Oscillation is an essential function for many common devices. Now a new generation of generators will extend their range of applications even further.
Crystal oscillators have for decades kept harmony and order on circuit boards using the piezoelectric characteristics of quartz. Quartz crystals flex when pulsed with electricity, a property that - with the right electrical feedback - makes them resonate at a precise frequency that depends on the crystal's size and shape.
However, much in the way silicon memory chips have replaced photographic film, new types of silicon timing-device may slowly be usurping crystal oscillators.
Mainstream crystal oscillators are low-cost and relatively energy-efficient. True, their response changes with temperature and age, but that can be sorted with calibration circuitry to provide standard devices offering anything from 5 to 100 parts per million (ppm) frequency error over temperature. On the downside, crystals cannot be integrated like transistors, and neither can they be combined with the processors and communications chips into which they deliver clocks and frequency references. Their sensitivity to noise means they need an exclusion zone around and below them to keep out interfering signals.
Looked at like this, using crystal oscillators is like having separate conductors with their own podiums for each section of an orchestra.
Smaller, smarter, more robust silicon timing devices are emerging from two technology camps. One uses vibrating microscopic electro-mechanical silicon structures (MEMS) such as beams, rings, and squares, mainly to generate MHz frequencies. The other uses resistor-capacitor (RC) and inductor-capacitor (LC) oscillators to provide kHz and MHz frequencies by either tuning these intrinsically variable oscillators against a more accurate external reference or using circuitry that precisely compensates for the oscillators' frequency fluctuations.
Both technologies involve carefully playing-off accuracy, power consumption, phase noise, and jitter. Phase noise is a measure of how much energy is expended at other frequencies offset from the centre frequency. Jitter is the rapid, short-term fluctuations of the signal waveform (which can be thought of as phase noise's time-based 'alter ego'). Jitter and phase noise influence signal-to-noise and bit error rate. As data transmission rates for wired and wireless networks have increased so has demand for flexible, low-jitter clocks.
MEMs resonators were described as far back as 1965 in a paper by HC Nathanson and RA Wickstrom. What has taken decades is getting the resonators into hermetically- sealed packages that are cheap enough for the mass-market. The problem has been that even small amounts of surface contamination can significantly change the MEMS resonator frequency. Over the last few years a number of companies including Discera, IDT, Sand9, Silicon Labs, and Silicon Time have nonetheless introduced commercial devices. One key step has been the ability to seal MEMS resonators inside tiny vacuum cavities within a silicon wafer.
MEMS resonators do not work on their own; they have to be packaged with a CMOS driver chip. Since MEMS stability varies with temperature, the chip includes temperature compensation circuitry (typically a temperature sensor, temperature to digital converter and fractional-N phase locked loop, which also generates the output frequency). MEMS companies do not integrate this chip with the resonator because the MEMS processing steps are harsh, and the driver chip can benefit from mainstream Moore's Law scaling.
The growing popularity of MEMS timing'devices became clear when Silicon Time, a spin-off from the automotive firm Bosch, became the first MEMS company to enter the top ten of basic oscillator suppliers as listed by CS&A, an US market research firm. (It has shipped 120 million devices since 2006). Silicon Time is making MEMS oscillators that generate output frequencies of up to 800MHz (depending on the specific device), and that can meet anything from 0.1 ppm to 50 ppm frequency error over temperature.
Its two main resonating structures are a crisscrossed square that vibrates at 5MHz and a new 48MHz circular structure that resembles a spaceship. "When we shift the raw frequency of a resonator from 5 to 48MHz, we get ten times improvement in phase noise and jitter," explains Piyush Sevalia, Silicon Time's vice president of marketing. As a result the company's latest MHz frequency SiT820X programmable oscillators feature 10ppm stability with sub picosecond jitter.
Most MEMS resonators are driven electrostatically by applying a DC bias voltage across a small gap between the resonator and its chamber. IDT, however, is using piezoelectric actuation technology licensed from the Georgia Institute of Technology. IDT's pMEMS function by applying electricity to a coating of aluminium nitride that has been deposited on top of the resonator beam. IDT quotes a frequency stability of better than '50ppm using this latest approach, and very high resonation frequencies at over 1GHz with sub-picosecond jitter.
Many MEMS timing devices are now sold as pin-compatible replacements for crystals, but there is no reason they cannot go in smaller packages according to Harmeet Bhugra, managing director of IDT's MEMS division. "IDT's resonators come in a wafer-level packages measuring 500x500µm so devices could be that small," he says. "Moreover, there are parts in the pipeline that integrate far more on chip. So within that space you could have products with multiple frequency outputs rather than a single reference."
While gaining ground in computing, networking, industrial, and some consumer applications, MEMs oscillators have not reached smartphones, which are prime targets for any crystal oscillator substitute. "There is about $1.50 or so of timing content on any smartphone. If you multiply that by 700 million or so smartphones, you have a $1bn market," explains Sevalia.
Standby for smartphones
Typically a smartphone will have one or two 3 x 1.5mm low-power 32.768kHz (usually referred to as 32kHz) quartz crystal oscillators, one mainly used for sleep-mode timing next to the baseband processor and the other commonly used to drive the power management chips. Each will occupy around 0.75cm2 of space including support capacitors and a signal exclusion zone.
There will also be higher frequency MHz crystal oscillators providing a clock for the application processor, an accurate ('2.55ppm) reference for the RF transceiver to meet the stringent RF compliance requirements of wireless standards and an even more accurate one ('0.5ppm) for the GPS chipset. MEMS devices could meet the frequency stability and phase noise specs for a phone's cellular and GPS references (currently handled by temperature-controlled crystal oscillators TCXOs), says Silicon Time's Piyush Sevalia, but they are still too power hungry. "We think we can cut our power consumption by an order of magnitude and we are working on that right now," he adds. This would bring the current drawn down to a few milliamps.
Designing a miniaturised substitute for 32kHz standby timing is more challenging. Such a device must only draw micro-amps of current, provide '100ppm or lower frequency error over temperature, and have jitter performance below 100ns. No MEMS device has yet come close to this specification. Low power CMOS timing reference chips built around RC oscillators are showing more promise although historically, these circuits consumed a lot of power.
This is because the output frequencies vary from silicon wafer-to-wafer and (while operating in a circuit) with temperature, age and mechanical stress. Because of this intrinsic variability, the chips need to integrate complex tuning circuitry to continuously correct for changes.
One workaround developed by Silego is to use another high frequency, higher accuracy crystal on the circuit board as an external reference and to periodically re-tune the on-chip RC oscillator with respect to it. As Silego's John McDonald explains"If it takes 1mA to run a MHz clock, if you only turn it on one thousandth of the time, you have only micro-amps of total power consumption."
By taking this approach Silego's 32kHz GreenCLK chips can deliver performance over temperature of 25 ppm to 30 ppm while drawing a minimum of 1.8µA. The company ships over a million of these per month. Silego's latest version, GreenCLK2, is a six-output 2 x 3mm device that includes a PLL for generating other frequencies from the MHZ input. "Our chip can take the signal from a single 25MHz crystal and use the PLL to generate, say, 27MHz for a graphics clock," says McDonald. "The phase noise is slightly reduced because the signal is going through the PLL but it is good enough for graphics. It uses the RC methodology to generate the 32KHz clocks." Six of the top ten consumer electronics companies are using Silego's technology in portable equipment such as tablets and notebooks. "Saving space is their main driver, but we also save them around 10 to 15 per cent of the existing bill of materials," says Silego's McDonald. For phones, the technology is less popular. "Tablets and notebooks have a lot of crystals and a lot of outputs so the value equation is much stronger than for a smartphone."
UK start-up EoSemi has gone back to first principles with its CMOS 32kHz ATOC (Accurate Timing Oscillator Circuit) chip to see how analogue circuits behave under mechanical stress and temperature. It has designed its ATOC chips specifically for smartphones and sees scope for including them in system-in-package assemblies with other phone chips. EoSemi is applying sub-threshold techniques to be able to fine-tune the ATOC's on-chip RC oscillator in real-time with minimal power. Circuits in sub-threshold mode use a supply voltage that is below the point where a transistor is supposed to turn off. Traditionally the threshold voltage represented a lower voltage limit but in fact the gate remains partially switched on and conduction continues in this region.
It is tuning against two sets of calibration codesone set relates to the oscillator's unique basic frequency and response to temperature measured during wafer test; the other accounts for ongoing changes in the silicon detected by on-chip temperature and stress sensors. The sensing technique exploits the different sensitivities to mechanical stress of metallic and semiconductor materials in terms of resistance. EoSemi has patented several parts of the design including the sensing circuitry and its approach to tuning (see 'View of making MEMS' panel, left).
The resulting chips are accurate to 30ppm and consume around 8'A active current. Compared with a 3x15mm sized 32kHz crystal, an ATOC chip will occupy at least 50 per cent less space - and this excludes the savings in 'exclusion area' on the PCB. Delivering such a circuit in a small die only became cost-effective when 180nm CMOS process technology moved to the mainstream, according Steve Cliffe, EoSemi's vice president of marketing.
EoSemi sees its design approach being equally applicable to making dual frequency chips with a 32kHz RC and a MHz frequency LC oscillator. This next step is 12 to 18 months away, according to Cliffe. "The challenge with two oscillators is solving the compensation for stress for both as each oscillator will be on a different part of the chip and undergoing different physical conditions," he says.
The market for silicon MEMS timing devices is tuning-up fast at 72.3 per cent CAGR, according to market research firm IHS ISuppli. Meanwhile, by combining its own figures with data from two market research firms, IHS iSuppli and CS&A, IDT predicts that by 2016, MEMS oscillators could take 45 per cent of the $446m standard crystal (XO) oscillator market, which is 5 per cent share of the total $4.06bn frequency control market (see market share breakdown, p90). Mobile phones makers are the ones to watch because when they abandon crystals, silicon timing will have moved centre stage.