a current mirror
A current mirror does exactly what its name says: you set up one current, and the mirror copies it into another branch, holding that copy steady no matter what the load does. It is how you make a good, stable current source out of transistors — and inside integrated circuits, where precise resistors are hard but matched transistors are easy, it is the standard way to distribute bias currents. Think of it as a pantograph for current: trace one value, get an identical one beside it.
The basic version uses two matched transistors sharing the same base-emitter voltage. The first transistor is 'diode-connected' (base tied to collector) and a reference resistor sets its current — say a 10 kohm resistor from a 10 V rail through the 0.7 V base-emitter drop gives a reference of (10 - 0.7)/10k = about 0.93 mA. Because the second transistor has the very same base-emitter voltage and is matched, its collector current mirrors that same 0.93 mA. The beauty is that the output transistor's collector current stays at 0.93 mA whether its collector sits at 2 V or 8 V — it behaves like a current source, not a resistor. The key insight is that the two transistors must be well matched and at the same temperature, which is why mirrors shine on a single silicon chip where the transistors are made together and sit microns apart.
Where it shows up: as the tail current source that biases a differential pair, and as an 'active load' that replaces a plain collector resistor with a current source to get enormous voltage gain (a current source has very high output resistance, and gain is gm times that resistance). It is one of the most-used building blocks in every op-amp and analog chip. The honest caveats: real mirrors are not perfect — finite output resistance (the Early effect) lets the copied current drift a little with output voltage, base currents steal a small error, and mismatch between the transistors shows up directly as a current error. Improved versions (Wilson and cascode mirrors) exist precisely to fix these.
Inside an op-amp, one reference current set by a single resistor is mirrored to half a dozen places to bias every stage, all tracking together over temperature. Try to build the same with discrete transistors on a breadboard and the mismatch between two random 2N3904s will give you a sloppy copy — mirrors really want to be on one chip.
Set one current; the mirror reproduces it elsewhere as a stiff current source.
A current mirror is only as good as the matching and temperature tracking of its transistors. On a breadboard with separate parts it is mediocre; on monolithic silicon it is excellent — context decides whether to trust it.