Communications applications, fueled by the Internet and e-commerce expansion, are emerging as the driving force in oscillator development. The need for ever-wider data "pipelines" brings along the demand for ever-higher reference frequencies. Unfortunately, at roughly 40 MHz, conventional crystal oscillators hit a frequency ceiling.
A crystal's frequency limitation stems from the relationship between resonant frequency and crystal thickness. The ratio is approximately T = 60/F, where T is crystal thickness in thousandths of an inch and F is fundamental frequency in megahertz. The thickness of a crystal designed for 40 MHz is therefore 60/40 = 0.0015 in. = 1.5 mil. Crystals thinner than this 1.5-mil threshold are too fragile to withstand the mechanical stresses of oscillator production.
Several available techniques outflank the 40-MHz limit. Using these methods, commercial crystal oscillators can generate outputs of up to several hundred megahertz. Yet there's still a significant gap between the crystal-stabilized reference frequencies available and the needs of today's Gigabit Ethernet, ATM, and SONET systems. In these applications, the datacom equipment builder must multiply, or upconvert, the crystal-oscillator reference frequencies to reach the gigahertz range.
Frequency multiplication tends to increase jitter as well as frequency, however. To minimize jitter at the gigahertz level, equipment builders have to start with super-low-jitter reference oscillators. But the higher the reference frequency climbs, the lower the final gigahertz signal's jitter. Naturally, crystal-oscillator manufacturers are developing generations of clock oscillators that raise frequency while simultaneously reducing jitter. Doing so requires advances in quartz-crystal technology, oscillator circuit design, packaging, and manufacturing (see "Oscillators Are More Analog Than Digital," p. 118).
Those manufacturers have long depended on two basic techniques, overtone and multiplier, to overcome an ordinary crystal's 40-MHz frequency ceiling. In the past year, a third and until now theoretical concept—inverted mesa—has proven eminently practical for achieving high oscillator frequencies (Fig. 1).
The clock oscillators rooted in frequency-multiplier techniques have long been the workhorses of modern time- and frequency-reference applications. These very economical devices use custom ASICs for crystal excitation, frequency multiplication, and output waveshaping. They're available for wide temperature ranges and achieve room-temperature stability to 10 ppm without compensation.
Multiplier-based oscillators also can operate over the −40° to +85°C extended industrial temperature range. They're available in through-hole and robust surface-mount packages. In addition, they can be provided for commercial off-the-shelf (COTS) use.
Phase-locked loops (PLLs) provide the basis for frequency multiplication in commercial oscillators. The PLL configuration is based on a simple voltage-controlled oscillator (VCO).
A 1/N digital divider is interposed between the VCO's output (FOUT) and the phase detector's input (FOUT /N). An error signal locks the VCO's output frequency to N times the reference frequency input so that FOUT = N ×FIN. Because the VCO operates at frequencies beyond crystal stabilization, it lacks crystal stability. The multiplier's VCO contributes phase noise, or jitter, to the output signal, resulting in jitter measurements of about 10- to 15-ps rms for multiplier-based oscillators. As a result, these oscillators can't satisfy the low-jitter requirements of commercial SONET, ATM, and Gigabit Ethernet transceivers.
Fortunately, a crystal has the ability to resonate at odd harmonics of its fundamental frequency. Overtone oscillators exploit this capability by using filters in the excitation circuit, eliciting resonance at odd harmonics. When properly excited at a specific overtone, crystal vibration contains negligible components at fundamental or other sub-overtone components.
Compared to multiplier types, the major benefit of overtone oscillation is its excellent jitter performance. A jitter specification below 5-ps rms can be readily achieved. Nevertheless, crystal fragility limits the maximum frequency of overtone oscillators to roughly five times the 40-MHz fundamental-mode limit, or 200 MHz. This range is sufficient for many existing applications. The combination of performance, reliability, and cost provided by overtone oscillators often makes designers choose them.
Of course, when using them, developers must be aware of some pitfalls. A crystal "prefers" to resonate at its fundamental frequency. While it can be resonated at a higher overtone, circumstances do exist that cause it to revert to the fundamental. One such instance can arise on startup, if the oscillator's dc supply voltage ramps up gradually rather than reaching the rated voltage "instantly" (Fig. 2). During such undervoltage conditions, the oscillator transistors experience reduced gain and bandwidth. Lacking bandwidth at the overtone frequency, the oscillator may excite oscillation at the fundamental frequency.
High operating temperature also may lower transistor gain and bandwidth, triggering a reversion to fundamental frequency. Quite mysteriously, though, the crystal may lack "activity" at a specific combination of rated temperature and supply voltage. It could then refuse to oscillate under those conditions.
The decrease or absence of crystal activity, along with reduced transistor bandwidth at low voltage and high temperature, all pose future challenges for oscillator designers. Innovative designs, superior transistors, and ASICs will be required for systems operating at 2.5 V.
Eventual operation at 1.8 V will present even stiffer challenges. The design of 2.5- and 1.8-V oscillators might even require that outboard components augment the capabilities of oscillator ASICs. This is especially likely for overtone oscillators built to meet the −40° to +85°C extended, industrial temperature-range specifications.
Reprints
