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.

LM3150: Unity gain bandwidth

Part Number: LM3150
Other Parts Discussed in Thread: TPS53355

Hi,

I'm working on a design that uses the lm3150 to generate 1.0V 12A (from 12V) for the core of a CPU. This works fine, however the CPU has a voltage scaling control that is connected to the FB pin of the device through a resistor. I am told (by the CPU vendor) that this should work once the unity gain bandwidth of the DC/DC is in the range 15-300kHz. I can't find this info for this device - can you help?

I am finding the 1.0V is scaling up to 1.6V, which it shouldn't so I'm trying to debug and it's obviously to do with the feedback control from the CPU - can you help with the query above?

  • Be aware the LM3150 is based on a Constant On-Time control loop. There is not a typical error amp and reference. There is a comparator that trips when Vout has sagged to a lower limit. This has an effect that Vout will rise at light load. Also, this class of control loop requires ripple voltage injection into the FB input to operate with regular (pseudo constant frequency) switching action rather than being totally hysteretic.

     LM3150_FB.docx

    Attached is a schematic that should improve light load regulation but it has not been tested for a situation of dynamically adjustable Vout.

  • OK so this part doesn't sound suitable for my AVS application as I've described. The TPS53355 is probably a better bet?

  • Ciaran ,

    What is the Switching frequency of your buck design ? My suggestion would be to choose on time such that the frequency is in 100Khz-150KHz range . You may have the issue of Min on time if you are using at higher frequency .
    The COT regulation architecture(like in LM3150) allows for fast transient response and requires no loop compensation, which reduces external component count and reduces design complexity and is well suited for AVS application .

    ---Ambreesh
  • Hi Ambreesh,

    Thanks for your reply. I would prefer to use the LM3150 as I have used it in 5 other places on the board - just not with AVS in the other cases. In this case it switches at 100.24kHz according to original Webench design. I had thought the feedback mechanism of the LM3150 wasn't suitable from previous comments below? Here's the schematic.

  • Hi Ambresh, Alan,

    Do you have any further feedback as to the suitability or not of the LM3150 for AVS as I've described and shown in the schematic?

    Ciarán

  • Hi Ciaran,
    Are you using the current sink at AVS_VDDFB to change the output voltage ?
    Further LM3150 is a COT controller and Typical COT hysteretic controllers need a significant amount of output capacitor ESR to maintain a minimum amount of ripple at the FB pin in order to switch properly and maintain efficient regulation. You can try adding a ripple injection circuitry across inductor which ac couples into FB node . Please follow below app note for more details .
    www.ti.com/.../snva166a.pdf
    The idea here is to get stable switching .

    ----Ambreesh
  • Hi Ambreesh,

    The net at AVS_VDDFB does change the output voltage - or at least that's the intention. It's connected directly to the CPU so I have no actual control over it. Unfortunately there's little documentation on the AVS signal from the CPU.

    I've noted your point about the ripple injection but I don't think this is an issue as I have a stable output and the switching waveform looks stable to me. Furthermore, doesn't the 'ERM' in the LM3150 eliminate the need for artificial ripple injection to the FB pin? I should also mention - when the AVS_VDDFB pin is disconnected I have a stable 1.0V also.

    I've attached a waveform showing the 1.0V jumping to 1.85V. You can see the feedback pin signal also. It drops from 0.6V to ~0.5V but is never driven back up. Once the CPU comes out of reset the rail and FB drop slightly.

    Here's the same waveforms taken from a TPS53355 1.0V implementation with the same CPU. This works as expected. Why is the LM3150 failing to operate in the same manner or why would the AVS signal drop so low in the case of the LM3150 and not the TPS53355?

    Ciarán

  • Ciaran ,

    When AVS_VDDFB is connected , Is the Switching stable?

    Please remove AC coupling capacitor C1371 and reevaluate the performance . AVS-VDDFB is messing up the feedback and that would be aggravated by the AC ripple injection through C1371 .


    Further LM3150 does have ERM but since the Low side FET impedance is very small , so the ripple generated would also be very small and insufficient to ensure the stable operation .

    ---Ambreesh
  • Hi Ambreesh,

    OK some progress today thanks to your hints. When I remove C1371 I get a 1V output (rather than 1.85V) but it has a lot of ripple/noise and is unstable.

    I tried different values of C1371 then with higher impedance than the 4.7nF and found 1nF to be best. This is quite good except for a period of instability early on:

    I'm looking into adding the ripple injection circuit that you suggested. I'm unsure of the value for R582. Do you think this will stabilise the system with the AVS now?

  • Ciaran ,

    Before making any further changes , can you add 200milliohm resistors in series wirh C1379 ,C1380 and C1381 . THis should be enough to generate the ripple for stable switching .

    Once you have it , you can easily then try to recreate the same with ripple injection circuit .

    Please go over page 7 and 8 of below app note for designing the ripple injection circuitry . Also you may use Webench too for the same . Choose Low o/P ripple option in Webench for the design .

    www.ti.com/.../snva166a.pdf


    -----Ambreesh
  • Hi Ambreesh,

    Thanks for suggestion. I actually had a similar thought yesterday and added 30mR series resistors to the output caps. It resulted a similar unstable condition to where we started - the 1V jumping to 1.85V. I tried 150mR today on each of the 2 output caps (C1381 isn't fitted) and the result was similar.

    We then modified the circuit to add the ripple injection as per the Webench suggestion but this wasn't any better than when I modified C1371 to 1nF. ie it was generally stable except for a period during the first 50ms. The mods were quite neat so I don't think there's additional pickup. It's beginning to look like this can't be made stable for this CPU?

  • Ciaran ,

    You can increase the magnitude of ripple injected to the feedback by decreasing  the Rr resistor . Can you try decreasing the Rr resistor to 100k and re evaluate the solution .

    Further the SGND connection is very important and sensitive to the layout ie  it Should be connected to PGND at a single point.

    The resistor needs to be referenced to SGND(similar to Css cap ) .Further the coupling cap Cac should be placed closer to FBnode .

    Please refer to below layout of the EVM for more details .

    http://www.ti.com/lit/ug/snva371d/snva371d.pdf

    ---- Ambreesh 

  • Hi Ambreesh,

    Thanks for your assistance on this. I went through all of your suggestions and in the end I found that the AVS from the CPU was operating incorrectly and my original circuit worked fine once I fixed this.

    Ciarán