32KHz oscillator issue with MSP430f149

Guys,

This has become critical for a customer who is now it a production stop because of this.

Whilst they are a small company the unit using the MSP430 is expensive ($4000) and they are currently at production stop due to a 20c crystal.

Please, can someone have a look at this for me?

Thanks,

Steve

It could be the 20c crystal that makes the difference.

Agreed, or it could be the layout, or it could be the loading cap requirement, or it could be something which has changed in the MSP, or it could be...

If we eliminate possible causes, whatever is left is the culprit.

I am trying to eliminate the crystal itself.

BR,

Steve

On the 1611, which has basically the same crystal circuit, we experienced problems with an 8MHz crystal recently. We didn't change the layout and all was well for quite some time, but suddenly we had stopped crystals.
We suspected the crystal, even if we ordered from the same source and the distributor offered them as the same brand/model as before. Or the capacitors (but a replacement with definitely correct ones didn't help in most cases)
The solution was a change in the nominal load capacitance.
However, reading this not for the first time now, I start to suspect a silicon change to be the culprit. No idea what or why. Or any other change in the production process of the MSP (e.g. case plastic material)

Well, changing the capacitance solved it, so we didn't invest more time into the issue.

Thanks for the response.

We also had one board recently stop oscillating on the 4MHz crystal too which was previously working (was going through final test and stopped half way through the testing).

I will experiment with various capacitors to see if I can find some 'safe' values, but the customer is a little concerned about this.

The 32K is not supposed to need any capacitors but we tried both loading caps and also a high value resistor to ground, to supply and across the crystal in various combinations and nothing seemed to make it stable.

Again, thanks for your reponse.

BR,

Steve

OK, more testing results...

We lifted the pins of the oscillator and connected the watch crystal directly to the pins so there are no additional board loadings or capacitances and the oscillator STILL did not oscillate.

Everything is pointing to the crystal, but as far as I can tell it should be OK. The ESR is 35K which is less than the "required" 50K.

They also seeing some instability on the 4MHz oscillator but this needs more debug.

We are now at the point where we will have to design out the MSP on the next revision of the board :(

Any and all help or comment on this issue would be very much appreciated. I would hate for them to switch processors.

BR,

Steve

Steve Clynes said:

... We are now at the point where we will have to design out the MSP on the next revision of the board :(

Any and all help or comment on this issue would be very much appreciated. I would hate for them to switch processors...

Do you mean design out the entire MSP family? I thought MSP430F24x is much improved over MSP430F14x and the migration path is not that bad.

It really comes down to a confidence issue. The customer has never had any issues with the processor used in their other products (Silicon Labs processors).

Switching processors would require them to then keep two software forks since the firmware is field upgradable.

BR,

Steve

And always read the ‘Device Erratasheet’!

-Leo

I did, and there was nothing relating to the LF crystal, so I was/am running out of ideas.

BR,

Steve

I was looking to BCL5, but you are probably not using the DCO.

But anyhow my experience is to look carefully to other bug reports even if they seems not relevant to the experienced problem.

-Leo

Yep, I understand your point :)

I did see an errata relating to reset which I had high hopes for but it didn't have any effect :(

BR,

Steve

FYI: Handling this one offline - Steve or I will post back with resolution info once we track down the problem (board, crystal, system, etc).

Regards,

Katie

Shouldn't this have been on the internal forum? Steve are you aware of the internal forum?

Steve Clynes said:

I was/am running out of ideas.

Personally, I never used a watch crystal on an MSP, but I used lots of them for realtime clocks over the last 25 years (based on PCF8583 I2C RTC chip).

I never experienced a problem. neither in the 5 or 6 different projects/layouts nor during mass production. I cannot remember a single unit that failed. And I really didn't care much for the layout. The first one, an RTC for the C64 user port and later its tape port, used a very simple layout, the prototypes were wired with tin-solder on a grid-style drilled board.

Taking my C64 cpu clock as reference, I was able to trim them to <10ppm with an adjustable capacitor and they worked for years and years (the few units I got back for service were because of the soldered batteries)
So unless quality of crystals has dropped on the last 25 years, I don't see the crystals as the offending parts. (even the latest designs with an RTC worked from scratch)

You said that the latest batch won't work. What has changed?

Any layout changes? Changes in the PCB production? (crosstalk etc.) Is the MSP revision the same? Same brand and model of the crystal?

If you still have a board from the previous batch, swap the crystals. If the old board still works and the new doesn't, the problem is somewhere else.

In our production, about 10% of all devices fail to bring the 8MHz crystal online. Here, the layout didn't change, nor did the crystal (at least according to the distributor: same brand and model). Usually a change of the crystal or a tweak on the capacitors helps. A smaller percentage fails almost immediately after installign the devices. Maybe due to changes of humidity or temperature that crosses a critical border. We haven't checked the MSP revision
(I got this idea while reading this thread, so thanks for not posting it internal only)
We too do not know how to fix this permanently. Luckily it doesn't stop our production, it is just a (greater) nuisance. But knowing the reason would be nice. It's just that there are more important things to do than digging deep into it.

At the moment nothing seems to make sense.

The boards are all from the same batch and the components are the same and from the same batches. No layout changes etc...

I took a different crystal from the EVM I had... didn't work. I put a 2491 in the EVM with both old and new and it didn't work. Different device revisions... same issue.

Temperature and humidity have certainly changed, but this just makes me even more worried about the units the customer has already shipped!!!

For the moment they have changed their software to use the 4MHz for their timing routines but they would still like to get to the bottom of this.

We have a few more things to try with super cleaning the boards being next on the list. After that comes giving Katie a board :) :) :)

I too have worked with crystals for quite some time and never paid too much attention to layout etc... and even the EVM is not as clean as their board I don't think, but I could be missing something critical.

I really appreciate your feedback and comments. I will keep you updated with our progress in the hope that anything we find out might help you too :)

BR,

Steve

Following your discussions it looks to me that all has no sense.

I would take a very close look to the PCB. Is there any short, even high resistance. Caused by copper, tin or solder past fading/shadow or solder mask and silkscreen lacquer.

But don’t focus too much on you 32Khz problem. A neighbor pin or even more far away could cause the problem.

-Leo.

Leo,

I agree :)

I have even removed a device, cleaned it up then only re-soldered the power and crystal pins, and on another board lifted the crystal pins completely and soldered the crystal directly to the XIN/XOUT and it didn't work either.

My next thought was power supply noise, so I injected a lab supply, still no go :(

It is certainly a strange issue.

I am really hoping for that "dohh" moment bu no joy yet.

BR,

Steve

Steve,

Just giving some hints maybe you already tried it;

Trying blank software which only initializes the 32Khz clock & internal DCO, leaving other modules unused and all port pins floating.

Hardware Reset the device after power-up.

Measure the current to the board and, if available, compare it to a good working board.

-Leo.

This appnote may help...

http://www.ti.com/lit/an/slaa322b/slaa322b.pdf

I did try the 32K examples from the product folder and they don't work either :( although this was through the IDE. I will try without the IDE so that it can free wheel.

I will also try measuring the current to the MSP.

Thanks for your thoughts and input. I appreciate your time.

BR,

Steve

Mike,

This is actually where I found the information recommending no load caps for the 149, and also to try a load resistor for low voltages (which I tried just for the fun of it).

Adding a load resistor worked on one board, but it is possible that it was actually something else that helped since it didn't work on any other boards :(

Thanks for continuing to think of things to try though. It is appreciated.

BR,

Steve

You might try this one as well...

http://cache.freescale.com/files/microcontrollers/doc/app_note/AN3208.pdf?fsrch=1&WT_TYPE=Application%20Notes&WT_VENDOR=FREESCALE&WT_FILE_FORMAT=pdf&WT_ASSET=Documentation&sr=8

I haven't thoroughly looked through this, but it sounded like it may help.

Excellent. Thanks.

We will work through this.

BR,

Steve

Another one with crystal problems:

In this thread, Joe reports that he has problems with the watch crystal accuracy.
During this thread, it seems that the production date of his MSP chips is related to this. Old chips (produced in 2005) did have +-10ppm but thew chips (produced in 2010/2011) have almost +-50ppm, on the same board.

Has TI changed something in their production process that might affect the whole span of MSPs? The package plastic, the foundry, the packager? Something that might affect pin capacitance or pin-to-pin isolation resistance? So that previously working (but maybe fragile) layouts no longer work or show decreased performance?

It's just another wild guess, but since we're out of ideas...

Hi Steve,

As a sanity check, could you try one of these parts on a TI target board like this? It depends on if you are using the right package of course (64-pin LQFP (PM)). Give me a call if you don't have access to one and we can figure out how to make this happen.

Regards,

Katie

Katie,

I do have one of these (very old though) and I am sure I tried without success. I also tried a 2491 and it didn't seem to work either.

I will re-try the experiments again though as a sanity check.

BR,

Steve