Teardown of a quartz crystal oscillator and the tiny IC inside | Hacker News Reader

Teardown of a quartz crystal oscillator and the tiny IC inside

righto.com

136 pointspicture35 comments

I just spent some time understanding the shape of the quartz crystal (don't be fooled by the broken disk in the picture -- it was originally a circular disk.)

I expected a tuning fork shape. Which indeed are used for low-frequency crystals like 32 kHz.

This one is a disk that vibrates in shear mode. The crystal is cut on an angle (about 35 degrees) to the crystal grain structure. When voltage is applied across the thickness of the disk, it creates a force parallel to the crystal axis. The crystal is extremely stiff in compression, so it can't get thinner, but it can shear so the top moves one way and the bottom moves the other.

The reason behind this complicated setup is that it's the most stable over temperature. The stiffness of any material changes with temperature, but when you get the angles exactly right the changes cancel out.

Further reading:

https://en.wikipedia.org/wiki/Crystal_oscillator#Crystal_str... https://www.jauch.com/blog/en/its-all-about-the-angle-the-at...

If you're interested in this, watch this [1] very in-depth 1943 movie called "Crystals go to War" which documents crystal oscillator manufacturing step by step. Honestly, I was amazed by this. I tend to think of technology as pretty primitive in the 40s but this is a great reminder of how sophisticated technology could be even back then.

[1] https://www.youtube.com/watch?v=fKprsCNLUlE

That's an amazing video. I was surprised by how many steps it took to make a crystal. Also noteworthy is how many different tests and checks they did on each crystal.

The technology is an interesting mixture of sophisticated and primitive. On the one hand, they had X-ray machines to determine the crystal alignment, and lots of specialized mechanical and electronic machinery. On the other hand, people were spooning abrasive out of a pot with a dinner spoon. Everything was very manual and labor-intensive, with no automation. An interesting mix of highly-skilled precision labor and low-skill tasks.

Would be very interesting to see a current-day behind the scenes video of 2021 manufacturing process and compare to 1943. I wonder how many humans touch current day crystals. In the video it’s at least at least a dozen.

I found a video of modern quartz crystal manufacturing. It doesn't have explanation, so it's basically fancy high-tech machines doing mysterious things: https://www.youtube.com/watch?v=Own-5Qz_vSs

It's interesting to compare the scale. In 1943 they were packing a box with a few dozen crystals, while the modern manufacturer fits probably 18,000 tiny surface-mount crystals in a box.

By the end of the second world war the average Lancaster bomber was flying with beam navigation, air to ground radar, air to air radar, electronic counter-measures etc.

They had pretty similar ideas to us, we can just make things smaller.

There is a famous book, to give you an idea of the sophistication, the "US Navy Handbook of Preferred Circuits":

https://usermanual.wiki/Collections/Help-Docs/books/www.rsp-...

There is a vacuum tube circuit called a "Phantastron"- a triggered voltage to time converter, you could do this with a 555 timer today... (or um, in 1972..).

https://www.radartutorial.eu/17.bauteile/bt52.en.html

Here, in a three tube oscilloscope:

http://www.hanssummers.com/tinyscope/tinyscopek3imw.html

Back in the 60s and early 70s, ham radio people used quartz crystals that were simply a sliver of crystal mounted in a little box with no additional circuitry:

https://i.ebayimg.com/images/g/VmAAAOSwdjZfEVOx/s-l640.jpg

A common circuit called a "Pierce Oscillator" (as Ken described in this article) was used to enable their oscillating ability:

https://en.wikipedia.org/wiki/Pierce_oscillator

Now all of this comes in a can which is more of a "clock module" than a quartz crystal, as Ken Sherriff explained.

Author here for any question :-)

Thanks for the article; it took me back to memories of bootleg replacing the 40MHz oscillator for a 50MHz on my work’s IIsi over my boss’ objection. Now I have some idea of what was inside.

Looking forward to any future work you might do to similarly explain the modern PLL oscillators as well as the even more modern programmable MEMS oscillators.

You mention it at the very end, but any idea why a more complex (and presumably expensive) oscillator is used? Maybe availability of the IC?

It seems strange unless there is a performance (e.g. voltage stability due to stray capacitance sensitivity) advantage. It's not like they're worried about IP or litigation.

The different oscillator types have various tradeoffs, so maybe this more-complex design was better for this application, as you suggest. Since the circuit complexity is all on the integrated circuit, the cost difference is probably very small.

My other theory when people make strange design choices is that patents might be involved.

Possibly ease of manufacturing because you had fewer unique parts to make/keep track of. Also allows market segmentation, differentiated by the multiplier and whether the disable pin was connected.

I saw an eevblog teardown of an oscillator module like this, but that was newer, not this old stock.

Question for you. This component is technically not a crystal component right? I.e. you can hook the output straight into a clock input or anything that accepts a digital signal? There are other 2-terminal components where the circuit to drive the crystal is external, right?

Are those 2-terminal components literally just electrodes attached to a crystal or do they have extra circuitry in them as well?

(Sorry not Ken but:) Two terminal crystals are pretty much just electrodes attached to a crystal. The drive circuitry for those are usually integrated in the device

Is this related to why film cameras got crystal sync motors? When crystal sync motor runs and you record sound separately in a crystal sync sound recorder, do they need to run at same frequency to be syncable?

I don't know any more about camera motors than what I just read [1], but yes, the quartz crystals keep the motors running at the exact same speed. This kept the film camera and the magnetic tape recorder synchronized.

A semi-related thing that I find interesting is the clapperboard that film makers use at the start of each scene they record. I had assumed this was a cliche tradition, but it's an effective way to synchronize the film and audio. You can see on the film the frame where the clapper closes and easily synchronize this with the sudden loud noise in the audio.

[1] https://www.filmmaking.net/filmmakers-faq/130/what-is-crysta...

This works for short recordings, but for longer recordings explicit continuous sync is needed, almost universally using Timecode, because normal quartz oscillators are not precise nor stable enough to guarantee single frame accurate sync over, say, an hour (which would be 10 ppm or so, 1/3600*30)

No question here - just wanted to thank you for your always well-written and fascinating posts. I’m only a hobbyist with electronics but I’ve learned a lot from your tear downs.

As an aside it never ceases to fascinate me that it’s literally a tiny slice of quartz being used to create these precise timing signals.

Even i wanted to thank you for your contribution. I enjoyed your Apple 5w adapter teardown which I think is still popular. Wonder if the new 20w USB-C apple charger is still interesting.

When I see an HN link from "righto.com" I reflex-click without hesitation. Always fascinating, always learn something (or a lot of things). Thank you!

In the "How CMOS logic is implemented" section, the text about the NMOS transistor says the source and drain are "P+ silicon", but the diagram labels them as "n+ diffusion". Is "P+ silicon" somehow the same thing as "n+ diffusion"?

Oops, thanks for pointing that out. It's now fixed.

Thanks! It seemed confusing, but I didn't know enough about the terminology of silicon doping to know if it was correct. I probably should by now, I've been enjoying your teardowns for quite some time!

You say this:

> Although a multiplexer can be built with standard logic gates, the implementation with transmission gates is more efficient.

But why is that? Sorry, my knowledge of electronics is still pretty limited.

Also, in footnote 2, you’re missing a period (47174 MHz) :)

If you implement a multiplexer with logic gates, you'd probably do something like `(A and SEL_A) or (B and SEL_B) or (C and SEL_C) or (D and SEL_D)` where A-D are the inputs and SEL* are the select lines.

Using De Morgan's law, the gates could all be NAND gate. You'd have four 2-input NAND gates (4 transistors each) and a 4-input NAND gate (8 transistors), for a total of 24 transistors if I counted correctly. Using four transmission gates (2 transistors each) would only be 8 transistors. (There are various other tradeoffs I won't get into.)

If you didn't need the SEL lines elsewhere, you could fold the decoding into the multiplexer, eliminating the decoding gates, but the multiplexer still comes out ahead.

(And thanks for letting me know about the typo. It's nice to know someone reads the footnotes :-)

Because transmission gate literally are two transistors. Standard two input logic gate contains four. So for example you can build two to one multiplexer from two transmission gates for total of four transistors while implementing the same function from standard totem-pole logic would require at least three gates totaling 12 transistors. (In both cases assuming you already have complementary control input, which you usually either have or can share with something. This kind of sharing optimization is the reason why transmission gate usually has complementary input and does not include the inverter internally)

Related - Tuning Fork Clock: https://hackaday.io/project/177317-tuning-fork-clock

It’s great knowing our grandparents were designing these things before we were born.

The more I learn about historical technology, the more I see that the complexity of logic built on top of old processes seems to be about equivalent to the complexity of contemporary software systems.

The only thing that seems to be really changing is the technological substrate underneath. It leads me to wonder if manufacturing ability and materials engineering are really what’s driving higher order improvements in technology, not “algorithms” so to speak. Electrical and mechanical engineers of olde otherwise seem to be no less capable of designing complex systems than software engineers.

It's a very interesting positive feedback cycle. Application identified. Some manufacturing process gets developed to automate. Scale up production. Challenges in scaling process are discovered and improvements are made with new tools made by manufacturing process. And so forth it goes. See DevOps for the software analogue.

@CyberRabbi

Given the way most modern software is made, refering to the makers as "software engineers" is a bit of an exaggeration. If you could visualize the average software project, a Rube Goldberg machine consisting of cogs, wheels, relays and bricks-and-pipes, all held together with duct tape.

Electronic schematics are similarly ducktaped copy pasta from previous designs with tons of superfluous and redundant components that nobody understands why they are there. That is the nature of engineering, nothing specific to software engineers.

A MEMS oscillator might make for a nice teardown as comparison.

Thank you for detailed post. Lot of it taken for granted when seeing schematics and it is tiny component named “XO” by designers.

Reply on news.ycombinator.com