PORTLAND, Ore.— Integrated Device Technology Inc. has introduced what it claims is the world's most accurate all-silicon CMOS oscillator with the industry's highest frequency accuracy, measured in parts per million (ppm).
Synchronizing high-speed digital circuitry needs rock-solid time bases, which usually means quartz-crystal based oscillators. CMOS oscillator makers, however, claim to be pioneering a new breed of digital time bases that are faster than quartz crystals yet smaller and lower power than MEMS.
"We are launching a 100 ppm all-CMOS oscillator that makes us competitive with crystal oscillators," said Michael McCorquodale, founder of Mobius Microsystems which invented the all-CMOS oscillator technology acquired by IDT (San Jose, Calif.) earlier this year. At IDT, McCorquodale is general manager of the Silicon Frequency Control (SFC) business. "We have already shipped 3.2 million units just in the last quarter."
IDT is promising 50 ppm parts by 2011, but has been quietly seeding OEMs with advance models of its current 100 ppm IDT3C02 oscillator, which is already replacing quartz crystals in a wide variety of timing applications. IDT claims design wins for Gbit Ethernet, Display Port clock, subscriber identity module (SIM) card, encryption token key, smart-card, microcontroller reference, peripheral component interconnect express (PCIe), serial advanced technology attachment (SATA), solid-state drive (SSD), universal serial bus (USB 3.0), flash drives and card readers.
Early reports from OEMS, according to IDT, claim that their digital circuits are achieving superior bit-error rates with its IDT3C02 despite its lower cost over quartz crystals and tiny 5-by-3.2-by-0.9 millimeter size (with an even smaller footprint--2.5-by-2-by-0.9 millimeter—planned for its 50 ppm part due out in 2011).
IDT3C02 oscillator (on top) can be wirebonded to the OEMs application specific integrated circuit (ASIC, on bottom)
"We have put the right compensation circuitry on our all-silicon CMOS oscillator to generate very high frequencies with good accuracy," said Tunc Cenger, senior manager of product marketing at IDT. "And our added value is that you can integrate inside the package which you can't do with quartz."
IDT delivers its all-silicon CMOS oscillators on wafers before dicing, so that its customers can stack the IDT3C02 on top of their application specific integrated circuit (ASIC) in a multi-chip package, or for ultra-price-sensitive applications, like flash drives, OEMs can affix the IDT die to a chip-on-board (CoB).
IDT has 35 issued and pending patents on its unique compensation circuitry and die encapsulation that hermetically seals its CMOS oscillator to protect it from stray electrical fields and changes in the environment, such as humidity, that had prevented previous designs from achieving 100 ppm frequency accuracy and less than 457 femtosecond phase jitter.
The frequency-trimmed, temperature-compensated, environmentally-stabilized IDT3C02 achieves -140dBc/Hz phase noise by beginning with a 3GHz self-referenced LC oscillator that it divides down to user-programmable range of from 4-to-133MHz. By using no power-consuming phases locked loop (PLL), the the IDT3C02 consumes less than a quarter of the power required by quartz, MEMS and other all-CMOS oscillators--just two milliamps active and 200 nanoamps in stand-by with 100 microsecond start-up--plus can run at any supply voltage from 1.8-to-3.3 volts. The no-moving-parts design, compared to quartz or MEMS, also leads IDT to claim superior shock and vibration resistance for its all-silicon CMOS approach.
IDT mixed signal die for its all-silicon CMOS oscillator houses a large inductor (top) surrounded by its compensation circuitry.
Very nice solution. I guess with the noise problems, it will never be integrated onto the ASIC. Can it be used in a separate package in the system or does it have to be packaged with the ASIC to avoid stray capacitance and ground loop issues?
noise problem? The phase noise seems a little lower than in other types of CMOS oscillators and should be much lower at the frequencies at which it is user configurable.
I'm also pretty sure it will have an option of being separately packaged, since wire bonding is also used for linking silicon to the package. That's my 2 cents anyway.
Lots of people have integrated high frequency LC oscillators on CMOS SoCs, so in reality the noise problems are not such an insurmountable barrier to integration.
But IDT is not in the SoC business and is not offering this as an IP block for others to integrate -- hence, the discussion in the article about IDT selling bare die so you can do stacked die in one package.
Either way, you achieve ths same result -- eliminating the quartz crystal -- which is the whole point. IDT has taken this nicely compensated LC oscillator plus dividers and built it as a stand-alone chip that targets quartz oscillator replacement for clock frequencies up to 133 MHz -- a clever marketing approach that could be very successful for them.
I do think, however, that the article title is somewhat misleading. "beat quartz" are not the words I would use to describe an oscillator that is "almost as accurate as quartz."
Forever is a long time Timing Guy. Not too many years ago it was a given that a good radio receiver required a SAW filter. But since the SAW couldn't be integrated in silicon, clever engineers found a way to design it out in many applications. Never say never...
This is a rather odd comment from Timing Guy. First, I've seen one of IDT's talks and they use a massively parallel test to trim every single device so yield should be in the high nineties in terms of percentage. They certainly don't "throw" anything away other than typical semiconductor yield loss due to defect density which should be very small with such a small die. Second, and assuming I am understanding correctly, by offering the die product, they are selling directly against passive resonators. That ASP is probably around $0.10 or less. Interestingly, that suggests that the CMOS oscillator is much lower cost than commoditized quartz. It seems like a real threat to quartz to finally make the function in standard silicon.
Thanks Kris. I will visit that site. This is a veyr interesting development indeed. It will be a relief to many board designers, no extra routing if its going to sit on ASIC. But I would love to know if the performance remains same as in conventional designs.
Wow, this is really cool. So we will not see those bulky oscillators when we open our digital watches :). Overall the spec is pretty gud and is comparable to the conventional oscillators available in the market.
Having 100ppm accuracy and less than 457 femtosecond phase jitter along with greater immunity to stray electrical fields and protection against changes in the environment, such as humidity, the IDT3C02 looks to be the best CMOS based oscillator available, isn't it?
Can anyone explain how this technology works? Any grad student can build LC oscillator that oscillates at hundres of MHz or few GHz (I know first hand, I had a few students who built that)...by neither L or C are predictable so you will get a value of oscillation frequency which will be 20% off or so...you can tune that in with C (using varactor) or less likely with L (using some switches to connect more L if needed)...but how do you maintain that tuning over life of the product, temp changes, VVD variations etc...any hint is appreciated...Kris
I see IDT claim 73c/10K, which is ok, but not much below even Digikey Crystal Osc prices.
There are also 1mA Crystal Oscillators, so their claims of lower power is more 'average', rather than comparing leading edge, with leading edge.
I can see it has a place, replacing the very bottom spec'd crystals, but Ceramic resonators are more threatened here.
The wide supply range is impressive.
Missing from their data sheet, is the Base Freq, I see mentioned above as 3GHz.
Also missing is any breakout of the 100ppm components : no dF/dT, dF/dV, etc
I agree with you. The title could be over-optimistic for the IDT3C02 CMOS oscillator. But this "all-silicon" oscillator could certainly beat the 100 ppm quartz crystal oscillators available in the market...isn't it?
The length change of silicon is 2ppm/Kelvin
Length change of Aluminium is 23ppm/K
Copper has 16ppm/K
Lets assume that the silicon length change overrides the length change of the metal (whatever is used) , a temperature change of 50 Kelvin causes an inductance change of 100ppm
Same for the capacitors, I guess.
I have no idea how they trim the circuit to compensate the +/-20% process spread of the capacitors. Laser trimming? I see no space for a fractional PLL.
Yes, I've been following these guys since they published their work at the ISSCC 2007:
when they were still Mobius Microsystems.
If I remember correct, the paper mentions using an elaborate digital calibration with the
aid of an external reference clock to achieve sub-100ppm setting accuracy.
Aging? What about the Aging? Will they spec 1 year and add that in the 100 PPM spec? What does it do 3 to 5 years down the road.
Short Term Stability?
457 femtosecond phase jitter. Yes but defined over what range? The range where the LC oscillator is optimum?
Achieves -140dBc/Hz phase noise; yes but at what offset frequency?
Yes but I have 1 more question to ask; if this is such a great technology then what happened to Mobius Microsystems? Why was it acquired by IDT earlier this year.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.