltage and package footprint that match your requirements. In this case, lead time is fairly short, but it still may require buying from multiple suppliers to purchase all the oscillators needed for a complete system.
If your oscillator specifications do not match up with stock in inventory, it is necessary to purchase oscillators made to order. As the cutting step that defines the operating frequency has to be done at the beginning of the manufacturing process for quartz oscillators, the entire process cannot begin until the order is received. Lead times tend to be on the order of 6 to 16 weeks. Since MEMS oscillators use the same resonator and frequency is programmed through the CMOS oscillator chip, the lead-time is much shorter. Manufacturers of MEMS oscillators can configure products to meet customer specifications with production lead times of typically 2 to 5 weeks.
For an application that does not have stringent requirements and can use an off-the-shelf quartz oscillator, the up-front component cost will be low. If the oscillators need to be special ordered, however, that adds cost, as does the addition of high performance features. This is especially true in the case of quartz oscillators. Since MEMS oscillators are programmable, they require no additional processing steps or components to add features, and the incremental cost will be much lower.
The up-front cost of an oscillator component is not the only factor that affects total cost. If initial testing of prototypes indicates a need to add oscillator features to improve timing performance, it streamlines the process and saves money if the new oscillator can be a drop-in replacement for the existing component. This minimizes the need to redesign the board to accommodate an oscillator with a different footprint.
There are many factors to consider when choosing timing components. For high performance applications, selecting an appropriate frequency is not sufficient. It is also important to consider the performance of the oscillator over expected operating temperature range and how the oscillator will contribute to the overall jitter budget. Special features such as temperature or voltage compensation, spread spectrum, or multi-frequency capability may improve performance sufficiently to justify added cost. Finally, the timing component needs to fit within space constraints imposed by board layout and be available within a time frame that does not adversely affect product launch. MEMS-based oscillators can be programmed with a wide range of features, making it possible to select the right timing device for any application.
About the Authors
Steve Pratt holds the position of Director of Marketing at SiTime. Steve has been in the semiconductor business for more than 20 years and prior to SiTime, has worked for various analog semiconductor companies including Maxim, Micrel, and Monolithic Power Systems. Steve holds a BS degree in Industrial Engineering and Technology.
Mehdi Behnami, Director of Product Marketing at SiTime, has over 20 years experience in the semiconductor and electronics industry. Prior to SiTime, Mehdi held management and engineering positions spanning marketing, applications and research. Mehdi received his Master’s degree in Electrical and Computer Engineering from the University of Iowa.