Inside nearly every op-amp, comparator, and analog IC you've ever used, dozens of current mirrors are quietly copying a single reference current to bias differential pairs, output stages, and gain nodes. Understanding them unlocks how analog ICs actually work — and lets you build precision bias networks on a breadboard with matched transistor pairs.

The basic BJT mirror uses two matched transistors (Q1 and Q2) with their bases tied together and emitters grounded. Q1 is diode-connected (base shorted to collector), forcing it into the active region where its V_BE sets the shared base voltage. Since Q2 sees the same V_BE, it conducts the same collector current — assuming matched devices and equal temperatures.

The reference current is set by a resistor from V_CC to Q1's collector:

I_REF = (V_CC − V_BE) / R

For V_CC = 5 V, V_BE ≈ 0.65 V, R = 4.3 kΩ → I_REF ≈ 1 mA. Q2 then sources (or sinks) 1 mA into whatever load you connect to its collector, regardless of load voltage — as long as Q2 stays out of saturation (V_CE > ~0.2 V).

Real-world non-idealities matter:

• Base current error: Both transistors steal base current from I_REF, so I_OUT = I_REF × β/(β+2). For β=100, that's a 2% error. Fix with a Wilson or cascode mirror.
• Early effect: Q2's output current rises slightly as V_CE2 increases, giving finite output impedance (~50–200 kΩ). A cascode mirror boosts this to many megohms.
• Thermal mismatch: A 2 mV V_BE difference causes ~8% current error. Use a matched pair like the BCM847 dual NPN or the venerable CA3046 transistor array — same die, same temperature.

Practical example: Biasing a differential pair tail current. You want 200 µA flowing through the diff pair's emitters, independent of supply variation. A simple resistor wouldn't do — its current would shift with V_EE. A current mirror referenced to a stable voltage (a bandgap, say) holds the tail current rock-steady. This is exactly how the LM741's 19 µA tail current is set internally.

The Widlar variant adds a small emitter resistor to Q2, letting you generate microamp currents from a milliamp reference without using a 5 MΩ resistor: I_OUT × R_E = V_T × ln(I_REF/I_OUT). With I_REF=1 mA, R_E=5 kΩ → I_OUT ≈ 20 µA. Beautiful logarithmic compression in one resistor.

MOSFET mirrors work identically but with no base current error — perfect for CMOS ICs, though they suffer worse V_GS matching than BJTs.