We have several hundred of this design now operating and from time to time the output from the 48V/12V supply disappears. The time between occurrenences is long and appears to be random eg 2 or 3 occurences in total a week over the entire population.
Reapplying 48V input restores normal operation.
Our configuration is PFM mode. We have a quite long softstart period (large CSS) but had previously noticed this identical condition on a development unit a long time ago which used just 26u for CSS (see previous thread where we raised it). This particular unit failed every few days on average. We dismissed the issue (unfortunately) as a one off damaged component. As a precautionary step we also decreased the FB resistive divider values to reduce any noise on this high impedance circuit.
We were fortunate to have one of our lab units in a test bed system exhibit this fault. Inspection of the unit found no obvious external conditions to the chip:
- Vin: Ok, and noise/spike free [also during normal operation]
- Enable(EN): 6.64V [also during normal operation]
- VHYS: 174mV [also during normal operation]
- FB: 0V
We note howover that:
- VSS: 185mV, and steady (noise free + constant)
- No activity on SW pin. In fact SW is HiZ indicating that neither pull up or pull down FET ever switch ON
- Pulling EN low and releasing restarted correct operation of the "latched up" unit.
It is noted that other designers here have experienced and reported similar issues with the LM5166.
It appears the LM5166 component has internally latched up. Can anyone assist?:
- What is the trigger event, and if this is external stimulus how to eliminate it?
- Is there a workaround to ensure the device exits this condition and restarts?
Thanks, Craig
UPDATE 12th April 2021: ( I am modifying original post because I cannot for some reason insert images within a reply)
We have been investigating this further and found if noise is injected onto either the EN pin or the FB pin, the LM5166 chip will quickly shut down and not attempt to restart until a power cycle. It would appear that the internal logic of the LM5166 perhaps latches up.
As a starting point white noise of about 3V pk-pk was superimposed onto the 7V DC present on the EN pin. The EN voltage is always much greater than the enable threshold. The switcher 12V output can be seen to colapse after several seconds and removing the white noise does not see the LM5166 restart - a full power cycle is required.
From other measurements the EN input seems to require a state to be valid for 1.3us or longer to be recognised logically as an enable or disable condition. With this in mind we tried applying step changes rather than noise to the EN input. Pulse widths of roughly 0.3 to 0.8us at 3V pk-pk also cause the LM5166 to shutdown and stay latched in this condition. 1V pk-pk has no effect and can be tolerated indefinitely.
The preceding effect can also be replicated by applying transients to the 48V supply input of our design (ie 48V down to 4~8V spikes of 0.3 to 0.8us duration at some tens of khz). Application of the stimulus with a lower repetition rate has not been effective in quickly reproducing the problem.
Placing a 100pF capacitor from EN to gnd provides a significant improvement but the LM5166 will still latch from time to time. Alternatively with 100pF on the FB pin to ground we have to date been unable to effect a latch up with stimulus apllied to the 48V input..
We are wondering if:
- the LM5166 is susceptable to internally generated switching noise under certain operating conditions , which will also be dependent on the design implementation and so not seen in all designs.
- addition of the 100pF on the FB pin is providing a low impedance path and reduces internal silicon noise which otherwise results in occassional random device latch-up from time to time
Any thoughts? Thanks.
UPDATE 20th April 2021: