How Does a Current Mirror Work?

Detailed Explanation

A current mirror shows up inside almost every analog IC — op-amps, comparators, voltage references, ADC and DAC front ends — as the way the chip biases its own internal amplifier stages, but it is referenced elsewhere on this site only as a passing feature of matched transistor pairs (see BJT vs MOSFET) without an explanation of how the circuit itself works or why it's built this way. This page covers the mirror circuit directly: how it forces a copied current, the basic BJT and MOSFET implementations, and the two standard variants (Widlar and cascode) that fix its two main limitations.

The Basic Two-Transistor Mirror

The simplest current mirror uses two matched transistors of the same type, with their base-emitter (BJT) or gate-source (MOSFET) terminals tied together:

        +V
         │
        R_ref
         │
    ┌────┴────┬─────────── Output (mirrored current)
    │         │
   Q1        Q2
 (diode-    (output
 connected)  transistor)
    │         │
   GND       GND

Q1 is diode-connected — its base (or gate) is tied directly to its own collector (or drain) — which forces Q1 to operate at whatever base-emitter voltage produces the reference current set by R_ref: I_ref = (V+ − V_BE) / R_ref for a BJT mirror. Because Q2's base is tied to the same node as Q1's base, Q2 sees the identical base-emitter voltage. If Q1 and Q2 are matched devices (same geometry, same die, same temperature — true for transistors fabricated side-by-side on an IC, and approximated by using a matched transistor pair or dual package in a discrete design), the same base-emitter voltage produces the same collector current in both, so I_out ≈ I_ref regardless of what's actually happening in the output branch — within limits covered below.

Sizing Q2 larger or smaller than Q1 (or, in an IC, using a different number of parallel unit transistors) scales the mirrored current by that same ratio: a 3:1 area ratio produces I_out ≈ 3 × I_ref, letting one reference current be scaled up or down for different bias branches on the same chip without needing a separate resistor for each.

MOSFET Current Mirrors

A MOSFET mirror uses the same topology with the gate-source relationship in place of base-emitter: M1 is diode-connected (gate tied to drain), setting a gate-source voltage that satisfies the square-law relationship I_ref = k(V_GS − V_TH)², and M2, sharing that same V_GS, conducts the same current (scaled by the W/L ratio between the two devices). MOSFET mirrors draw no steady-state gate current, so — unlike a BJT mirror, where base current subtracted from the collector current introduces a small systematic mismatch between I_ref and I_out — a basic MOSFET mirror's output current depends only on the matching between the two devices' threshold voltage and W/L ratio, not on current gain.

Two Practical Limitations, and Their Fixes

Output impedance (why the mirrored current isn't perfectly constant). An ideal current source's output current is completely independent of the voltage across it. A real BJT mirror falls short because of the Early effect: increasing V_CE on the output transistor causes a small increase in its collector current even at a fixed V_BE (equivalently, a MOSFET mirror has the same limitation from channel-length modulation). The output transistor's finite output impedance (r_o) means the "constant" mirrored current actually drifts slightly as the output branch's own voltage changes with the load it's driving. The standard fix is a cascode current mirror — stacking a second transistor in series with the output device, which holds the output device's own collector-emitter (or drain-source) voltage nearly fixed regardless of what the output node itself does, multiplying the effective output impedance by roughly the added device's own gain. The trade-off is compliance voltage: a cascode mirror needs more headroom (typically two V_CE(sat) or V_DS(sat) drops instead of one) before it can maintain its improved output impedance, which matters in low-voltage designs.

Generating a small bias current from a practical reference resistor. A simple mirror scales output current by device area ratio, but scaling down to a very small microamp-level bias current this way would need an impractically large area ratio (or, equivalently, an impractically large reference resistor to set a very small I_ref directly). The Widlar current mirror solves this by adding a resistor in series with the output transistor's emitter (or source), which develops a small voltage drop proportional to I_out. Because that voltage subtracts from the output transistor's own V_BE, and V_BE sits inside BJT's exponential current-voltage relationship, even a modest emitter resistor produces a large reduction in output current relative to the reference — letting a microamp-level bias branch be generated from a reference current large enough to be set accurately by a normal-value resistor.

Where Current Mirrors Are Actually Used

• Biasing analog IC input stages. A differential pair (the input stage of essentially every op-amp and comparator) needs a stable "tail current" bias — a current mirror is the standard way that bias current is generated and distributed inside the chip. See what is an op-amp? for how the differential input stage uses this bias current.
• Active loads. Replacing a plain resistor load on an amplifier stage with a current mirror acting as a high-impedance current source load significantly increases the stage's voltage gain, because gain is proportional to the load impedance the stage works into.
• Distributing one reference current to multiple bias points. A single precision reference current (often itself derived from a bandgap reference — see what is a voltage reference IC?) can be mirrored, at different ratios, into several bias branches across a chip without needing a separate resistor and reference for each.
• Discrete constant-current applications. Outside of IC design, a discrete matched-transistor mirror (or a dual-transistor package sharing one thermal environment) is a simple way to build a constant-current LED driver or a bias current for a discrete amplifier stage without an op-amp-based active current source. See how do you design a constant-current LED driver circuit? for how this compares against linear and switching driver ICs at higher power.

Design Considerations

• Matching quality sets accuracy, not just device selection. A current mirror's accuracy depends on how well its two transistors actually match in practice — a monolithic IC current mirror matches to a fraction of a percent because both devices are fabricated side-by-side on the same die at the same temperature; a discrete design built from two separately-packaged transistors will not match nearly as well unless a matched pair (two transistors on one die in one package, sold specifically for this purpose) is used.
• Thermal tracking matters as much as initial matching. Even a well-matched pair mismatches if the two transistors run at different temperatures — a discrete mirror where the output transistor dissipates significantly more power than the reference-side device (because it's carrying a much larger mirrored current) will drift out of the initial match as it self-heats. Keep the two devices thermally coupled, and avoid large current-scaling ratios in discrete designs where thermal tracking is hard to guarantee.
• Check compliance voltage against the actual supply headroom available. Every mirror topology needs a minimum voltage across the output branch to keep the output transistor out of saturation (BJT) or the triode region (MOSFET) — a basic mirror needs only about one V_CE(sat)/V_DS(sat), while a cascode mirror needs roughly double that. In a low-voltage design (3.3 V rail or lower), this headroom requirement can rule out a cascode mirror or force a wide-swing cascode variant.
• A BJT mirror's base current introduces a small, correctable systematic error. Because base current is subtracted from collector current, a basic BJT mirror's I_out is slightly less than I_ref by an amount related to 1/hFE — usually negligible for signal-level bias currents, but worth accounting for in a precision reference-distribution design; a MOSFET mirror doesn't have this term since MOSFETs draw no steady-state gate current.

Common Mistakes

• Assuming a discrete two-transistor mirror will match as well as an IC's internal mirror. Two separately-purchased transistors from the same reel can have meaningfully different V_BE and hFE at the same current, even from the same manufacturing batch. For anything beyond a rough bias current, use a matched dual-transistor package or verify actual matching on the bench rather than assuming datasheet typical values apply to both devices equally.
• Ignoring compliance voltage and finding the mirror silently stops working as the supply drops. A mirror that works correctly at 5 V can fall out of its compliance range at 3.3 V if the output branch's own voltage drop leaves too little headroom for the output transistor — the mirrored current degrades gradually rather than failing obviously, which can look like an unrelated bug elsewhere in the circuit.
• Using a large current-scaling ratio in a discrete design and getting poor tracking. A 10:1 or greater area/current ratio between reference and output transistors amplifies any thermal or matching mismatch between the two devices proportionally — scaling ratios that large are routine inside a monolithic IC (where matching is excellent) but unreliable in a discrete design built from separate packaged parts.
• Forgetting that a diode-connected transistor's own V_BE (or V_GS) is temperature-dependent. A basic current mirror's reference current, and therefore its mirrored output, drifts with temperature because V_BE (or V_TH) drifts — typically a fraction of a percent per degree for a BJT mirror's I_ref. For a temperature-stable bias current, that reference must itself come from a temperature-compensated source (a bandgap reference) rather than a plain resistor-and-diode-connected-transistor branch.

Zeus Design's analog and mixed-signal engineering team designs precision bias and reference circuits — including discrete current sources where an off-the-shelf reference IC doesn't fit the application — as part of complete product electronics design.