I have designed a custom pcb utilizing the ENC28J60 ethernet chip to be used with raspberry pi. The board has 5V in which powers both the ENC28J60 (through a AMS1117-3.3 LDO regulator) and the pi. Power and internet is received from a POE splitter which has RJ45 and microUSB 5V out.
This design works fine during initial testing, but when the modules are left to run 24/7, after a few weeks of operation the ENC28J60 chip gets damaged (still detected by rpi, but does not detect ethernet cable). However this only occurred on roughly half of them, the rest work fine.
On the broken boards I have diagnosed, that the 5->3.3v voltage regulator is now outputing almost 5v, basically acting as a pass through. So the ENC28J60 (rated for 3.3V) gets overvolted (5v) and breaks, that part seems clear, but I cannot figure out what could cause the voltage regulator to act/break in such a way.
The ENC has a draw of 250mA max according to datasheet meaning it is well within the AMS1117 3.3 rated 1A current. I have read about "floating ground pins" however running an additional ground to the regulator did not affect the almost 5V output. I do not think overheating is the issue, as in the enclosure, the cpu temp of the pi does not surpass 50C.
Is the regulator insufficient quality for this application of constant usage? Could the POE splitter (amazon link) have spikes in voltage that cause this? or is something else the problem here?
Schematic of ENC28J60 (green net ports go to rpi):