• State Resolved
  • Locked Locked
  • Replies 10 replies
  • Subscribers 62 subscribers
  • Views 731 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

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.

TPS61093: TPS61093: Bad transient response at nominal load current in EVM as well as custom PCB

Adam Pasternak
Intellectual 620 points
Part Number: TPS61093

I have a TPS61093EVM  evaluation board from TI, as well as a custom made PCB featuring this part in an embedded application.   Both the EVM as well as the custom PCB are suffering from fairly extreme excursions in the transient response and I cannot figure out what is happening.    The fact this occurs on both the EVM and the custom PCB suggests this is not a layout issue on the custom PCB.  

First, the evaluation board TPS61093EVM.    Input voltage is 2.2V.  Output voltage is set for 6.5V ( R1 = 121k, R2 = 10.2k )  by unsoldering/soldering appropriate value R's.    Load is pulsed between 0mA and 70mA, 250Hz, 50% duty cycle.   The load is an electronic load, verified through thorough testing to be accurately loading this output at 70mA.    Everything else on the TPS61093EVM is untouched and stock as in the datasheet & user guide.   Here's the output waveform, measured using a low-inductance barrel ground lead, directly across the output capacitor, on a 1GHz Agilent scope:

As you can see, I'm seeing negative excursions of almost THREE VOLTS when the 70mA load is pulsed on, and then almost 600mV of overshoot and slow recovery when it is turned off.   This looks almost nothing like the waveform in the TPS61093EVM datasheet.  

Adding a Cff of 10nF seems to quell the problem somewhat, but WHY is such a large value for Cff required here?   

Now, the TPS61093 on the custom PCB is laid out in more or less identical fashion as in the datasheet suggestion and the EVM.   Component values are a bit different, however.  In detail:    Cin=4.7uF 0603.  Cout=2.2uF 0603.   Inductor is 4.7uH LQH2HPZ4R7MJRL with 1.2A Isat rating ( this is 2x the maximum L current in this design, btw.  I think L saturation is not an issue here)   SS pin is connected to 220k/220nF in parallel. Cp cap is 220nF.  Vo cap is 220nF.  The green waveform is the applied load current of 70mA.  Yellow waveform is the Co output waveform, without using a Cff capacitor:

Now, adding Cff = 10nF to the custom PCB helps matters somewhat, but still seems out of spec with regards to the transient response:

 In the yellow waveform above, it seems Vout doesn't even recover in time to reach the target voltage ... the long slow RC ramp of ~4ms seems quite out of spec, as well as all the other transient behavior of this design. 

Can anyone at TI or external provide any insight to what is happening here? 

Why the large excursions negative/positive on both the EVM and the custom PCB?  Am I not allowed to pulse between zero load and nominal load with the TPS61093?  For the record I have tried all sorts of inductor values, all sorts of output capacitor values, and a wide range of values for Cff;  the values stated above in the custom PCB give the  "best" response according to my empirical observations, but that is still very short of expected behavior.  

Thank you for reading!

  • 0
    TI__Mastermind 46905 points
    Hi Adam,
    Thanks for the detail experiments in the EVM and recording the issue clearly.
    could you also help to measure the VIN voltage with the VOUT? make sure there is no large voltage drop in the VIN.

    The output capacitor is not enough for this application. I would suggest to increase the output to 10uF or higher to help the load transient and stability. After that, we can following this application note to optimize the Cff "www.ti.com/.../slva289b.pdf" .
  • 0 in reply to Jasper Li
    Intellectual 620 points

    I have been researching this issue some more and I have found that many boost converters are subject to a minimum load;  that is to say, below some minimum load, the output is not guaranteed to be in regulation.   This may be because the device enters DCM or pulse-skipping mode, or because the minimum switch on time creates a lower current limit below which the control loop cannot function ... but I am not sure if that is what I see occurring in this device. 

    I am indeed pulsing the load current 70mA to 0mA, and might need a minimum load resistor. 

    Does TPS61093  have a minimum required load to operate properly?    I cannot really find this stated in the datasheet directly, but the datasheet does state:

    In my applicaiton, Vin = 2.2V, Vout=6.5V

    So iout_skip = 4 * (65e-9)^2 * 1.2e6  /   2 * ( 6.5 + 0.8 - 2.2 ) * 4.7uH      =    512 uA  

    If I keep the load current > 512uA would this help the response I am seeing?   I am going to try and run this experiment this week and check it out. 

  • 0 in reply to Adam Pasternak
    TI__Mastermind 46905 points
    Hi Adam,

    The device doesn't require minimum load resistor to regulate the output voltage. it enters skip mode to regulate the VOUT if the load is very small or no load. But it need time to charge or discharge the output capacitor in load transient condition.
  • 0 in reply to Jasper Li
    Intellectual 620 points

    Here is the input voltage waveform and the output voltage waveform before the Cout modification ( from the EVM ):  

    Now see below: I changed Cout to be  2x 4.7uF X7R 0805 25V  capacitors in parallel.   The waveform shows practically no improvement: 

    I doubt that Cout is the problem here.   Using WEBENCH as well as equations from the datasheet, to satisfy my requirements for Vin = 2.2V Vout = 6.5V  Iout = 100mA   I need L  = 10uH   Cout = 1uF Cin = 4.7uF      I used a smaller inductor of 4.7uH and a larger Cout of 2.2uF     but this is within the tolerances in the datasheet.    As an aside, I also tried using L = 10uH  but that did not change the response at all.   It seems to me something deeper is happening here causing the very long dropout and recovery times shown in the scope waveforms.     

  • 0 in reply to Adam Pasternak
    Intellectual 620 points
    Can you or anybody there at TI verify this operation of the EVM? I am wondering if I am the only one who sees this type of transient response.
  • 0 in reply to Adam Pasternak
    TI__Mastermind 46905 points
    Hi Adam,

    I will double check the behavior with EVM.
    is the behavior after adding Cff=10nF acceptable? you need any improvement or just want to understand the behavior.
  • 0 in reply to Jasper Li
    Intellectual 620 points

    Unfortunately, the behavior after adding Cff = 10nF is still not acceptable.   There is a large excursion of 400 mVp-p between voltage sag, recovery, and overshoot.   It is not this absolute value in sag/overshoot that is the problem, but the recovery time.   The recovery time appears to be very different from the TPS61093 datasheet & TPS61093-EVM .   Even 2ms after the load is applied, it appears the output voltage does not recover/stabilize at its setpoint of 6.5V

    I have tried adding a dummy load of 6.5k  for  1mA load at the OFF time period in the PWM, but this does not help the transient response at all.  I was concerned about the device entering DCM or PSM and its recovery from that, but it appears not to be the case.

    I would try to find an alternate part from TI or another vendor, but the TPS61093 seems unique in that it has all the following features I require:  (a) At least 1.8V min Vin (b) At least 6.5V Vout (c) <= 3mm x 3mm area (d) Synchronous rectifier (e) Can support 0.1A output current (f)  Input/Output isolation switch      So it seems that at least for now, I am stuck with the TPS61093

    Jasper: thank you for all your help so far.   Much appreciated!   Look forward to hearing about your findings.

  • 0 in reply to Adam Pasternak
    TI__Mastermind 46905 points
    Hi Adam,

    I found similar behavior in the EVM with 4.7uH inductor, 22uF/0603/10V ceramic capapacitor and 10nF feed-forward capacitor. I found further increase the Cff help. could you try to increase the Cff?
  • 0 in reply to Jasper Li
    Intellectual 620 points

    Ran some more experiments today in the lab.  I think I have come to a solution. 

    For reference, the best transient response I was able to get with Cff = 10nF & Cout = 10uF was:

    Changing Cff = 100nF resulted in:

    Going even higher to Cff = 220nF  did not improve anything:

    Now, I needed to go back to my original Cout = 2.2uF.  I was worried this would cause instability or again long transient response, however, going from Cout = 10uF --> 2.2uF  did not affect the transient response that much, in fact, it was almost identical to the Cout = 10uF case:

    For future reference or anybody else who has this problem, the design parameters for the above response are:

    R1 = 121k   R2 = 10.2k    Cout = 2.2uF  Cin = 4.7uF   Cff = 220nF     Css = 220nF   Rss = 220k    L = 4.7uH   Cp = 220nF

    Thank you Jasper for verifying these findings and helping me get the proper response.   Even though this was solved, I am still very curious as to why in fact such a very, very large value for Cff is required.   Usually Cff are in the range of pF to a few nF ... 

  • 0 in reply to Adam Pasternak
    TI__Mastermind 46905 points
    Hi Adam
    I think the Cff can't further improve the transient response if it is higher than a special value. the Cff actually add a zero and a pole, the frequency of the zero is lower than pole. that is why Cff can improve the response. if the pole frequency is much lower than the crossing frequency, increasing the Cff will not help the response any more.

    so I would prefer using the minimum value of Cff that can meet you requirement. for example, if 100nF (or maybe lower) work, I would like to use 100nF (or maybe lower) instead of 220nF. large Cff may impact the startup time