• State TI Thinks Resolved
  • Locked Locked
  • Replies 4 replies
  • Answers 1 answer
  • Subscribers 98 subscribers
  • Views 492 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.

AM5748: Power consumption spike on move from AM5728 to AM5748

Michael Shklyarman
Genius 12740 points
Part Number: AM5748
Other Parts Discussed in Thread: AM5728,

Champs,

Customer moved from AM5728 to AM5748 on exact same board and runs exact same (RTOS based ) binary that ran on AM5728. With AM5748 they observe power consumption going over 2W higher than AM5728 (1.36 vs 3.6W)! To understand the reason we ran PET on their use case for both devices and PET simulation confirmed the measurements. The use case is fairly simple ( single A15 core, DSS, QSPI/SPI/UARTs). Questions I have are

1. it looks like AM5748 PET does not permit running single A15 and shutting down the second one. Is it not possible in the AM5748 silicon?

2. AM5748 PET does not permit disabling C66s but puts them into AutoCG mode - what exactly does it mean? Do C66x consume considerable power in this state? is it possible to shut them down completely?

3. Can powered-up second A15 not running any firmware and C66x in AutoCG state account for >2W extra consumption?

The extra 2W causes serious unanticipated thermal issues in customer's system, so any insights will be greatly appreciated!

thanks

Michael

  • 0
    TI__Guru**** 393215 points
    Hi,
    Please note that due to US holiday the responses may be delayed until the week of November 26th.
  • 0
    TI__Genius 13567 points

    Hi Michael,

    1. The AM574x silicon has power switches removed for PD_MPU, PD_DSP, PD_EVE, and PD_CAM. These domains will always be consuming leakage power even if they are not clocked.
    2. In use cases where all these peripherals are used anyway then there is no power increase. If you're not using these peripherals, such as in your single core use case, then the leakage power of these domains now needs to be accounted for. Your overall power consumption has always been sensitive to the junction temperature but it is more pronounced in AM574x.
    3. Depends if you are using the TMS or TMX silicon.

    Self heating of the chip and an increased junction temperature contribute to the power consumption and that in turn pushes the temperature. So accommodating the AM574x will benefit from more cooling. That can come in the forms of a larger heat sink, more copper per PCB layer, or forced air.

    Regards,
    Ahmad

  • 0 in reply to Ahmad_Rashed
    TI__Genius 12740 points

    Hi Ahmad,

    Could you please clarify what you mean by "depends if you are using TMS or TMX silicon". I understand the leakage current will have to be added to overall consumption but in this particular case we have over 2W added! Can this really be attributed to leakage current in DSP/EVE/second A15 ? it seems way too much

    thanks

    Michael

  • 0 in reply to Michael Shklyarman
    TI__Genius 13567 points

    Hi Michael, I replied to soon! Did not mean for that to answer your 2 Watt question. This is more FYI... 1) early chips did not have the optimized voltages programmed and only the max voltage per rail was set and 2) the process target during TMX phase was not finalized and that had large effects on leakage. Those two points were to say that power measurements on TMX parts are not fully indicative of expectations from TMS production parts. 

    On the 2W question, you mentioned that the PET confirmed the bench measurements? What were the operating conditions that were entered? When you take power measurements can you separate them per-rail?