MPM3833C max output capacitance

Integrated Inductor Power Modules
1

I see that there are many topics in the support forum concerning this issue, but I have not seen an exhaustive discussion of maximum output capacitance for integrated power modules.

Search results for ‘max output capacitance’ - Monolithic Power Systems’ Technical Forum

I am using an MPM3833C as a supply regulator for AU15P Xilinx Artix UltraScale+ 0.72V VCCINT core voltage rail. The expected load is ~1A. I have over 500uF of capacitance on this rail, per Xilinx recommendation from their PCB design guidelines document.

I am observing that the output is unstable and continuously oscillates at 20kHz with 50mVpp amplitude under 1A load. Regulation is marginal for the Artix voltage rail requirements. I can replicate the behavior on an EVM3833C evaluation board by installing 6x 100uF Murata MLCC capacitors in parallel with the output. After observing this oscillation, I notice in the MPM3833C datasheet that the max operating output capacitance is 100uF.

image
image1254×751 128 KB

One post on the support forums suggests that COT devices are inherently intolerant of excess output capacitance.

MPM3632S, R4 and max capacitive load - Integrated Inductor Power Modules - Monolithic Power Systems’ Technical Forum

Meanwhile, MPS has several reference designs intended to power an Artix UltraScale+ AU10/15P, which all use multiple-output COT regulators. I don’t see any discussion of maximum output capacitance in the datasheets, but nothing in the test report documents suggests that tests were done with anything other than the 44uF on the module evaluation board, significantly lower than the Xilinx recommended capacitance.

Another project team within my organization has had success with using an MPM38222 as the 0.72V core voltage regulator with the same AU15P Xilinx FPGA. Notably, this regulator uses peak-current-mode control, as opposed to COT-mode control. Also, a separate post on the MPS support forums states that the MPM3834C can support up to 1.5mF of output capacitance. So I will try one of these modules for my application.

My questions are the following:

  • Is there a recommended single-channel regulator module for the Xilinx AU15P FPGA core voltage? Many MPS regulators do not list a maximum output capacitance in their datasheets, so it is difficult to determine which will be suitable. If there is no recommendation, I will try to move forward with the MPM38222 or MPM3834C.
  • Some sources state that inserting a small series resistance or inductance in the output before the bulk capacitance will isolate the bulk capacitance from the control loop, but I have not seen a procedure for selecting appropriate values of resistor/inductor. Is there an app note for this? I am hesitant to add series impedance to a digital voltage rail.
  • The question of maximum output capacitance has come up frequently on the MPS support forum, as noted above. Is there an app note discussing the impact of large bulk capacitance on the module control loop? It is common with FPGAs to have core voltage rails with several hundred uF of capacitance, so it would be easier to select appropriate regulators with a better understanding of the effect of large bulk capacitance, especially with COT-mode regulators.
2

The big issue with PSU’s is when the phase angle of the error amplifier (feed back pin) is moved so the regulator simple shuts down. You can avoid this by placing an inductor before the final output cap.
Have a look at this article:

it explains about how to configure a second stage filter to avoid the phase shifting and maintain the stability of the converter regardless of the capacitance load. I use this in all my designs.

Simon

3

Hi Simon,

Thanks for the link. However, I don’t think this applies, as the article suggests adding the second-stage filter within the control loop. The 600uF load capacitance in my application pushes the power stage primary double-pole down to ~6.5kHz vs. the ~33kHz with 22uF, regardless of whether it’s part of a single-stage or two-stage filter. The problem is (I presume) the internal compensation is not designed to provide the necessary phase boost at that low of a frequency. Putting the secondary filter outside of the control loop would isolate it somewhat, but introduces a series impedance to the power distribution network.

The article states that its filter is insensitive to load capacitance, but I think it’s still making the assumption that any added load capacitance is equal to or small relative to the large (2.5F!) output capacitor. In my case, I am adding ~26x the suggested output capacitance.

4

In the case of an FPGA, it’s the output capacitance which dictates in the impedance not the filter. Xilinx usually specify 5 or 6 different capacitors of various sizes for each power pin, all low ESR, so each cap covers a different part of the frequency band. This in effect creates an extremely low impedance which the PSU tops up.
In past FPGA Designs, I’ve used several 2A/5A Ferrite bead for the same function, no ill effects (ARM Devices with 3 x 1G serdes Ethernet). From memory, there were 30 or so capacitors doing the decoupling.

Of course, if there’s an eval board, grab one and try it. It will give you some idea, usually unloaded with lots of capacitance will show if it will regulate or go to a funny voltage.