Computing stuff tied to the physical world

Pretty picture, ugly effect

With the startup voltage being monitored, we’ve now seen a clean startup as the harvested energy comes in. But we’re not out of the woods yet, alas – here is another startup example:

SCR75 3

The voltage rises, at about 2.2V we switch on Vdd and then… the supply collapses?

This happened on several occasions, while repeatedly switching AC mains on and off to look for edge cases. And it occurs more often with lower current levels, i.e. when the harvested energy is not coming in as quickly.

What we see here could be called “zombie state”: a failed startup, which continues forever, unless there is a sufficiently sudden change in AC current levels to force a new equilibrium.

That’s not all. Some failure modes are quite surprising, and lead to pretty scope images:


The circuit fails to start and enters a slow oscillation pattern about 11 times per second.

One reason for this could be the R1 and C4 filter, which acts as a delay. C4 was probably increased to 1 µF in this case, while exploring different ways to get reliable startup. Delays circuits can be a problem – this needs to work no matter how fast or slow Vres increases.

Another reason is probably the lack of hysteresis – the NPN transistor is switching on at a certain voltage level, but it doesn’t take much to drop that level slightly, causing the transistor to switch off again. We need to turn on at say 2.5V, but once on we should allow the voltage to drop to say 2.0V before switching off again. That will give the µC much more headroom to consume the energy stored in the two reservoir capacitors C1 and C2.

Hysteresis is about latching to the new state, and not letting go of it too soon. It’s easy to add with a bit of positive feedback. Everyone likes positive feedback, right? Same here.

In practice, a 1 MΩ resistor between the P-MOSFET’s output and the NPN’s base does the trick: when Vdd starts rising, it pulls the base slightly more up, making it rise a little faster.

A third reason could be those two red LEDs we’re using as surrogate zeners: they do work, but the voltage levels are not well-defined and their switch-on behaviour is very gradual. That just makes it harder for the rest of the circuit to make up its mind.

We’re back to the drawing board, but with some ideas on how to improve things.

[Back to article index]