TI Claims Breakthrough in BAW On-Die-Oscillators | Hacker News Reader

TI Claims Breakthrough in BAW On-Die-Oscillators

eetimes.com

101 pointsbaybal248 comments

This is a pretty big change for digital design. With high accuracy 48MHz clocks on chip, parts that do USB become that much cheaper and more reliable. Single chip USB-3.1 chips become possible as well.

The low phase noise is something that makes them very useful for wireless systems that are all software radios these days anyway. Phase noise, or clock jitter, results in difficulty in receiving closely spaced signals. And since many (if not most) of the modern wireless protocols today are multi-channel, it means you can make wireless protocols that support more clients in the same spectrum, or more data to clients in the same spectrum.

I am hoping that as a minimum TI releases a series of 4 pin oscillators to replace the ones like from Vectron that can be very finicky to keep working precisely.

The only thing I didn't see was what is their vibration resistance standard? The small 'can' oscillators can be pushed of their center frequency by yelling at them ;-) It is a really hard problem to both secure a quartz crystal from outside vibration influences and still allow it to vibrate at its intended frequency!

I think it's so interesting that the best oscillators we can use in electronic circuits, atomic clocks excluded, all operate in the mechanical domain. If I knew nothing, I would assume we could build good RC oscillators, which are purely electrical.

The big advantage of mechanical oscillators vs RC occilators is mechanical ones have very high Q. Q is related to the ratio of resonant energy vs the amount lost per cycle. High Q oscillators allow you to produce spectacularly pure and or accurate signals. That's important for RF applications.

On the other hand RC oscillators can start up very fast. In uS vs milliseconds. Lot of microcontrollers have an internal RC oscillators for that reason. Go to sleep and wake up instantly.

TI's announcement is interesting since external crystals for RF transceivers is a pain point in design and manufacturing. You can have inexplicable variation from one batch of crystals to another. Batch's of IC's. Variations in the PCB material, etc.

What does "purely electrical" mean at this scale? Quartz resonators are piezoelectric devices, after all. The same field of solid state physics that explains a MOSFET is required to explain a oscillator.

Quartz resonators still move and are affected by the direction of gravity

Yes! Here's a video: https://www.youtube.com/watch?v=zILwgQhjC_Q

Piezoelectric respnators are using mechanical resonance.

You can build LC resonators. And also resonant cavity resonators. LC are electrostatic/magnetic. Cavity resonators work off standing waves.

So do electrons and nuclei.

It means that it doesn't literally physically vibrate, like quartz oscillators do. It's only the electric field that vibrates in it, not the physical thing itself.

I know what it means. I was pointing out (charitably) that that's a silly distinction. An electronic oscillator ABSOLUTELY vibrates "physically". Those electrons are "physical things" and have mass. They have momentum. Those things can be measured and have an impact on the circuit.

Even your assumption that the transistor body doesn't move is wrong in practice. Those electric fields absolutely produce strains (that is, real physical forces) on the semiconductor junction, and those strains affect electronic properties and are included in fancy circuit models (or at least I saw a paper about it once, anyway).

The idea of "solid state" machines having "no moving parts" is a lie told to you by your electronics textbooks. It's a useful abstraction. It's not the truth.

And in this regime, it's frankly harmful to your understanding of how things work because it's led you to believe that crystals are doing something fundamentally different than circuits.

Maybe if you can show me that if you flick an RC oscillator you can detect it, then I'll believe it's a silly distinction. Or if RC oscillators can detect gravity: https://www.youtube.com/watch?v=zILwgQhjC_Q

I'm also curious what it would mean to you for two things to be doing "fundamentally different" things. Do you believe in the difference between a mechanical oscillator and an electrical oscillator? Mechanical oscillators like a spring/mass exchange energy between kinetic and electrical energy (a moving mass and a strained material). An RC oscillator exchanges energy between elements that store energy in an electric field and a magnetic field. That feels fundamentally different even in a four-fundamental-forces kind of manner.

Quartz crystals oscillate like a slab of jello vibrates. It’s not moving like a pendulum or anything...rather the bulk material is expanding and contracting and energy is stored in the deformation of the crystal lattice.

Same with a spring-mass oscillator, right? A spring stores energy in the deformation of materials.

Yes. A spring reacts to force by storing energy in the strain (aka stretching) it experiences.

A crystal oscillator has the additional property that when it stretches, it develops an electric field across it--and conversely, when an electric field is applied across it, it stretches.

So you can make the crystal vibrate by putting a voltage across it, and this vibration is visible when you monitor the voltage. It's like striking a bell...the bell's vibration creates an audio vibration.

To be honest, I'm not really sure the reason you're explaining this. I had thought you were drawing a contrast between pendulums and quartz oscillators.

What I find even more mind-blowing is that GHz range frequency filters in cellphones are mechanical. They have excellent selectivity and need very little energy. Check it out: http://www.telecomabc.com/s/saw-filter.html

I suppose once you're in the domain of crystal oscillators, the line between "mechanical" versus "purely electrical" is a bit fuzzy. The mechanical/thermal effects of passing a high-frequency signal through passive components (or a transmission line) might be comparable to the mechanical vibrations of a quartz clock, though somebody better at physics could probably correct me.

At a "couple of hundred microamps" penalty, this more or less kills a wide range of potential products, ie battery-operated products like smart sensors or remote controls and the like.

One of the most important factors for a long lifetime is to minimize the static power draw, and running at 0.1--10 uA is quite common.

A common CR2032 has about 225mAh. If you can get 80% out of it, you have about 225 * 1000 * 0,80 / 24 = 7500,0 uA-days. At 10uA, you have roughly 2 years (depending on usage etc). At "a couple hundred uA", you'll have two months.

I, too, chuckled at the "couple of hundred microamps" quote. However he refers to the high-speed clock, which is turned on only during transmit/receive, and is indeed negligible compared to the total power consumption during RF communication.

That being said, their low-frequency on-chip oscillator specs are horrendous (50ppm PER DEGREE). You can't maintain a BTLE connection or mesh node with that kind of clock, you still need a 32kHz crystal oscillator, so TI haven't "solved" the crystal oscillator problem just yet...

Indeed; they clearly haven't solved accuracy. But they've done a remarkable job of cleanliness. In particular, the close-in phase noise (e.g., at offsets from carrier < 1 KHz) these BAW-endowed parts are amazing.

With the CC26xx parts you can use a 32kHz RCOSC and get accurate enough timing for BTLE connections, so with the CC2652RB you should be able to have true "crystal-less" Bluetooth LE.

Not necessarily. The application in question is for the baseband reference frequency in a radio. If the oscillator can be powered down with the radio, and powered on during transmit, it would be a great savings.

A good callout re: low power applications, however.

Its speaking of it as 2% as well. The 2% right before it means the 200uA is relative to starting at 10mA. Your CR2032 is already not going to make it a day. That you lost about half an hour off of 22 hours doesn't mean much. I suspect this application is not useful to ultra low standby systems, but rather around RF applications of some sort that are already much hungrier for power than 10uA.

Oof, didn't notice that. The 2% penalty seems to be relative to the entire chip's power consumption with the radio on and heavy CPU activity. If you compare with the 680 uA specified typical power consumption with the radio off and CPU idle but no other power-saving features activated, that extra couple of hundred microamps seems like a much bigger deal.

How does it fare in a slightly-helium-rich environment? That story about every iPhone in a hospital shutting down for a week after an MRI shutdown turned out to be due to not-quite-hermetically sealed packaging of the silicon mems resonators.

Based on the video here [1] and some of the links on that page, plus assorted diagrams of other bulk acoustic wave resonators, I don't think there will be any problem.

Helium can interfere with devices that either depend having a vacuum in a volume or depend on having that volume contain a gas mixture with specific properties. Helium getting in messes that up.

The BAW on-die-oscillators do not appear to have any places that are supposed to be vacuum or supposed to contain gas. The have two piezoelectric thin films, with acoustic reflectors behind them. The gap between them appears to be solidly filled. All the relevant acoustic waves are carried in that solid stuff, not in air. (This also means, I think, that these things should be OK in vacuum).

[1] https://training.ti.com/tis-bulk-acoustic-wave-clocking-tech...

The helium problem was because of helium working its way into the crystal structure of the resonator and affecting its springiness.

From this old paper, I think quartz crystal oscillators are still vulnerable. It all depends on the packaging.

https://tf.nist.gov/general/pdf/906.pdf

Ceramic piezo oscillators and silicon mems oscillators are a very different technology based on different physics.

What matters here is the speed of sound in that ceramic piezo material. Can helium change that by much? I don't know.

...Yes? That's why I asked.

Technically, you are not answering GP's question.

This is so funny to me. I instantly thought of the CC26xx series (FSK transceivers) from TI when I first read the title, "Surely the next version of their chips will use this amazing new tech and destroy the need for that external epson oscillator!".

Sure enough, that's the first chip quoted in the article. That explains how the newest CC2652 has been available in sampling quantities only for what, a year now?

GG, TI.

Weird, the LMK05318 datasheet shows an external crystal oscillator on pins 31 and 32. Is this an older part and the integrated oscillator is not yet released?

I went looking for phase noise specifications, as the article and CC2652RB data doesn't have much to say about how the phase noise of the integrated oscillator compares to an external oscillator.

http://www.ti.com/lit/ds/symlink/lmk05318.pdf

http://www.ti.com/product/LMK05318

In that particular part the BAW is part of a secondary, faster oscillator. It's leveraged for its spectral purity - its phase noise - rather than its accuracy. The datasheet you linked is for a clock chip that's designed to take in a reference clock and transform it into multiple end-use clocks.

A really neat thing you can do with that part is use a GPS receiver's pulse-per-second output as a high-accuracy, high-noise reference to generate high-accuracy, low-noise, high-frequency clocks. It's something you would see in datacenters or cell sites or distributed sensing systems.

Will this technology slowly replace external crystals for generic microcontrollers? Or just for high-accuracy applications.

If it is as good as the article claims, I would say more like “instantly”.

Bringing the resonator on die eliminates the crystal from the BOM, one of the pricier passive components. Also probably two capacitors are now gone, again for xtal circuits you need better caps, cheapies are not good for oscillator circuits. Plus the PCB area for the oscillator is eliminated, and that is an area of PCB that requires careful layout to meet the constraints on paracitic capacitance, etc.

Process people will take their sweet time to turn this into a marketable fab product.

I'd say that it can come to market quicker as a packaging house product: they grow resonator dies with BAW separately, dice them, and put into existing RF SoC packages.

The increase in power draw is significant for a lot of applications which will make these unfeasible. That being said there are a lot more applications that could fit that within their power budget.

I suspect in the cases where the power isn't an issue it will come down to the designers choosing between the cost and space trade off where a crystal is necessary (it isn't always).

Of course currently only TI can make these (though no doubt everyone else is going to make a real effort to replicate). The chip in question is rather pricey though, $11.44 in thousand quantities vs. $3.54 for the very similar NRF52840 + about $1 for the crystal and other components needed.

You're looking at the price of their 400Gbps network synchronizer chip. The other chip they mentioned with this technology, the wireless microcontroller, is only $3.55.

Or TI will license this to others.

For many low-accuracy applications, just an RC oscillator is used. This sounds like a good way to reduce part count for systems that need the accuracy of a crystal - if the main IC is made by TI.

Or somebody that licenses the patent(s) from TI.

I assume this is the technology Apple is already using in their custom silicon, and why there was that fairly recent story of a hospital where Apple devices failed in the presence of helium leaked from their MRI installation.

Apple is/was using silicon mems resonators, which is basically a fancy tuning fork on flexures cut out of silicon. This technology ("bulk acoustic wave") sounds like surface acoustic wave, but perhaps with the vibration mode being through the bulk material rather than just the surface.

It also seems to fill a different niche from the MEMS resonators Apple are using. Those are ultra-low-power 32KHz oscillators for keeping time in standby mode, but the CC2652RB still relies on an external quartz crystal for that, perhaps because this tech is even more power-hungry than the already too power-hungry high frequency quartz oscillator it replaces.

Sounds like it is slightly higher power. From the article:

> Asked if there is any tradeoff by integrating a BAW resonator in the wireless MCU package, Wong noted a potential power delta at about 2% — “a couple of hundred microamps.” He called it “a reasonable tradeoff, [considering] its benefits outweigh [it].”

But, I guess if you can invest the money and area you saved by not having an external crystal and buy a 2% bigger battery, which is probably a net win.

The thing preventing this to be a home run is the power usage increase. +2% vs

Reply on news.ycombinator.com