• State
  • Locked Locked
  • Replies 8 replies
  • Subscribers 52 subscribers
  • Views 1233 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

cc2640 32k oscillator stability over temperature

Kevin Braswell
Intellectual 295 points

Other Parts Discussed in Thread: CC2640, CC2650

I am observing a temperature-induced failure in the CC2640 32k oscillator, where the oscillator will stop while the CC2640 is being cooled (or while warming back up to room temperature after being cooled).  And it will not restart except after a power cycle.  At room temperature there are no indications of oscillator instability, so I need a way to verify that my test condition is valid and if so, how to resolve this issue.

My 32k crystal is CM13032768DZF (12.5pF load capacitance).  I'm using 22pF caps.  On the small quantity of boards that I have tested (at room temperature) the 32k oscillator starts up in 1~2 seconds, the 32k frequency is accurate within 20ppm, and the oscillation signal is about 150mVpp.  The oscillation signal level seems like it might be a little low, but I have assumed it is okay since all other performance seems as expected.  Is there any indication that this signal level is "too low"?

As for my cooling method, I'm lining up a temperature chamber for further testing, but my initial results are based on using cold spray.  I am monitoring the CC2640 case temperature with a thermocouple, and applying the cold spray from the opposite side of the PCB to minimize direct thermal shock to the part.  But it's clear that the temperature gradient is not well controlled and can be extreme.  The datasheet indicates that the temperature gradient is limited to 5C/s, but apparently this only applies to temperature rise (i.e. positive temperature gradient), and only when the CC2640 is in Standby mode.  I have wondered if I may have exceeded this positive temperature gradient, but I have also seen the oscillator stop during a negative temperature gradient.

Any suggestions of things to check or additional guidance would be great.  Here's some more picky background details.

Since size was not a constraint, I used the (larger and cheaper) CM13032768 instead of one of the "approved" 32k crystals (as per Rev C of SWRA495, which has now been deprecated).  I checked that it met the requirements of the 32k oscillator, at least to the degree that it made sense.  The datasheet indicates that the 32k crystal load capacitance should not exceed 12pF, but one of the approved 32k crystals (FC-12M 32768KD-A3) in the previous rev of SWRA495 has a load capacitance of 12.5pF.  And that rev of SWRA495 also pointed out that while the "preferred" crystal load capacitance is 7pF, increasing it to 12.5pF would just mean a current increase of about 0.2uA.  So that all made sense and there did not seem to be any issue.

This (or a similar issue) was initially reported by my customer, who currently has about 200 of these boards out in the field (in California, exposed to weather).  Considering the current time of year (early Spring), it's hard to imagine how what they are seeing is related to what I am seeing.  But the symptoms appear to be the same.

  • 0
    TI__Mastermind 18780 points
    Hi,
    This sounds strange, so I think we need to do some more debugging. How do you know that the 32 kHz crystal stops? Can you monitor the 32 KHz signal on a DIO while doing your test? Can you try turning off the DCDC in SW while running your test and also monitor the VDDR signal? Do you have one of our kits that you can try to reproduce the test on? We have used cold spray here on our kits before without seeing the issue that you experience. What power supply do you have?
  • 0 in reply to CHS
    Intellectual 295 points

    Hey Charlotte,

    Thanks for the reply.  First off, may I reiterate the only definitive question I have come up with so far?  I'm sure it's easily lost in my many words.

    At room temperature, I measure the 32k oscillator signal level to be about 150mVpp.  Does that seem reasonable?

    As for your questions, I can answer a few things right off.

    I am monitoring the oscillator by probing the X32K_Q2 net (at the crystal load cap).  I'm using a Tektronix TAP1500 active probe (1pF max) with a minimal inductance probing technique.  It seems to have little effect on the oscillator.  This allows me to not only know whether or not the oscillator is running, but how significantly the AC signal level is affected by temperature.  Of course, I can't put this whole test setup in a temperature chamber.

    As for my power supply...  I'm supplying external power (3.2Vdc) from a bench power supply.  In the hardware power architecture I have a regulator that provides 2.5Vdc to VDDS.

    As for investigating software changes, that is a challenge.  I designed the hardware for a customer, and they fully control the software.  They reported a similar issue on units in the field, which is why I'm investigating this.  But they have not yet committed to providing a custom software build to assist in this investigation.  I do have an XDS100 v3 JTAG debugger, which I have previously used to control certain aspects of the CC2640 during my hardware verification.  So presumably I can use that to achieve some of the changes you suggest.  I'll look into that.

    As for some TI reference hardware, I have two CC2650EM-5XD boards.  I do NOT have the SmartRF06EB.  I have hacked up one of the CC2650EM-5XD boards to directly power it.  I assumed that the 32k oscillator would be enabled by default, but I observed no 32k activity.  Perhaps I need to program the part?

    Finally, some of the behavior I see seems extremely strange for silicon to exhibit.  I.e. after a cool-down and the oscillator has stopped, I let the board sit for at least a half hour to insure that it is back to thermal equilibrium at room temperature.  When I re-apply power and the oscillator starts back up, the AC signal level is very low (maybe 80mVpp) and it takes several minutes for the AC signal level to return to its normal level.  It only seems to make sense that this kind of behavior could only be attributed to something with a mechanical aspect (i.e. the crystal itself).  I have another crystal on order and will repeat my test with it later this week.

  • 0 in reply to Kevin Braswell
    TI__Mastermind 18780 points
    I will check some more, but as you saw, the 32 kHz crystal will not be on when just adding power to the board. It would be good if you could do any of the test I suggested previously or tried to see if you saw the same behavior with a different crystal. I see that testing with our HW would requiring you adding wires to use the XDS100v3 debugger, but it would be a good exercise.
  • 0 in reply to CHS
    Intellectual 295 points

    Being a hardware guy, I have had no issues making progress in a couple of areas.

    1. I replaced the original crystal (CM13032768DZF) with one that I believe was "approved" in Rev C of the Crystal Selection App Note (FC-12M 32.7680KD-A3).  The AC signal level was notably increased (from 170mVpp to 200mVpp).  Unfortunately the oscillator still stopped when I cooled the part down to 0C.

    2. I modified the CC2650EM-5XD so that I can operate it standalone and see if I can replicate the failure on that hardware.  Using the XDS100 v3 and SmartRF Studio 7 I can control the RF (BLE) activity.  But I never see any external 32k oscillator activity.

    Maybe I'm missing the obvious, but now that I have a debug interface to the CC26xx, how do I enable the external 32k oscillator?

    I also want to use SmartRF Studio on my own hardware to disable the DCDC (as suggested).  But the customer expressed concerns about software security early on and it appears their software loaded into the part is blocking the JTAG connection.  SmartRF Studio and SmartRF Flash Programmer 2 both suggested erasing the flash, but I'm hesitant to do that until I know I can manually enable the external 32k oscillator.

  • 0 in reply to Kevin Braswell
    TI__Mastermind 18780 points
    Hi,
    I have gotten confirmation that voltage sounds low on your board, yes. There is an amplitude regulation loop, so with a range of crystals there should be ~150mVpk (300mVpk-pk) on X32K_Q1 and higher than that (~200mv) on X32K_Q2. Given that the crystal takes 1-2 seconds to start up, the oscillator probably doesn’t have enough negative resistance to function reliably with this crystal. If it had sufficient negative resistance, it ought to start in ~100ms, or at least less than 1 second. SmartRF Studio will as you saw not start the 32 KHz crystal. But if you download CCS and TI RTOS from the TI web sites you will find code so that you can start the crystal and do some more advanced testing. Can you post your schematic and layout (or send me a friend request if you want to send it to me only:))?
  • 0 in reply to CHS
    Intellectual 295 points
    Finally got some time to get back on this issue. I got the TI CC2650EM-5XD programmed with a minimal code base so that the oscillator is running, and made some interesting observations.

    1. I monitored the 32k oscillator the same as I have done on my own hardware (Tektronix TAP1500 Active Probe with low-inductance probing technique on the X32K_Q2 net at the load cap). The signal level I measure on the TI board is about 200mVpk-pk. This is comparable to what I measure on my own board (at least with certain crystals) and is much less than previous post indicated that it should be. So is my understanding from the previous post incorrect and 200mVpk-pk on X32K_Q2 is okay? Or does this indicate some concern with my measurement technique?

    2. I observe a similar issue on the TI hardware. I was able to cool the CC2650 silicon down to freezing without any issues. But when I cooled it down to about -20C, it stopped oscillating as it was warming back up to room temperature.

    As I was testing the TI hardware, I noticed all of the frost that accumulated, especially during the extended cold spray exposure. And how it melted into a notable quantity of liquid condensation. So as a test I just put a few drops of water on the PCB to simulate excessive condensation, and again the oscillator stopped.

    So it seems that this is not a temperature issue at all, but a condensation issue. Does TI suggest any coatings or treatments to minimize excessive leakage current between the X32K_Q1 and X32K_Q2 pins when excessive condensation occurs? Perhaps conformal coating is the best option?
  • 0 in reply to Kevin Braswell
    Mastermind 9135 points
    You can consider to put some "waterproof glue" on exposed pins.
  • 0 in reply to Kevin Braswell
    TI__Mastermind 18780 points

    The water will, as you saw, be and issue as this will gives you a high resistance that will stop the crystal. I would really recommend you to use a temp chamber or similar for your testing at low temp. For products intended to be used in high humidity environments, I would look into conformal coating like silicon potting material or similar.