@Edo Jayakusuma First circuit has positive feedback so it goes full switched on, transistors go out of saturation region so current mirror is not mirroring currents as it should. Second picture is with negative feedback so you are mirroring the same current into the 2 sides of the circuit and biasing M3 and M5 sources to the same voltage, with

kT ln(2)

voltage on R1. Also consider that even the correct circuit in real word may not start since the off state is a metastable point. You must include a startup circuit that ensures Mos transistors go in the ON state. I also suggest incrementing the BJT ratio, instead of 2:1 go for 10:1 ratio (or more) so the thermal voltage increases to

kT ln(10)

. This reduces mismatch errors. Also, do not use 1v8 transistors in a 5V design. These are for 1.8V max.

Hi @Stefan Schippers thank you for the input🙂, I think those are solid suggestions. For the positive feedback, I think you are correct but I can't see where exactly is the positive feedback

The trick is to plot the V-I characteristics of the voltages on the source terminals of M3 and M5. Left BJT is smaller so has higher voltage at small currents. However right BJT although bigger has a series R1 resistor, so at high currents it will show higher voltage. If you go above the cross point, M5 source goes high, this turns off M5, so M5 drain can go higher. This will increase gate of M3 so M3 current will increase. The mirror will increase the current on M5 and you are in a runaway condition (positive feedback).

Ooh wow, that's a great explanation. So if the current of M5 source and M3 source is always below the crossing voltage I can imagine that it is stable? Which have the corollary that, it is possible to make this circuit to be stable by reducing the resistance R1 massively?