IDT CMOS oscillators are going to replace crystals
Crystal free Oszillators on a CMOS solutions is a alternative for crystals and crystal oscillators.
Unlike in classic analogue electronics, where voltage levels as information carriers pass asynchronously through the signal chain, digital circuits require precise clocking sources, if possible coupled and phase-locked across the entire component.
Crystals and crystal-based oscillators have had triumphal success due to their precision and low temperature drift.There is hardly an electronic device now a days without at least one crystal ticking away somewhere, from digital watches, mobile phones and notebooks to supercomputers. The market is vast and therefore much coveted. The young semiconductor company Mobius Micro systems, recently taken over by IDT, has taken on the task of bringing down the “crystal fortress” by using amonolithic CMOS approach.
The Achilles’ heel of crystals
 |
The weak point of crystals, their “Achilles’heel”, is the mechanical sensitivity of these components and their cost of manufacture, which are still around a few 10 US$ cents per unit, despite automated production.
Since crystals derive their reference property, i.e. their capability to precisely oscillate at a given frequency, from their geometrical and crystalline properties, this also represents the weak point: crystals react strongly to mechanic shock and vibrations and can even develop microfissures which lead to erratic behaviour.
Due to their electro-mechanical nature, they also resist integration into semiconductors, which means that electronic circuits still require crystals, peripheral elements such as oscillators, PLLs etc., aside from the actual “effective” IC. The idea to fully integrate monolithic clock generators seems obvious, but for a number of reasons this has never been successfully realized. One of these reasons is stability, as it was hither to impossible to replace the high precision and temperature stability of crystals with pure CMOS solutions. However,demands on precision are rising continually, even for consumer products. A few examples of known PC communication standards serve to illustrate this point:
- USB 1.0, 2.0: 500 ppm
- Firewire: 100 ppm
- Ethernet: 100 ppm
- PCI Express: 300 ppmExample from thetelecommunication sector
- SONET: 20 ppm
Particularly modern communication interfaces such as Firewire or Ethernet are quite demanding with regard to the stability of the clock generator. Solutions other than crystals or crystal oscillators are either not sufficiently precise or are simply too expensive.
The monolithic method
No wizardry is required in order to integrate an oscillator into CMOS technology, since capacities and inductances as well as all types of active amplification modules circuitries can be realized. To master manufacturing methods in view of the resulting frequency and to compensate for temperature variations is more difficult. Mobius has presented a solution for this problem which it calls the capacitor bank in a suitable mannersuffice as modulation oscillator. This enables frequency modulations with adepth of a few percent and in any type of desired modulation.
![Schematic of the CHO - Cf[12:0] symbolizes the 13b condenser array for frequency alignment and for SSCG; Cv[5:0] symbolizes the Varaktor temperature compensation](uploads/pics/Schaltbild_01.jpg) |
Compensating temperature, voltage and ageing
The temperature variations of the CHO, which are negative but largely linear, are compensated by a programmable MOS Varicap bank. The active Varicap diodes are connected to a voltage source with positive temperature variations, their capacity thus decreaseswith rising temperatures, balancing the negative TC of the CHO.
Apart from the temperature, the supply voltage greatly affects the stability of the frequency generation, appropriate measures therefore had to be taken: The CHO, buffered by a differential-2-single ended converter (D2S) and the divider array are powered through an internal 2.5 Vrail, while the configurable output drivers are supplied from the external 3.3V power source. The internal 2.5 Vare generated by a highly stable LDO with bandgap-reference.
And finally, in order to prevent ageing effects, mainly caused by damage to the oxide layers and favoured by excessive signal amplitudes, the Mobius circuit features a controller in order to guaranteen arrow amplitude tolerances.
Output timing by simple division
 |
In order to achieve common output frequencies of 10, 12 or 24 MHz etc., a divider chain with integer division ratios is used. For example, 24MHz can be generated witha factor of 40, or 10 MHz with a factor of 96. For other commonly used frequencies such as 14.318 MHz, an integer divider is also used, in this case with a factor of 67 and a slight detuning of the CHO frequency to 959. 306MHz through the Varicap bank. This method does not require fractional divisions as used in PLLs, but has the advantage of avoiding the typical PLL weaknesses of additional jitter by using phase comparators and VCO.
 |
Jitter and stability
The test results in this regard are actually quite impressive - the CHO shows jitter values which are comparable with pure crystal oscillators and which area lot better than those featuring integrated PLL. However, it has to be mentioned that PLL technologies such as FemtoClock NG(IDT) are nowadays available which no longer cause significant deteriorations of the crystal properties.
However, with regard to the commercial prospects of Mobius oscillators, the only relevant issue is not to perform worse than crystals in all categories, since the main advantage results from the circumstance that oscillators can be manufactured with the tools of IC manufacturers. The first generation of Mobius oscillators achieves a frequency stability of +/- 400 ppm, which is sufficient for many applications such as USB, but not yet sufficient for communication interfaces such as Ethernet or PCI Express. The next generation will provide a stability of +/- 100 ppm and will therefore be suitable for industrial temperature ranges as well.
Tiny dimensions
The first three oscillators from the “all silicone” range, the MM8102 (300 ppm), the MM8202 (400 ppm) and the MM8203 (2000 ppm) are now available. IDT is thus the only supplier of pure silicon oscillators with crystal-like performance for the manufacture of wafers and casings.
The new components replace crystalbased oscillators in consumer applications, computing and storage which require small component dimensions, and offer excellent characteristics for creating the usual serial interfaces, including S-ATA, PCIe, USB2.0 and USB 3.0. These components are also available as wafers, enabling manufacture and packaging methods such as CoB (ChipOn Board) or MCM (Multi-Chip Module) in order to save significant amounts of space. Standard 4-pin packages measure 5 x 3.2 x 0.9 mm or 2.5 x 2 x 0.6.
These chips are manufactured using standard CMOS technology and therefore offer users a fully integrated alternative to crystal-based resonators or oscillators. Further more, the MM8202 is particularly suited for very flat consumer devices such as SIM cards with high storage capacity or USB memorysticks.
Both the MM8102 and the MM8202 offer excellent frequency precision (less than 300 ppm for the MM8102) and high peak frequencies (up to133 MHz), which is ideal for the usual wired serial data transfer technologies employed nowadays. Both ICs consume only very little power when active (2mA at1.8V) and support a standby-mode which reduces power consumption to less than 1μA. The monolithic “all silicone” swingers also offer excellent shock and vibration resistance, since they generate frequencies by electronic means and without moving mechanical parts.
Competing technologies
 |
According to a survey by the market research company “Visant Strategies”, worldwide turnover of crystals totalled 3.1 billion US$ in 2008, with 4.3 billion forecast for 2013. The potential for replacement technologies is therefore very high, but there are also difficult obstacles to overcome. The manufacture of crystals and crystal oscillators has progressed through a learning curve over many years, it is therefore mature and thus very cost-effective.
Furthermore, a competing technology has entered the market a few years ago, which also wants a part of the cake. These are MEMS-based oscillators, which are now ready for commercial exploitation and which are supplied by manufacturers such as SiTime or Discera. MEMS oscillators can be manufactured with methods similar to those employed in CMOS production, they are also very small, incredibly precise (+/- 10 ppm and better), but not very cheap.
A 12 MHz oscillator with a precision of+/-50 ppm, for example, costs approx. 1.40US$ in 1k quantities, while IDT indicates a unit price of approx. 0.55 US$ for their first oscillator MM8202 (+/- 400ppm) for 1k quantities. MEMS oscillators also have the disadvantage that they are mechanic oscillators and that they are affected by shocks and vibrations exactly in the same way as crystals. Furthermore, base frequencies are not freely configurable, the usual values are in the area of a few MHz. MEMS oscillators therefore usually require PLL circuits in order to obtain specific frequencies, plus specific excitation circuits for electrostatic excitation and, similar to CHOs, extensive compensation electronics, since their temperature dependence is not nearly as good as with crystals.
All this also affects power consumption, pure CMOS solutions are almost unbeatable in this area. For example, the MM8202 requires a maximum current of 2.5 mA (without output load) when powered at 1.8V and of 1 μA with the output switched off. MEMS oscillators, however, tuck in a lot harder - a comparable oscillator with this technology consumes approx. 3 to 5 times more power.
