LTC3118 power converter issue

Hi!
We use LTC3118 in our system for the voltage buck converter/battery back-up because of its dual power path facility. The output voltage is 5V. The input voltages are 12V and 3.6...4.1V (Li-Po battery).

We used the same design in previous PCB production in the last two years without problems, but unfortunately, within the latest board production batch we’ve met the following strange power supply behavior:

  • The IC temperature is significantly higher than it is in our boards of previous batches (30% or more).
  • The power supply that based on LTC3118 barely bears its typical load. The actual output voltage value is always less than 5V. It depends on the load current and can vary within the limits 4.0...4.9V. The output voltage can reach 5V only without any load.
  • The power supply completely refuses to work as a boost converter from a Li-Po battery.

The strange thing is that the boards of previous batches were normally working. We just moved the on-board power block a bit and slightly fixed up the PCB layout concerning this. There were no PCB layout reworks or/and significant reducing of the heat-sinking copper foil polygons. We tested the open boards, not inside a cramped enclosure. We even used a fans but it hasn't solved the thermal issue.

The steps we already made about investigating and fixing this issue:

  • We checked the PCB design for errors and compared the old and the new board versions
  • We measured the actual current consumption. It is seldom exceeds 1A that is normal for LTC3118 
  • We examined the board with a thermal imager for problem circuitry and unexpected current consumption. Nothing.
  • With a manufacturer we examined the IC soldering quality in X-rays. Especially paying an attention to thermal pad welding. It is okay.
  • We even replaced the IC with the one from an old board that has been working.
  • We checked the BoM and the actual components, especially in the feedback and compensating circuitry, the bootstrap capacitors. Everything’s alright.
  • I modelled the scheme with LTSpice simulating a bad vias conductivity in a current paths. Their impact  is not so significant and doesn’t cause an overall scheme work fail (as we can see in the real board).

Are there any important things to pay an attention? What else to look for?

I am actually in a standstill. We ordered a new PCB with a thicker copper foil (70μm vs. 35μm in actual board). The manufacturer said that he changed a PCB supplier in the current batch.

Thanks.