Active Components

Current Mirror (RF)

Q: How does a current mirror set the bias of an RF power amplifier?

A reference transistor is diode-connected and fed a known reference current, typically 0.5 to 5 mA, derived from a precision resistor against a regulated supply or a bandgap reference. Because the reference and output devices share the same gate-source (or base-emitter) voltage, the output device conducts a scaled copy of that reference. In a GaAs or GaN MMIC amplifier, the mirror output drives the gate bias network so the final stage runs at a defined quiescent current, for example 80 to 150 mA per millimeter of gate periphery for a Class AB stage. Because both devices sit on the same die and track temperature together, the quiescent point drifts far less than a fixed gate-voltage scheme, holding the bias to within a few percent across a minus 40 to plus 85 degrees C operating range.

Q: What sets the mirror ratio and how accurate is the copied current?

The ideal mirror ratio is the device size ratio, the emitter-area ratio for bipolar or the gate-width ratio W2/W1 for FETs, so a 1-to-10 mirror copies ten times the reference current. Real accuracy is limited by three effects. First, finite output resistance from channel-length modulation (the Early effect or lambda parameter) makes the copied current rise with output voltage, typically 1 to 5 percent per volt for a simple mirror. Second, base-current loss in bipolar mirrors introduces a 2/beta error that a beta-helper or Wilson topology corrects. Third, device mismatch and threshold-voltage offset add a few percent. A plain mirror holds about 2 to 5 percent matching; a cascode mirror raises output resistance by the device gain (roughly g_m times r_o) and improves matching to better than 1 percent at the cost of voltage headroom.

Q: Why use a cascode current mirror instead of a simple two-transistor mirror?

A simple mirror has output resistance equal to the output device r_o, so its copied current changes noticeably as the output voltage swings, which degrades supply rejection and bias stability in an RF stage. Adding a cascode device stacks a second transistor on the output, multiplying the output resistance to approximately g_m times r_o squared, often raising it from tens of kilohms to several megohms. This greatly reduces current variation with output voltage and improves power-supply rejection by 20 to 40 dB. The trade-off is reduced voltage headroom, since the cascode consumes an extra V_DS,sat or V_CE,sat of roughly 0.2 to 0.5 V, which can matter on a 3.3 V or 5 V MMIC supply. Wide-swing cascode variants recover most of that headroom.

/KUR-uhnt MIR-er/

Built from two or more matched transistors that share the same control voltage, this active bias circuit copies a precisely set reference current into one or more output devices, so the output current becomes a fixed multiple of the reference set only by device size ratio. In RF and MMIC design it is the workhorse for establishing repeatable, temperature-tracking quiescent current in amplifier stages, replacing a brittle fixed gate-voltage scheme with one that follows process and thermal variation. Reference currents typically run 0.5 to 5 mA, output devices are scaled 1:5 to 1:50, and copy accuracy ranges from a few percent for a simple mirror to better than 1 percent for a cascode version.

Category: Active Components

Typical I_ref: 0.5 to 5 mA

Copy Accuracy: < 1% (cascode)

How Current Mirrors Lock RF Bias to a Reference

The core idea is simple but powerful. A reference transistor is diode-connected, meaning its gate (or base) is tied to its drain (or collector), and forced to carry a known reference current generated from a precision resistor, a bandgap reference, or a current digital-to-analog converter. Because the output transistor shares the same gate-source voltage V_GS (or base-emitter voltage V_BE), it is driven to the same operating point and conducts a scaled replica of that reference. The scale factor, called the mirror ratio, is set by the ratio of device sizes: emitter areas in bipolar mirrors or gate widths W₂/W₁ in FET mirrors. This makes the output current depend on a geometric ratio rather than on absolute threshold voltage, which is exactly the property that makes the bias repeatable from die to die.

In a GaAs or GaN MMIC power amplifier the mirror output feeds the gate-bias network of the RF stage, holding the final transistor at a defined Class A or Class AB operating point, often 80 to 150 mA per millimeter of gate periphery. Because the reference and output devices sit on the same substrate and share thermal and process gradients, the quiescent current tracks together. A well-designed mirror holds bias to within a few percent over a minus 40 to plus 85 degrees C range, where a naive fixed-voltage bias could drift by 30 percent or more as threshold voltage shifts with temperature.

The limitations come from non-ideal device behavior. Channel-length modulation (the FET lambda parameter, or the bipolar Early voltage V_A) gives the output a finite resistance, so the copied current rises slightly as the output voltage increases. In bipolar mirrors, the base currents of both devices are stolen from the reference, creating a fractional 2/β error that beta-helper and Wilson topologies were invented to cancel. Device mismatch and gradient effects add further offset, which careful common-centroid layout reduces.

Governing Relationships

Ideal mirror ratio (FET):
I_out = I_ref × (W₂/L₂) / (W₁/L₁)

Output current with channel-length modulation:
I_out ≈ I_ref × (W₂/W₁) × (1 + λV_DS2)

Simple-mirror output resistance:
R_out = r_o ≈ 1 / (λ × I_out)

Cascode mirror output resistance:
R_out,casc ≈ g_m × r_o²

Bipolar base-current error:
I_out = I_ref / (1 + 2/β)

Where W/L = device aspect ratio, λ = channel-length modulation, V_DS2 = output drain-source voltage, r_o = output resistance, g_m = transconductance, β = bipolar current gain. Example: 1:10 GaAs FET mirror, I_ref = 2 mA, λ = 0.04 V⁻¹, V_DS2 = 3 V → I_out ≈ 22.4 mA.

Current Mirror Topology Comparison

Topology	Output Resistance	Copy Accuracy	Min. Headroom	Key Drawback	Typical Use
Simple (2-device)	r_o (10 to 100 kΩ)	2 to 5%	~V_DS,sat (0.2 V)	Low R_out, β error	Coarse MMIC bias
Beta-helper	r_o	1 to 3%	~V_BE + V_CE,sat	Bipolar only	Multi-output mirrors
Cascode	g_mr_o² (MΩ)	< 1%	~V_T + 2V_ov	Reduced headroom	Precision RF bias
Wilson	g_mr_o²	< 1%	~V_BE + V_CE,sat	One added pole	Cancels β error
Wide-swing cascode	g_mr_o²	< 1%	~V_DS,sat + V_ov (0.3 V)	Extra bias node	Low-voltage MMIC

Common Questions

Frequently Asked Questions

How does a current mirror set the bias of an RF power amplifier?

A diode-connected reference transistor is fed a known current (0.5 to 5 mA) from a precision resistor or bandgap reference. Because the output device shares the same V_GS or V_BE, it conducts a scaled copy that drives the RF stage gate bias, holding a defined quiescent point such as 80 to 150 mA per mm of gate periphery for Class AB. Since both devices share the die, the bias tracks temperature to within a few percent over minus 40 to plus 85 degrees C.

What sets the mirror ratio and how accurate is the copied current?

The ideal ratio is the device size ratio (emitter area for bipolar, gate width W₂/W₁ for FETs), so a 1:10 mirror copies ten times the reference. Accuracy is limited by finite output resistance from channel-length modulation (1 to 5% per volt), the bipolar 2/β base-current error, and device mismatch. A simple mirror holds 2 to 5% matching; a cascode reaches better than 1%.

Why use a cascode current mirror instead of a simple two-transistor mirror?

A simple mirror has output resistance of only r_o, so the copied current shifts as output voltage swings. Stacking a cascode device raises R_out to roughly g_mr_o², often from tens of kΩ to several MΩ, cutting current variation and improving supply rejection by 20 to 40 dB. The cost is an extra 0.2 to 0.5 V of headroom, which wide-swing cascode variants largely recover.

Related Terms

← Current Minimum Current Probe →