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.

OMAP3530 / TPS65920 CLOCK SOURCE / FREQUENCY

Other Parts Discussed in Thread: TPS65920, OMAP3530, OMAP3503, TPS65950

Hello.  I am designing a portable system that will use the OMAP3530 and the TPS65920.  I have no audio in this application so I am not limited to the 26 MHz usually required in these situations.  I need to minimize power consumption as much as possible, hence I have several questions about the clock source and frequency between these two ICs.  Please answer all of my questions, if possible, but they all boil down to this question:

1) Is it better from both a power savings AND performance perspective, to have the OMAP3530 or the TPS65920 as the clock master?  If the OMAP3530 as the clock master is the better option, is it better to use a crystal, CMOS clock source or clock squarer source, and at what frequency?

The nitty-gritty details questions:

2) If TPS65920 is the clock master, it requires either a 19.2 MHz, 26 MHz, or 38.4 MHz HFCLKIN (sine wave or square wave clock, not a crystal).  Is there any power savings in using a sine wave input clock vs. a square wave input clock?

3) If the OMAP3530 is the clock master, it may be either a crystal, CMOS digital clock, or a clock squarer source.  Of these three, is there any power savings in using one vs. another?  Can you give a relative ranking of these three as far as power savings?

4) The OMAP3530 may use crystals ranging from 12 MHz to 19.2 MHz.  Is there any benefit to using a faster crystal vs. a slower one?  With the flexible range of the DPLL circuits it is possible to generate the fastest ARM and DPS clocks from even a 12 MHz crystal.

5) As far as the OMAP3530 / TPS65920 power saving modes (e.g., sleep, performance points [OPP], etc.) does having the OMAP3530 as the clock master result in any reduced power saving modes by the TPS65920?

6) From a performance standpoint, is it better to have the TPS65920 as the clock master or the OMAP3530?

6) From a power savings standpoint, is it better to have the OMAP3530 as the clock master or the TPS65920?

7) If the OMAP3530 as the clock master is the better option, is it better to use a crystal or a clock source, and at what frequency?

8) Is there any downsides, or points I am missing in the above question list?

Thanks!

Charles

  • Charles, overall i think it would be better to have the TPS device as the clock master.  In general, this is the typical mode of operation for most OMAP35x + TPS659xx designs.  Moreover, the TPS659xx has many power management features that work hand in hand with the OMAP35x to allow many different levels of power optimization.  Most of the open source software available will also support this configuration.  SIgnals between the two devices (such as SYS_CLKREQ) were designed with each in mind, so the overall experience of managing power will be easier with the TPS device as the master.

    As far as choosing the input oscillator for the TPS device, i don't think the frequency will impact final power numbers too much.  A lower frequency oscillator will probably consume less power, but the key is that most of your power savings will come from choosing optimal OPPs for your application scenarios. 

    Also i'm not sure sine wave vs. square wave will make much of a difference.  When choosing sine wave, the clock slicer inside the TPS will be turned on to produce the needed square wave to the OMAP35x.  When inputting a square wave, this slicer can be put in power down mode, reducing power consumed by the TPS device.  However, the clock source may be using this same method to square its clock, so really you may be just moving consumed power from one device to the other, with the net effect being zero.

     

  • JJD, thank you for your reply!  I'm really under the gun to get this design finished up so I would really appreciate a prompt response.

    I understand that life would be much simpler if I just "go with the flow" as use an external oscillator on the TPS65920 and I would be fine with that choice if the TPS65920 accepted the range of crystals that the OMAP3530 may use to generate the HFCLK.  As it is, the TPS65920 must have a 19.2MHz, 26MHz or 38.4MHz oscillator clock and doesn't accept lower frequency crystals like the OMAP3530.  Generally speaking, the lower the input clock frequency the lower the power consumption.  In this portable application I really need to conserve power.  So if you please, here are some additional questions:

    1) We do plan to use all of the TPS65920 and OMAP3530 power optimization features.  Are you saying that if the OMAP supplies the HFCLK that some (or all) of these features are disabled?  If so, what features are disabled?

    2) It appears to me from reading the TRM and DM for the TPS65920 and OMAP3530 that all the TPS65920 requires for the power optimization is a local 32kHz clock, and from that it generates an internal 3MHz clock for power supply switching functions, etc.  Is some of the TPS65920 functionality disabled if the HFCLK is not generated by the TPS65920?

    3) Does not using the SYS_CLKREQ from the OMAP3530 to the TPS65920 limit the OMAP3530's functionality or low power modes?

    4) It seems to me that once the TPS65920 puts the OMAP3530 into a power-savings state that the OMAP3530 still has the 32kHz clock applied to it so that it may interrupt the TPS65920, requesting in turn the TPS65920 supply the HFCLK when needed.  This precludes the necessity for the SYS_CLKREQ from the OMAP3530 to the TPS65920.  Am I not understanding this correctly or missing something here?

    And I think the following original questions are still valid and need answers:

    5) If the OMAP3530 is the clock master, it may be either a crystal, CMOS digital clock, or a clock squarer source.  Of these three, is there any power savings in using one vs. another?  Can you give a relative ranking of these three as far as power savings?

    6) The OMAP3530 may use crystals ranging from 12 MHz to 19.2 MHz.  Is there any benefit to using a faster crystal vs. a slower one?  With the flexible range of the DPLL circuits it is possible to generate the fastest ARM and DPS clocks from even a 12 MHz crystal.

    7) From a performance standpoint, is it better to have the TPS65920 as the clock master or the OMAP3530?

    8) From a power savings standpoint, is it better to have the OMAP3530 as the clock master or the TPS65920?

    9) If the OMAP3530 as the clock master is the better option, is it better to use a crystal or a clock source, and at what frequency?

    10) Is there any downsides, or points I am missing in the above question list?

    Regards,

    Charles

  • Charles, answers are embedded...

    Regards,

    James

    Charles McCarthy said:

    JJD, thank you for your reply!  I'm really under the gun to get this design finished up so I would really appreciate a prompt response.

    I understand that life would be much simpler if I just "go with the flow" as use an external oscillator on the TPS65920 and I would be fine with that choice if the TPS65920 accepted the range of crystals that the OMAP3530 may use to generate the HFCLK.  As it is, the TPS65920 must have a 19.2MHz, 26MHz or 38.4MHz oscillator clock and doesn't accept lower frequency crystals like the OMAP3530.  Generally speaking, the lower the input clock frequency the lower the power consumption.  In this portable application I really need to conserve power.  So if you please, here are some additional questions:

    1) We do plan to use all of the TPS65920 and OMAP3530 power optimization features.  Are you saying that if the OMAP supplies the HFCLK that some (or all) of these features are disabled?  If so, what features are disabled?

    [JJD]  Features such as audio codec, USB, ADC are not available without the HFCLK.  As for power management, without control of the HFCLK gating, transitions such as SLEEP-ACTIVE and vice versa may not work properly. 

    2) It appears to me from reading the TRM and DM for the TPS65920 and OMAP3530 that all the TPS65920 requires for the power optimization is a local 32kHz clock, and from that it generates an internal 3MHz clock for power supply switching functions, etc.  Is some of the TPS65920 functionality disabled if the HFCLK is not generated by the TPS65920?

    [JJD]  This is true.  But don't forget that the RC oscillators are idled if a HFCLK is present, so you will actually be consuming more power in the TPS device when you don't supply an HFCLK. 

    3) Does not using the SYS_CLKREQ from the OMAP3530 to the TPS65920 limit the OMAP3530's functionality or low power modes?

    [JJD]  Being able to gate the external clock is important for power management.  If the TPS device does not have the HFCLK, you could use the SYS_CLKREQ to gate an external oscillator.  However as stated before, there may be some issues with the TPS event sequencing going in and out of sleep modes relative to gating the clock.  Alternatively, if you are using a crystal, OMAP will take care of shutting down the internal oscillator. 

    4) It seems to me that once the TPS65920 puts the OMAP3530 into a power-savings state that the OMAP3530 still has the 32kHz clock applied to it so that it may interrupt the TPS65920, requesting in turn the TPS65920 supply the HFCLK when needed.  This precludes the necessity for the SYS_CLKREQ from the OMAP3530 to the TPS65920.  Am I not understanding this correctly or missing something here?

    [JJD]  This is not true.  The sleep/wakeup sequences use SYS_CLKREQ as an indicator when OMAP needs a clock.  These sequences are described in the PRCM chapter of the TRM.  There aren't any commands sent to the TPS device to turn on/off the clock.  You can choose not to use SYS_CLKREQ, but that would just mean that the clock will HFCLK will always be oscillating.

    And I think the following original questions are still valid and need answers:

    5) If the OMAP3530 is the clock master, it may be either a crystal, CMOS digital clock, or a clock squarer source.  Of these three, is there any power savings in using one vs. another?  Can you give a relative ranking of these three as far as power savings?

    [JJD]  I don't know the relative ranking.  The crystal is least, but then you would require the OMAP internal oscillator to be active.  With the others you wouldn't need the internal oscillator.  So now its a matter of comparing the power consumed with int osc. vs. and external oscillator.  I'm not sure of the answer here.

    6) The OMAP3530 may use crystals ranging from 12 MHz to 19.2 MHz.  Is there any benefit to using a faster crystal vs. a slower one?  With the flexible range of the DPLL circuits it is possible to generate the fastest ARM and DPS clocks from even a 12 MHz crystal.

    [JJD] Yes, you can hit any of the ARM/DSP frequencies with any of the valid input frequencies.  The power will depend on what M and N values (mostly M) that you have to choose for each PLL to get the appropriate internal frequencies for OMAP3.  The higher the multiply factor, the more power a PLL will consume. 

    7) From a performance standpoint, is it better to have the TPS65920 as the clock master or the OMAP3530?

    From a processor performance standpoint, i don't see any impact on where the clock comes from.  If you are talking about overall system power management performance, I think TPS as the clock master is better, as that was the intended use of a TPS659xx + OMAP3530 system.

    8) From a power savings standpoint, is it better to have the OMAP3530 as the clock master or the TPS65920?

    [JJD] I don't have enough comparison data, but some of the differences i have outlined above.  As i said previously, your biggest bang-for-your-buck concerning power savings is the efficient use of the appropriate OPPs for your application.  Whether you choose 12 or 26MHz, or oscillator or crystal, those choices will have little impact compared to implementing things like DVFS and AVS in OMAP3. 

    9) If the OMAP3530 as the clock master is the better option, is it better to use a crystal or a clock source, and at what frequency?

    I think i answered this above. 

    10) Is there any downsides, or points I am missing in the above question list?

    You raise some good questions, but from an overall power savings standpoint, i don't think the choices on crystal vs. oscillator or 12MHz vs. 26MHz will make much of an impact in your final power numbers.  I've tried to give you a general picture of how each choice will impact power, but sometimes it is an issue of moving the power around, with no real net gain.  I'm sorry i can't give you any hard data to compare, we just don't have those numbers.

    Thanks,

    James

    Regards,

    Charles

  • James, thank you for your very complete response.  This helps lots!

    Regards,

    Charles

  • Dear,

    For our application we will use OMAP3503+TPS65950.

    Talking about slicer module for HFCLKIN. We will use the 26MHz square wave for this high frequency pin.

    I saw in the TPS65950's spec "HFCLKIN is the high-frequency input clock. It can be a square- or sine-wave input clock. If a square-waveinput clock is provided, it is recommended to switch the block to bypass mode when possible to avoid loading the clock"

    How to bypass the slicer while using the square wave input at HFCLKIN pin?

    Kind regards,

    M.MIK