Almost every electronic device needs a clock source. For example, a microcontroller (MCU) uses an oscillator to advance the next instruction, and a radio needs an accurate oscillator to mix radio-frequency signals to baseband for processing.
The advent of smart, connected appliances has created higher demands on clock performance. This article explains how the designer can meet these challenges while reducing technical risk, design time and bill of materials. We look at the options of quartz crystals, quartz crystal oscillators (XOs) and highly-integrated clock solutions using quartz and MEMS-based technology.
Smart, connected devices need sophisticated clock trees
MCUs often include internal RC phase-shift oscillators for non-precision computing applications. These oscillators use an integrated resistor-capacitor pair to create the time constant that governs the oscillator frequency. Such oscillators have an accuracy of about 1% and exhibit high jitter (unwanted random fluctuations in the timing of the clock transitions). They are suitable for applications in which transition timing is not critical, such as clocking an MCU for computation and driving a simple numeric seven-segment liquid crystal display (LCD). The display requires multiple clock waveforms; however, transition timing tolerance is a few milliseconds. UART communication up to a few Mbit/second is also possible, with timing tolerances of hundreds of nanoseconds, but this represents the limit of a simple RC oscillator.
Smart, connected products communicate over a network to the cloud, via Bluetooth, wired Ethernet, Wi-Fi or other connectivity protocols. The inclusion of radio and/or high-speed data drives the need for precision clocks with an accuracy of a few parts per million (ppm) and low jitter.
A key component required in the generation of a precision clock is a stable reference frequency, and this requires a resonator. A resonator is an electrically passive device that naturally oscillates with larger amplitudes at certain (resonant) frequencies than others – the strings of a violin are a simple example. Common choices for electronic devices are quartz crystals and MEMS resonators. The requirements of a resonator are:
- Stability of the resonant frequency over time and temperature. This avoids clock frequency drift.
- A high quality factor, or Q, that ensures resonator response to only a very narrow band of frequencies.
- Ability to operate at a high signal level, which supports a good signal to noise ratio at the output.
Items two and three are key to ensuring a clock signal with low jitter, enabling stable timing transitions.
Since a resonator is a passive device, it requires controlled energy to oscillate and create the reference frequency. Coupling a resonator to a sustaining amplifier in a feedback configuration achieves this stable oscillation. Quartz crystals or MEMS resonators with the appropriate amplifier are very suitable as frequency references for data transmission in the 10 Mbit/s and above domain.
Quartz resonators have a high Q and high-output capability, and are suited to applications where jitter must be extremely low. 100 femtoseconds of phase noise (measured in the traditional 12 kHz to 20 MHz bandwidth) is achievable. MEMS resonators operate over extended temperatures with very stable frequency, offer very high reliability, are resistant to shock and vibration and enable very small clock solutions, close to 1 mm square. MEMS resonators have high Q with lower output; 500 femtoseconds phase noise is possible, and new resonator designs are pushing this lower. For many modern networking applications, such as PCIe, for example, a smaller integration bandwidth applies, and both technologies are very suitable.
Implementing a clock in an embedded system
In an embedded system, there are three common implementations of resonators to produce a clock signal.
- A quartz crystal connected directly to a “target SoC,” (which is to be clocked)
- A quartz crystal oscillator or XO, creating one clock output for the system as a whole
- A quartz or MEMS-based clock generator (creating one or multiple clock outputs, at both low and high [>50 MHz] frequencies)
Implementation 1: Crystals connected to the target SoC
In this case, the System on Chip (SoC) is designed with a sustaining amplifier and the crystal can be connected directly, usually with capacitors to arrange the right feedback and tune the frequency. The SoC amplifies and may frequency-translate the reference clock, according to its needs. Diagram 1 shows the schematic of both a high-frequency and a low-frequency crystal connected directly to an MCU.
Two crystals connected directly to an MCU, showing loading capacitors and series resistors.
A system requiring only one or two crystals connected in this way is cost-effective; PC board layout is easy with the crystal placed immediately next to the SoC, generally avoiding issues with signal integrity and electromagnetic interference (EMI).
However, there are some caveats to consider:
- The crystal must be carefully chosen to be compatible with the SoC’s internal sustaining amplifier circuit. If the equivalent series resistance of the crystal is too high compared to the negative resistance of the amplifier, the oscillator may fail to start.
- The crystal will likely require loading capacitors to ensure that the feedback is correctly phased, as well as setting the frequency accurately.
- Quartz crystals have a relatively large temperature coefficient. Applications requiring operation outside -40°C to 70°C may need to use a temperature-compensated crystal oscillator (TCXO) or an integrated MEMS-based clock.
- Standard quartz crystals operating in a fundamental mode have resonant frequencies at or below 50 MHz. Quartz crystals operating above 50 MHz in an overtone mode tend to be more expensive.
Implementation 2: Crystal Oscillators (XO)
An integrated crystal and sustaining amplifier in a single package is known as a crystal oscillator or XO. Diagram 2 shows how the crystal wafer blank is combined with an oscillator Application-Specific Integrated Circuit (ASIC) in a hermetic assembly. This pre-packaged solution, although more expensive than a single crystal, avoids the interfacing issues of caveats one and two above, so that reliable start-up and correct output frequency is assured.
Crystal oscillators consist of a quartz crystal blank, traditionally inside a ceramic package with a metal lid.
Again, a system requiring one or two XO’s can be cost effective. If multiple frequencies, additional buffered outputs or frequencies above about 50 MHz are required, the system can benefit from an integrated clock generator.
Implementation 3: Clock generators
A highly-integrated clock generator includes some or all of the following functions in a single package:
- The resonator, such as quartz or MEMS
- The sustaining amplifier
- Phase-locked loop(s) for frequency multiplication to one or more frequencies, usually in the 1 MHz – 1 GHz range
- One or more buffers to provide multiple copies of the same clock frequency
Diagram 3 shows the integration of all the functions schematically.
An integrated clock generator combines a MEMS (or crystal) resonator with an oscillator, and extends functionality with a programmable PLL and buffer output stage.
The clock generator shown in Diagram 4 consists of a molded plastic QFN package containing an integrated crystal and an integrated ASIC including the sustaining amplifier, phase-lock loop, two programmable dividers and five output buffers. A one-time programmable (OTP) ROM stores the customized configuration (for example, the frequencies and output protocols). The design does not use the traditional quartz XO approach of a crystal blank sharing a hermetic package with the ASIC; instead a fully-assembled hermetically-packaged crystal shares a lower-cost plastic enclosure with the ASIC. The separate packaging of the crystal isolates it from the ASIC and substrate, maintaining pristine clean conditions and avoiding any contaminants on the quartz surface, achieving very good frequency stability over time (also termed “excellent aging performance”).
Benefits of clock generators
While a crystal or XO solution can be cost-effective when only one or two frequencies and single copies of a clock signal are required, more complex systems can benefit from using a clock generator.
One of the most obvious benefits is lowering system component count and the overall Bill of Materials (BOM). Diagram 5 shows a 10 gigabit Ethernet switch requiring multiple clocks, above and below 50 MHz. The combination of multiple XOs, a phase lock loop and buffers can be replaced by an SM803 clock generator (requiring an external crystal) plus a DSC400 with integrated MEMS resonator.
Block diagram of a 5-output clock generator and photograph of the crystal mounted on a substrate with the integrated circuit before the assembly is encapsulated in a plastic package.
Additionally, an integrated clock solution is flexible and allows different frequencies and output protocols to be selected by programming an OTP ROM. Sometimes, I2 C or SPI inputs or hardware pin-control are included to allow the designer to adjust frequencies after the clock is installed in the embedded system. This is especially useful for systems with selectable settings, and can also be used for system margining – specifically speeding the clock for maximum performance under special conditions.
Recently, an automotive manufacturer contacted me with an EMI problem; the fast transition times of our programmable clock generator used to operate a surround-vision camera were creating harmonics which interfered with the car’s FM radio. However, our clock generator had been designed with variable output rise/fall transition times, and the customer was able to avoid expensive screening or hardware changes simply by using our re-programmed version of our product. Product delays and cost overruns were avoided, because of the flexibility of the clock design.
An important function often available today in clock generators is the ability to “spread” the output clocks (use spread spectrum). Spread spectrum applies a carefully designed frequency modulation of the clock output, small enough not to impact end-system performance, but sufficient to spread the sharp spectral peaks of a fixed clock output to a narrow distribution of frequencies several decibels lower. This reduction of peak spectral levels can prevent EMI problems. The spectra of spread and unspread clock signals are shown in Diagram 6.
Reduction of EMI can be achieved by modulating the clock signal and reducing the energy peak (spread-spectrum modulation).
Clock generators can also provide significant performance improvements. A “discrete” approach to generating multiple clock outputs, using a single oscillator followed by buffers and phase-lock loops, involves routing multiple clock traces on the PC board. These traces are susceptible to crosstalk and reflections; by contrast, the multiple clock signals leaving the pins of a clock generator are clean, with low relative skew and matched rise/fall time. For a complex clock tree, therefore, the use of a clock generator enables clean clock generation with reduced technical risk, design time and board space.
Designers used to placing crystals next to SoCs may be concerned about routing longer traces from a central clock generator and the possibility of signal degradation and EMI. When properly executed, however, with carefully designed transmission lines, proper termination and correct use of PC board stack up, a centralized clock solution avoids these issues and saves cost.
With the use of a clock generator, multiple components are replaced by a pre-designed packaged clock generator, produced in very high volume to stringent reliability and quality standards. This assures higher reliability and lower total cost of ownership of the whole system. Field returns are also minimized.
Architecting the clock tree
When beginning to architect the clock tree, it’s important to take a broad view of the entire end system. Diagram 7 illustrates this big picture approach, showing the available component options. Examples of the types of questions to ask yourself include: How many different frequencies, and how many copies of each, are needed?
Example list of all the clock signals required.
Consider which frequencies need to be synchronous. The Ethernet switch in Diagram 5 is an example of a free running clock tree. Independent channels of data do not need to be clocked synchronously. A single channel of data that is processed at different clock rates, when latency in the data flow can be managed by buffering, is also a free-running or independently clocked system.
A system using a single precision clock, replicated in multiple locations with tightly controlled phase locking, is a synchronous system. This design is common in high-speed data transport, where low latency is key. For these systems, two additional components are common.
The first is a jitter blocker , an integrated phase-lock loop with a narrowband loop filter. It does not generate a clock signal but is used to remove jitter from an existing clock. Additionally, low-additive jitter buffers enable multiple copies of an existing clock to be produced.
Diagram 8 illustrates a synchronous clock system.
A synchronous clock system.
It is crucial to research and understand the specifications of different clock signals. The most important electrical parameters are shown in Table 1. An understanding of frequency accuracy and allowable jitter is key. Jitter is measured in several different ways for different applications and must be well understood.
Once the frequencies and number of outputs are defined, there are many other clock parameters to be considered.
Distinct from the electrical parameters, special situations may drive choices. Form factor and temperature range are two examples of factors that often impact product selection.
Save time with tools
Highly-integrated clock generators must be easy to use, and many clock manufacturers offer online tools to help the designer. Two examples of these tools are Microchip’s “ClockWorks Configurator” and TimeFlash. ClockWorks allows the purchaser to specify customizable products, receive a custom datasheet and order samples. TimeFlash is a desktop programming tool, enabling “blank” programmable clocks to be programmed at the customer’s facility, requiring only the TimeFlash2 programmer and a PC. Tools like these greatly simplify the selection and specification process and enable system designers to obtain or create samples quickly for testing in their embedded system.
Many of today’s smart, connected systems can greatly benefit from a highly-integrated clock solution. While transitioning from multiple oscillators and architecting a centralized clock involves careful planning and layout, it can reduce cost, improve performance and add functionality and higher reliability. Adding a highly-integrated clock to your application is simpler than you think. So, is it time to integrate your clock solution?