• State
  • Locked Locked
  • Replies 1 reply
  • Subscribers 99 subscribers
  • Views 695 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.

CVDD and SmartReflex for TMS320C6657

David Mendel
Prodigy 15 points
Other Parts Discussed in Thread: LM10011, TPS54320, TMS320C6657, LM10010

We have had some issues with the CVDD SmartReflex voltage levels, and it seems that bumping up the voltage level is beneficial. The following is what we have observed and tested:-

 Problem:

  1. After correct power sequencing and coming out of reset, we had a problem where frequently some DSP registers could not be accessed through the  JTAG debugger. When the JTAG worked, we had many DDR3 bit error problems.
  2. We use a TPS54320 to drive the CVDD, and a LM10011 as a D/A of the 6-bit VCNTL signals to the TPS54320 feedback. See attached schematic.
  3. I looked at the various power rails and the noise was minimal. 
  4. I then investigated the VCNTL lines from the DSP that set the core voltage. This is a 6-bit digital code that is translated into voltage steps between 0.7V and 1.1V.
  5. After reset, the DSP delivers a “42” which translates to a desired 0.969V, and the delivered voltage was spot on.
  6. I first removed this VCNTL feature so that the core was a continuous 1.1V (LM10011 Rset=267K) and there was an immediate improvement. However TI warns that long term reliability can occur if the part stays in reset when 1.10V is normally delivered.
  7. I then re-enabled the VCNTL circuit, and changed the range to be 0.81V to 1.10V, and “42” delivered an expected 1.009V.
  8. JTAG loaded every time and memory tests to the DDR3 memory ran overnight.
  9. Attached spreadsheet shows calculations and effect of resistor changes on what DSP wants and what it gets.
  10. Subsequently modified to the .88V/1.1V range with Rset=200K ohm, and applied this to three boards.
  11. So the first/original board is doing great. The 2nd board still had issues, but when the LM10011 was removed from the circuit (R951 removed) and CVDD = 1.05V it always behaves when coming up, and gets an infrequent DDR3 bit error. The 3rdboard still has occasional JTAG start-up hiccups &/or DDR3 bit errors but it is better than it was.

 Questions:-

  1. While we are in the prototype phase with potentially multiple issues, is it OK to run the voltage higher than requested by the DSP?
  2. Is it OK to go below “63”/1.1V at startup while the DSP is in reset?
  3. Does the TMS320C6657 operate as a Class-0 SmartReflex (one-time fixed) or as a Class-3 SmartReflex (updated based on various dynamic factors)?
  4. Is it possible that the factory set SmartReflex voltages are too “tight” (targeting the lowest voltage for heavily filtered power)?
  5. What else should we be investigating with regard to this issue?

    1300.LM10010_TPS54320_Calc2.xlsx
    7142.930-1081-01_Superset_Sch_1_39V5_sht47.pdf
  • 0
    TI__Mastermind 42545 points
    • While we are in the prototype phase with potentially multiple issues, is it OK to run the voltage higher than requested by the DSP?

    Generally in a prototype phase it is alright to operate at a slightly higher voltage as long as the device is properly cooled.  The device will use more power then was calculated with the power consumption model.

    • Is it OK to go below “63”/1.1V at startup while the DSP is in reset?

    The specification for start up voltage is 1.1V.  If a voltage level lower then 1.1V is used it is possible that some units may not exit from reset correctly.

    • Does the TMS320C6657 operate as a Class-0 SmartReflex (one-time fixed) or as a Class-3 SmartReflex (updated based on various dynamic factors)?

    The TMS320C6657 operates as a Class-0 SmartReflex device.  It will present the voltage needed by CVDD once after the device exits reset.

    • Is it possible that the factory set SmartReflex voltages are too “tight” (targeting the lowest voltage for heavily filtered power)?

    All TMS320C6657 devices are tested at the voltage programmed into the device by the factory. 

    • What else should we be investigating with regard to this issue?

    Be sure that you've checked the voltage level of the power supply directly at the pins of the device or as close as you can get to the pins.  Often we've encountered customer boards that, due to layout or faulty decoupling, experience some voltage drop when measured at the vias under the processor itself.  Due to the nature of the device the current requirements can change very quickly. Check to be sure that you are not seeing a drop in voltage at the part or droop in voltage during operation.  You may be compensting for this by raising the voltage.

    Regards, Bill