Use this forum to chat about hardware specific topics for the ESP8266 (peripherals, memory, clocks, JTAG, programming)

User avatar
By rudy
#75455 I work for a small electronic manufacture. Another fellow and myself design our electronic products. We have programmers that write firmware for the products and some that write the computer applications.

We have a modular control system that has a lithium battery power backup for the linux based front end. The battery backup is used to provide power for a clean shutdown when the AC power fails. After the linux cpu signals that it is done the microcontroller on the power supply cuts the power. The power control switch is an ON Semiconductor NCP380.

Adjustable Current‐Limiting Power‐Distribution Switch
The NCP380 is a high side power-distribution switch designed for
applications where heavy capacitive loads and short-circuits are likely
to be encountered. The device includes an integrated 55 m (DFN
package), P-channel MOSFET. The device limits the output current to
a desired level by switching into a constant-current regulation mode
when the output load exceeds the current-limit threshold or a short is
present. The current-limit threshold is either user adjustable between
100 mA and 2.1 A via an external resistor or internally fixed.

We have the current limit set for the maximum 2.1 Amp limit. But we had a high percentage of failures. The solution was to keep changing the NCP380 until it passed the current test. This was a real pain but that is what our manufacturing department did for more than half a year. Finally enough was enough and I was asked to design a new replacement power supply with higher capacity. Unfortunately I was also asked to look at the current design and see what we could do to make it more reliable.

The core of the power supply was a chip that handled power conversion, battery charging, power loss detection, and seamless switching from the 5 volts coming in and the battery boost circuit switchover. Unfortunately it did not operate as seamlessly as advertised. I came up with fixes for all this but we still had issues with the NCP380 current limiter/switch. It was the last problem and I will show you what the fix turned out to be.

This is the example schematic from the data sheet on the part.

Looks pretty straightforward. Should be no problem. And the following is a shot of the switch on the circuit board.


The bottom of the board has a ground plane. All the GND indicated vias connect to the ground under this view. It basically is a full plane on this section of the board. The bottom of the NCP380 has a thermal pad that is soldered to ground under the chip.

The three capacitors on the bottom (22uF) are the filter caps from the boost converter. The top of the caps is 5 volts. It goes into the NCP380 and then out the top. Since the topic heading lists bypass caps (because it was the issue) I will try to make it perfectly clear what is happening here.

The top of the capacitors (+5 volts) goes into the chip. It also is the power connection for the chip. The negative side of the capacitors connects through the vias to the ground plane, to the via below the NCP380, and then into the chip. Unfortunately this was not good enough.

When I went through the data sheet for the NCP380 there was something that bothered me. Under pin function for the IN pin.

Power-switch input voltage; connect a 1uF or greater ceramic capacitor from IN to GND as close as possible
to the IC.

We certainly had more than 1uF. And we were close. But what bothered me is that if I would have done the circuit board layout I would have put a small capacitor across the input, everything on the component side. Not going through vias to make the connection.

So what I did was scraped away the mask on the ground via below the NCP380 and soldered a 1uF cap to the exposed via copper and the other side to the power input at the top of the 22uF capacitors. As shown below. And it solved the problems.

The layout had been correct as far as the schematic indicated. And it wasn't a horrible layout. And it just goes to show that even though it seems that everything was done correctly, it just wasn't good enough. Layout matters.

The problem seems to be some spikes, switching noise, from the up converter.
You do not have the required permissions to view the files attached to this post.
User avatar
By QuickFix
#75457 Impressive find, this does prove that details matter. :idea:

Thanks for the topic; it read like a detective story. ;)