Correction Request: AN2867 Oscillator design

[BarryWhit]

Rev 19 Page 16/59:

I believe this should read "increase" instead. The purpose of Rext is to limit the current and thus lower power dissipation in the crystal. So, lowering Rext increases power dissipation.

 

A separate issue is that this phrase is awkward at best. First of all, on many boards Rext=0. Since it certainly can't be any lower that that, it's a strange way of phrasing the issue to say that too low a value increases power dissipation. What the appnote is trying to say is that the value of Rext should be selected to meet two competing requirements: staying below Max Drive level (which may require increasing Rext), and having sufficient Gain margin (which via the resulting change in ESR, may require decreasing Rext).

 

CC @STOne-32 

[STTwo-32]

Hello @BarryWhit 

I've forwarded your rectification request to the concerned team so they review it and may enter the necessary updates if they have to (under internal ticket number 188914). Same for the potential typo.

Best Regards.

STTwo-32 

[STOne-32]

Hi @BarryWhit ,

Thanks for the valuable remark , when the first edition of this App note was written I believe in 2009 , it was a big revolution on the crystals world having STM32 HSE and especially LSE compatible with a small footprint 3215 (3,2 * 1,5mm) at that time it was the hype of IoT and consumer products battery powered. So we kept a big part of theory that was applicable for the famous tunning  fork robust 12,5pF CL crystal . 

Today, I supported thousands of customers and we provided xx Billions of our STM32 , it is rare to see a customer design adding a Serial Rext for LSE  Crystal , it creates only issues of leakage even kept 0Ohm .
So your remark is valid  !  
Guess why this Rext is still maintained in litterature  and design ? 

• This is the only easy way to measure the famous or so called « negative resistance » by our partners of Crystal manufacturers and customers and define  another ratio so called Safety factor  that was widely used in automotive industry to determine a stable oscillation over time . But not the same as oscillation start up we define with Gm .

crystal manufacturers and their market improved a lot today and we have a wide variety of crystals : Low power , trimmed and suitable with different sizes driven by Low power , Embedded and and Connectivity . I believe that that Rext should not exist anymore on a new design if the right crystal is selected with our LSE ( 4 drives are provided ) and also our HSE ( General Purpose or RF).

As said by @STTwo-32  we will try improve that in the AN.

Have a great end of Sunday.

STOne-32

[BarryWhit]

Is it also practically irrelevant nowadays to use Rext to resolve overdrive of HSE crystals? I just posted about this as a possible answer to a malfunctioning circuit, but by what you've now said perhaps I was way off.

 

Whether you remove the footnote or correct it doesn't matter to me, I would welcome either.

Thank you both for the very prompt response.

[BarryWhit]

Thanks for the kudos. I would have appreciated an answer even more. I'm honestly not sure what the answer is.

 

---

@BarryWhit wrote:

Is it also practically irrelevant nowadays to use Rext to resolve overdrive of HSE crystals? I just posted about this as a possible answer to a malfunctioning circuit, but by what you've now said perhaps I was way off. 

 

 

[STOne-32]

Dear @BarryWhit ,

we already escalated the improvement update for the AN2867, with this #188914 internal ticketing system.

Now, What I’m saying is to select from the start the right crystal for HSE or LSE to avoid to use Rext from design phase as the choice is  now very wide . Then if the choice was already done as example a preferred part and not possible to change this is required to avoid for MHz crystal overdriving and may affect the oscillation reliability over time .

Ciao,

STOne-32

[BarryWhit]

Right. But, that does mean that Rext is still occasionally useful, like when someone realizes too late that they overlooked a specification, which happens quite a lot. But this also means that

 

> why this Rext is still maintained in literature  and design ? 

> [Because] This is the only easy way to measure the famous or so called « negative resistance » 

 

is not quite true. You can avoid the need for Rext by selecting a suitable crystal, but you may still need it to fine tune your design if your crystal isn't spot on. Furthermore, in the crystal circuit more than in other parts of board design, it seems like there is more manual fine-tuning required to get the design right.

 

Finally, even if you're not technically exceeding the max overdrive of your crystal, lowering the drive level overall can make the oscillation cleaner (reduce distortion, make it more sine-like). I don't what effects if any does this "mild overdrive" have on a circuit in the long run, but I think practically it's just not realistic to be able to guess what the waveform look like from just the datasheet. it's more like you make the best first guess you can, and than you (hopefully) tweak in the lab to dial it in. So I think Rext is  something people definitely still need to know about, and the appnote should keep the discussion around this.

 

Thanks,

[PHolt.1]

A suggestion:

There are basically two xtal vendor groups:

AVX, Kyocera, and such. These are expensive (check out e.g. mouser.com for a 25MHz one and a 32768Hz one) although presumably 1M+ volume users pay only pennies/cents for these €0.30 (1k+) xtals.

Various chinese xtals. These are 1/3 of the price of the above and I bet most sub-100k volume mfgs use these unless it is a pricey product in which case they don't care.

It would be great to have some of the latter listed in the xtal appnote.