Has anyone used silicon oscillators? I'd like to stock a few SOT-23 oscillators to replace quartz crystal oscillators for non-critical stuff. Something cheap, of course.
--
John Larkin Highland Technology, Inc
jlarkin att highlandtechnology dott com
Perhaps not what you're looking for, but I put a Silicon Labs Si5340 into a product recently. It has a 14GHz internal VCO with four frac-N output dividers, and a frac-N feedback divider that allows it to synthesise just about any four frequencies up to 800MHz to better than 0.1ppm. If the frequencies you want are integer numbers of Hz, it will synthesise them with 0ppm error. At least it did that for all the test frequencies I tried. The frac-N dividers have dynamic phase adjusters to help reduce the jitter due to the fractional N division process. They don't say how this is done, but I assume that they get multiple phases from the 14GHz VCO and switch between them.
It's programmable over I2C or SPI. It also has a one-time programmable memory so you can get it to power up with the frequencies you want, but I'm not using that in my application.
I used a 48MHz TXCO as the input reference.
The datasheet doesn't give phase noise plots. (The part is new. I expect phase noise plots to turn up in future releases.)
I tried to measure the performance. I couldn't see any but the carrier above the noise floor of my spectrum analyser at any span / rbw setting. Ok, I need a better spectrum analyser. With the four outputs set to different frequencies, I measured crosstalk between adjacent outputs at about -80dB.
I don't have access to a phase noise test set any more so I can't do any real phase noise measurements :(
On my trusty Agilent scope (20GSa/s, not sure of jitter spec), I measured the relative jitter between the Si5340 outputs. This type of measurement shows up the jitter from the output dividers but not the VCO. At 600MHz, with all the test outputs at the same frequency I measured about 30ps p-p on infinite persistence. I repeated the same test with frequency offsets, e.g. output 1 was 600MHz, output 2 was 600.0000001MHz, etc. causing the traces to drift slowly past each other on the scope. The jitter seemed to be about 50ps p-p.
I think the performance is good enough for my needs. A decade back, that sort of performance on four channels would have cost a few hundred dollars [VCSOs, buffers, PLL per channel]. Now it's $20.
I've been using the Silicon Laboratories parts (like 501BAA16M0000CAF) but they cost about $0.70 @ 1k. Typical crystal oscillator type specs,
20ppm or 50ppm. Where can you get a SOT23 part with a decent (ie better than 0.1%) spec at anything near $0.10. I found some ST parts on Digikey at $0.26 on a
3k reel but they had no stock, 1.5% spec.
I would imagine 1% would be easy at roughly room temperature, but you could get a lot of temperature dependence for the frequency.
The built-in oscillators (which I /think/ are CMOS oscillators) in Freescale Kinetis microcontrollers are around 0.1% at room temperature, according to the samples I took. Since the cheapest of this chips is a Cortex M0+ microcontroller at about 1mm x 2mm for less than 50 cents, I would not expect a high price for a stand-alone oscillator.
Are you talking about the micromachined silicon resonator MEMS oscillators, or the tweaked ring oscillators of MCUs? Their jitter behaviour is going to be pretty different, I expect.
The "silicon oscillators" are just analog circuits, compensated RC oscillators of some sort. Power consumption and TC are low, so they are probably not ring oscillators, I'm guessing. Maybe an integrated
555 sort of thing? Silicon capacitors are excellent, and good current sources can be done on-chip.
MEMS oscillators generally cost more than XOs!
--
John Larkin Highland Technology, Inc
jlarkin att highlandtechnology dott com
In the former case, a shielding can is formed by tracks and vias in every metallization layer, then planar structures such as spiral inductors, transmission lines, etc. are fabricated inside. (The 'tub' structure may not be strictly necessary, but silicon probably makes an awful ground plane. RF chips incorporating tuners and amplifiers would still use shielding for internal isolation though.)
Such inductors have ratty Q (peaking at 5-15 or so in the 4-20GHz range), but they're good enough to make simple things with. So, add varactors and compensation circuitry, and you've got a reasonably stable oscillator (and, as it turns out, a pretty stable one in general).
The MEMS product uses electrostatic deflection on a tuning fork, or something like that, which I presume has much better Q, and probably lower operating frequency. I would think they're sensitive to rotation or acceleration, but that's probably nulled out by design (whereas, MEMS accelerometers optimize for the opposite effect..).
In either case, a DDS or whatever furnishes a programmable output frequency. Such devices are often programmed by the supplier as Value-Add parts.
Both components are comparable to crystal oscillators (which are arguably MEMS devices as well, but not monolithic), though no one really talks about the close-in versus far-out stability / jitter tradeoffs, if any.
"Jim Thompson" wrote in message news: snipped-for-privacy@4ax.com...
Really, I would've thought that too noisy; it's occasionally even suggested as a TRNG source in MCUs (compare internal RC to external XTAL, siphon off LSBs of frequency counting).
I'm surprised to hear you say that. Every time I've looked at them they were very price competitive. I think they would not sell at all if they weren't. It's not like they do much better than an XO in any way.
Everything according to the requirements. Many designs only need timing accuracy good enough to run a UART. Why worry about 100 ppm timing when you are operating a toothbrush, a microwave or many, many other devices.
Oh, I forgot about that special grade, there's A, B and "Shitty" in the oscillator families.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.