LM5021: Start up Problem

Hello Kuramochi-san,

To answer your latest question: your understanding is correct - the hiccup operation (for overload detection) should not function if COMP voltage is below ( Vss-ocv – 0.6V ), which is typically 4.6 V. But the datasheet indicates that this hiccup threshold could range from (6.1-0.4 = 5.7V) down to (4.3-0.8 = 3.5V), considering worst-case conditions. I doubt if the ICs’ actual thresholds are far from the typical value. I assume you ask this question because COMP was measured to be less than 4.6V, yet the circuit still behaves as if it is in hiccup mode. Is this a correct assumption on my part?

I had a similar thought, but the timing of the hiccups does not match the component values shown in the schematic diagram. Is the change from 0.1ohm to 0.2ohm the only change that was made? The schematic shows the 0.2R circled in red, but also shows 3 other components circled in red. Do these represent other changes that were made but not mentioned in the original post? Hiccup mode should cycle with timing based on the SS cap. With 47nF at SS (C89 473), the hiccup rate should be approximately 1 second, but the oscilloscope shows about 130ms. Either C89 is a different value than shown, or the hiccup may be due to a different condition than overload detection.

I suggest to look at the VIN voltage during start-up. This can be another source of hiccupping, if VIN falls below the UVLO level too soon. Here is how I suggest this might happen:
Increasing Rsense (R91) by 2x decreases the peak current to ½. In a flyback design, either CCM or DCM, this decreases the power throughput to ¼ of the previous level. So it will take 4x longer to charge up the output capacitors, and also decreases the energy available to charge the VIN caps C41 and C87. C87 is shown as 106 = 10uF. Conversely, decreasing Rsense from 0.2R to 0.1R increases the throughput power by 4x, which increases the energy available to charge C87 higher, faster.

Between hiccups, C87 must charge from ~7V back to 20V with current from R137-R139. Assuming 200Vac input, 282Vdc / (3x470K) = 200uA – 18uA (VIN start-up) = ~180uA / 10uF = 18V/s => 0.72s to charge up 13V (7V to 20V, each hiccup). This also does not match the ~130ms hiccup rate, if C87 = 10uF. But I believe that 20-V DC-bias on a high-value, small-dimension MLCC cap can bring the actual capacitance value quite low, so the “10uF” cap may only have 1~3uF incremental value which can account for the 130-ms restart repletion rate. It can also account for faster than expected droop of VIN during the start-up switching, when the load on C87 is high. I admit that there are several assumptions here for this scenario to work, but they are not unreasonable assumptions. Still, they need to be verified.

As a test of this, I suggest to add an external 10-uF aluminum electrolytic or tantalum capacitor in parallel with C87, and observe its effect on the hiccup. If my hypothesis is correct, this should keep the VIN voltage high enough for long enough time to start-up without hiccupping, even with ¼ the power capability. This then suggests to get a better VIN cap, or maybe reducing the value of R140, or some other means to keep VIN higher, longer. Of course, if I am wrong, then further investigation on COMP, SS, VCC and CS will be needed to determine the root cause of this hiccup and account for its apparent repetition rate.

Regards,
Ulrich