comment on this article

Quartz faces losing its lock

Wherever you go in electronics, you cannot avoid silicon, although its pre-eminent status is coming under threat. Its amorphous oxide is only just beginning to be displaced from leading edge integrated circuits as the gate material of choice. But, in the meantime, its crystalline form is still going strong as the primary timekeeper for electronics.

Unlike glass, quartz is piezoelectric – distorting when a electrical pulse is applied to it. And its distortion leads to the generation of an electrical signal. Oscillators use the two sides of the piezoelectric effect to induce resonance in the crystal – feedback forces the crystal to oscillate within a very narrow range of frequencies. The higher the resonance, Q, the narrower this range. By cutting and shaping the crystal, you can control this resonance frequency.

Low frequency crystal oscillators (XOs), such as the 32.768kHz generators used in digital watches and many consumer products, often used a tuning fork shape: one end is fixed; the other flexes. The frequency happens to be useful because a simple divider circuit can reduce the frequency to 1Hz: 32,768 is 2^15. An AT-cut crystal – one of the most common shapes in use for megahertz-frequency clock generators – is shaped more like a lozenge. It needs to be cut at an angle of 35° through a grown quartz rod to have the right crystal orientation to achieve its target frequency and reasonably good performance over changes in temperature.

Without external assistance, the quartz crystal would never be able to meet the stringent demands of GPS or even mobile-phone networks. The problem for quartz crystals is that their performance is limited by the physics of a resonating block of material. Quartz crystals can run comfortably at tens of megahertz and can offer outputs in excess of 100MHz range by using overtones of the fundamental frequency, although these can prove unstable. As a result, most crystals today simply provide a reasonably stable frequency source for phase-locked loops (PLLs) that then generate a much wider range of frequencies, although at the risk of increased jitter – small shifts in the clock edges seen as phase noise in the frequency domain.

The use of quartz oscillators and clock generators is complicated by the sheer variety of timing requirements across the range of electronics systems. A GPS receiver relies on a precise frequency source – accurate to 1ppm – or it will refuse to lock to any transmitting satellite in the system. But accuracy of less than that is just fine for the system clock of a PC. Many of these products will introduce artificial jitter – using spread-spectrum clock generators to reduce electromagnetic interference (EMI).

In the quest to build high-accuracy oscillators, the quartz industry has spawned a dizzying array of four-letter abbreviations. A regular XO is not good enough to support the demands of a GSM handset, for example. Initially, designers turned to the temperature-compensated crystal oscillator (TCXO). Traditionally, this uses a thermistor network to generate a voltage that compensates for the crystal's drift – usually around ±30ppm over the standard commercial temperature range.

In the past decade, handset designers have gradually replaced TCXOs with digitally compensated devices – DCXOs. Typically, the digital compensation circuit lies on the transceiver chip – often using a sensor-driven polynomial generator – and is fed by a standard crystal oscillator. Digitally compensated or microprocessor-controlled oscillators have appeared for systems where an on-chip compensation circuit is not available. These devices can also correct for long-term ageing effects, although a good AT-cut crystal uses demonstrates changes of less than 1ppm per year.

For greater stability, compensation is not accurate enough. The answer is to overcome variation with temperature by simply keeping the temperature constant. These oven-controlled crystal oscillators (OXCO), generally used in tethered network equipment because of their higher power consumption and areas of up to 5cm2 – larger crystals provide a more stable resonator. But these devices can produce a frequency that is stable in the range of parts per billion and so can be used in the satellites that generate GPS signals, rather than the DCXO or TCXO-driven units that receive the incredibly sensitive timing messages.

The tradeoff between size and performance is one of the reasons behind the search for an alternative to quartz. Smaller crystals are, in general, more difficult to handle, while higher frequency crystals have to be thinner, making them more delicate, which undermines manufacturing yield. As a result, a trend towards using very small crystals in portable devices – such as phones – has stopped, or even reversed, because of reliability issues.

To achieve stable operation, the quartz oscillator circuits needs to be protected from interference. This calls for the exclusion of most circuitry from the area that surrounds the crystal package and its supporting passive components – including the PCB layers underneath them. And there are issues with power consumption. Power consumption during standby time is now one of the most important concerns for designers working on low-energy systems. But, when almost everything else in a circuit has been put to sleep to conserve battery life, you still need a clock source for the small number of peripherals that remain active. The constant beating of the quartz resonator then becomes a major proportion of system power consumption.

Some companies have seen these limitations of quartz as an opportunity for IC-grade silicon to make inroads into the timing market. The oldest alternative is the on-chip resonator – sometimes used as a low clock-rate backup on microcontrollers for situations where the main crystal fails. The accuracy of these on-chip oscillators, traditionally, has not been good. Accuracy is often measured in percentages rather than parts per million: it is hard to produce a consistent clock output because of the variation inherent in chipmaking.

Rather than use electronic oscillators, attention first settled on micro-electromechanical systems (MEMS) technology (see fig 2). The reason for picking MEMS is that it offers the possibility to use photolithographic techniques, adapted only slightly from regular CMOS processes to build sources at much lower cost than the mechanical processes used to make most quartz sources.

As most resonators need some sort of signal conditioning and compensation, it is possible to use a CMOS process to integrate the two in one chip. However, manufacturers have tended to keep the two components separate to minimise the technology risk. For example, MEMS devices do not like moisture. As a result, they often need to be kept in hermetically sealed packages which are more expensive than those used for conventional CMOS chips – and the CMOS signal-conditioning die can be a lot larger than the MEMS portion, which is relatively simple in terms of mechanical structure. The mass of a resonating beam may be just 10^-13kg – a single molecular layer of water could alter its resonating frequency by 100ppm. The advantage of the MEMS resonator is that it is much easier to size and shape the resonating beam to generate a target frequency with a given Q. The size can also helps with reliability in shock and vibration-prone conditions: environments that can be tough on quartz crystals.

In principle, MEMS devices will perform better at higher frequencies, where quartz crystals need to use overtones, and in applications where the ability to program a target frequency digitally is needed. Temperature compensation is typically easier for a MEMS oscillator: the thermal response of the resonating beam tends to be more linear than is the case with quartz.

The piezoelectric effect could also turn up in MEMS-based devices. Instead of relying purely on the resonance of a silicon cantilever, this device uses a thin layer of piezoelectric, such as aluminium nitride or zinc oxide, to help constrain the frequency of vibration. This technique could be used to increase the Q of the oscillator, helping push MEMS devices into competing with OCXOs or to relax the need to seal the package hermetically – in exchange for a more modest Q value.

Some manufacturers are looking to MEMS technology to provide a boost for the quartz crystal resonator itself. Using photolithography, they etch the crystal into more complex shapes, such as the bi-mesa. Epson-Toyocom gave these devices the moniker QMEMS.

Silicon has been called in to give quartz another helping hand through phase-locked loop (PLL) products designed to reduce the jitter produced by a standard crystal. The crystal is used primarily as a frequency reference, feeding into a circuit that synthesises a higher and more stable oscillating output.

The next step is to remove the mechanical or piezoelectric resonator entirely and to use an electronic resonator. Rather than invent a new type of resonator, designers have gone back to the sloppy inductor-capacitor (LC) 'tank circuit' (see fig 3) favoured by low-speed microcontrollers and worked on ways to reduce a jitter than can be as bad as 15,000ppm by several orders of magnitude.

Because everything can be integrated onto one chip, the all-silicon LC oscillator promises to be the heart of the smallest possible clock generator. The LC resonator can run at gigahertz frequencies – these high frequencies are favoured by the comparatively low inductances and capacitances that are available to CMOS designers. The output frequency is usually many times lower than that: usually in the megahertz range. The key to the improved performance, however, lies in the signal conditioning and compensating circuitry that overcomes the inevitable manufacturing variations encountered in CMOS processing and the complex dependence on temperature that a tank circuit can exhibit.

During test and calibration, the changes in frequency over temperature are used to calculate a set of polynomial coefficients that can then be used to adjust the output frequency as thermal conditions change. For process variations, a bank of varactors programmed using calibration data stored in on-chip non-volatile can provide the necessary voltage bias.

Chris Edwards

Comment on this article

This material is protected by MA Business copyright See Terms and Conditions. One-off usage is permitted but bulk copying is not. For multiple copies contact the sales team.

What you think about this article:

Add your comments


Your comments/feedback may be edited prior to publishing. Not all entries will be published.
Please view our Terms and Conditions before leaving a comment.

Related Articles

FD-SOI breakthrough

CEA-Leti and Dolphin Design have announced the development of an adaptive ...

A question of scale

Is Moore's Law dead, dying or in rude health? It depends who you ask. In turn, ...

NI Trend Watch 2014

This report from National Instruments summarises the latest trends in the ...

Capactive sensing

This whitepaper looks at a number of capacitive sensing applications to ...

Digital consciousness

​Would you consider uploading your brain to the cloud if it meant you could ...

The changing face

A year of seismic social and political change 1968 saw anti-Vietnam war ...

The project begins

Recently, Stephen Doran took up his position as CEO of the Compound ...