TMS320C6655: TMS320C6655

Herewith some more information about this issue, as requested in the previous thread:

Using the feedback/questions from TI, I analyzed more in details what happen on a few “defective boards” and found the origin of the problem and a solution.

Explanation of the problem:

-          We know for a while that there are impedance variation on the C6655 DSP’s core power rails (CVDD and +1.0V). This impedance variation can be measured when the board is powered off and we can see a decrease by a factor of 2 or 3 when we compare a “good” and a “bad” board.

-          The symptom of the problem is an over-current flag from the DC/DC controller (IC16 – UCD9222).

-          I made a lot of measurements and see that:

• The IC5 (UCD7242) over-current fault is not triggered
• The current sense of IC5 is correctly sent to IC16
• During the startup time of both power rails, there is an overcurrent during the ramp that occurs at about 0.2V-0.3V.
• This inrush current has its peak value when the rail is about at 250mV and then decrease to the nominal current of the rail
• Example data:
  • The +1.0V has a nominal current of 2A, but practically its current is about 1A.
  • On most board, the inrush current is limited to about 2.5A.
  • On some boards (the problematic ones), the inrush current can peak up to 3 to 4A.
  • The detection point of IC16 is at 3A. So any board with inrush current higher than 3A is automatically powered-down an cycle because IC16 power down the rail and the PLD sequencer restart the whole board.
• The inrush current seams to decrease with temperature, so sometimes you can have a board that will restart all the time during 1 minute (because of our sequencer) and then stabilizes in a functional state
• On the attached screenshot (TEK00011.PNG), you can see:
  • In yellow, the rail voltage that starts ramping
  • In blue, an image of the rail current with 1A = 444mV
  • Peak inrush current occurs at 250mV
  • DC/DC is stopping because the current goes too high from the threshold (threshold at 3A = 1.32V)
• On the attached screenshot (TEK00012.PNG), you can see the same situation where the inrush current is a bit lower, so the voltage ramps up to the good value

Solution to this problem

-          I made a lot of investigation on the DC/DC controller IC16 (fully programmable digital chip). Basically I re-used the programming files from TI reference design since this part of the power supply was almost a copy/paste of their design. Since we have issues, I decided to go deeper in the parameters and check if the reference design was good or not. Basically I made some differences in the programming files to be sure it reflects 100% of our own design, but it didn’t change anything to the described behavior.

-          The whole design is very protected against overcurrents : PTC fuse at the input, overcurrent/thermal protection inside power chip IC5 and current monitor in the controller IC16. So I decided to change the detection threshold inside IC16 (moving them from 150% to 220% of nominal current). It was better, but sometimes some boards had inrush currents larger than the new threshold.

-          The final update I made was the good one : having the power rails rising faster. The inrush current is only present at specific very low voltages during ramp up. The original design had ramp up time of about 4ms, which is not very fast for such a DC/DC. Since nothing in the DSP’s datasheet restricts the minimum rise time (only restriction is to have both rails up in less than 100ms), I decided to lower the rise time to 1ms (which is the value used for others DC/DCs on the board and the result is positive: As the “bad zone” near 250mV is crossed faster, the inrush current is lowered. You can see the result on attached screenshot TEK00017.PNG

tek00011.png

tek00012.png

tek00017.png