Current-Feedback Amplifiers

Introduction

The current-feedback amplifier, commonly abbreviated CFB or CFA, is a high-speed operational amplifier topology that senses feedback as a current rather than as a voltage. To the user it presents the familiar package of a noninverting input, an inverting input, and an output, and it is applied with the same resistor networks as a conventional op-amp. Internally, however, its architecture and its governing equations differ fundamentally from the classic voltage-feedback amplifier, and those differences give it two prized properties: a closed-loop bandwidth that is largely independent of the chosen gain, and a slew rate high enough that it does not impose the usual hard ceiling on large-signal speed.

These advantages explain why current-feedback amplifiers populate the fastest analog signal paths. Video distribution, high-resolution displays, wideband data acquisition, pulse and transimpedance amplification, and the buffering of fast digital-to-analog converters all draw on the CFB's ability to combine high gain with wide bandwidth and to settle quickly after large transitions. The topology is closely related to the second-generation current conveyor, and indeed a current-feedback amplifier can be viewed as a positive current conveyor followed by an output buffer, a kinship that illuminates why it behaves as it does.

This article contrasts the current-feedback topology with voltage feedback, explains why its inverting input acts as a current-sensing node, develops the gain-bandwidth independence and slew-rate behavior that distinguish it, sets out the constraints that the feedback resistor must satisfy, and surveys the applications for which it is chosen.

The Topology Compared with Voltage Feedback

The two amplifier families differ in the nature of their inputs and in the quantity that feedback controls. A voltage-feedback amplifier has two high-impedance inputs and amplifies the voltage difference between them by a very large open-loop voltage gain. Feedback drives that voltage difference toward zero, and the error signal the amplifier acts upon is a voltage. Its inputs are symmetric: the inverting and noninverting terminals are electrically alike, both presenting high impedance.

A current-feedback amplifier breaks this symmetry. Its noninverting input is a high-impedance voltage input, but its inverting input is a low-impedance terminal connected to the output of an input buffer. That buffer forces the inverting-input voltage to follow the noninverting-input voltage. The amplifier does not respond to a voltage difference between its inputs; it responds to the current that flows into or out of the low-impedance inverting node. This error current is sensed and converted to an output voltage by a large transimpedance, a gain expressed in ohms because it relates an output voltage to an input current. Feedback acts to drive the error current, not an error voltage, toward zero.

The Internal Structure

The canonical current-feedback amplifier comprises three stages. A unity-gain buffer connects the noninverting input to the inverting input, giving the latter its low impedance and forcing the two input voltages to be nearly equal. Current mirrors sense the current delivered by that buffer into the inverting node and convey it to a high-impedance internal node. The voltage developed at that high-impedance node, equal to the error current multiplied by the transimpedance, is then carried to the output by a second unity-gain buffer. The buffer-mirror-buffer sequence is exactly the structure of a current conveyor with an added output buffer, which is why the two devices share so much behavior.

The transimpedance gain is deliberately made very large, just as the open-loop voltage gain of a conventional op-amp is made very large. A small error current therefore suffices to swing the output across its full range, and negative feedback reduces the required error current to a tiny value in normal operation, much as voltage feedback reduces the required input voltage difference in a conventional amplifier.

The Inverting Input as a Current Node

The defining feature of the current-feedback amplifier is that its inverting input is a current-summing node of low impedance. Because the input buffer holds this node at the voltage of the noninverting input, it behaves much like a virtual short to that potential, but it does so by sourcing or sinking whatever current the external network demands rather than by relying on a high open-loop voltage gain to enforce the equality.

In a standard noninverting configuration, a gain-setting resistor runs from the inverting node to ground and a feedback resistor runs from the inverting node to the output. Currents through these two resistors meet at the inverting node, and the amplifier supplies the small difference between them through its input buffer. That difference is the error current the amplifier senses. The action of the loop is to adjust the output so that the error current is driven nearly to zero, which forces the currents in the two resistors into the balance that defines the closed-loop gain. The closed-loop voltage gain that results is the same ratio of resistors as in a voltage-feedback amplifier, namely one plus the ratio of the feedback resistor to the gain-setting resistor for the noninverting case, so the two families look identical from the outside even though the internal error variable is a current in one and a voltage in the other.

This current-sensing character has a direct practical consequence. Because the inverting input is a low-impedance node fed by a buffer that can deliver substantial current, the amplifier can charge and discharge the capacitances in the feedback network rapidly, without being throttled by a fixed internal bias current. This is the seed of the high slew rate examined below. It also means that the two inputs are not interchangeable, so the standard practice of swapping inputs to invert a circuit must be applied with care, and the bias currents at the two inputs differ in origin and magnitude.

Gain-Bandwidth Independence

The most celebrated property of the current-feedback amplifier is that its closed-loop bandwidth depends primarily on the feedback resistor rather than on the closed-loop gain. This stands in sharp contrast to the voltage-feedback amplifier, whose gain-bandwidth product is essentially constant, so that doubling the gain halves the bandwidth.

The reason lies in where the gain-setting resistor appears in the loop. In a current-feedback amplifier, the loop gain is set by the transimpedance of the amplifier acting against the feedback resistance that converts the output voltage back into a feedback current at the inverting node. To first order, the loop gain and hence the closed-loop bandwidth are governed by the feedback resistor and the amplifier's internal transimpedance and node capacitance, but not by the gain-setting resistor. Changing the gain-setting resistor alters the closed-loop voltage gain without strongly altering the loop gain, so the bandwidth stays nearly the same as the gain is changed.

The designer exploits this by treating the feedback resistor as the element that sets bandwidth and stability, and the gain-setting resistor as the element that sets gain, two knobs that are far more nearly independent than in a voltage-feedback design. A current-feedback amplifier can therefore provide substantial gain while preserving a wide bandwidth that a voltage-feedback part could not sustain at the same gain. The independence is not perfect, since parasitics and the finite output impedance of the input buffer introduce a residual dependence, but over a useful range the bandwidth holds remarkably steady as gain is varied.

Slew Rate and Large-Signal Speed

A second decisive advantage of the current-feedback topology is its high slew rate. Slew rate is the maximum rate at which the output voltage can change, and in a conventional voltage-feedback amplifier it is limited because the input differential pair can deliver only a fixed maximum current, set by its tail bias, to charge the internal compensation capacitance. When a large step is applied, the input stage saturates at that fixed current and the output ramps no faster, regardless of how large the input becomes.

The current-feedback amplifier escapes this bottleneck. When a large transient appears at the input, the imbalance of currents at the low-impedance inverting node is large, and the input buffer delivers a correspondingly large error current to the internal high-impedance node. The current that charges the internal capacitance is therefore not clamped to a fixed bias value but rises with the size of the transient. The amplifier slews faster when it needs to, a behavior sometimes described as a dynamically demanded charging current. The result is a slew rate that can reach hundreds or thousands of volts per microsecond, far beyond what comparable voltage-feedback parts achieve, and a large-signal bandwidth that approaches the small-signal bandwidth rather than falling well short of it.

One practical consequence is that the slew rate of a current-feedback amplifier, unlike the fixed slew rate of a voltage-feedback part, also depends on the feedback resistor, because that resistor governs how much error current the inverting node delivers for a given output excursion. Reducing the feedback resistance therefore tends to raise both bandwidth and slew rate together, which is one reason the manufacturer's recommended value, discussed below, is the prudent starting point.

Because the slew rate is so high, the current-feedback amplifier largely avoids slew-rate-limited distortion and the associated settling delays that compromise voltage-feedback amplifiers handling fast, large signals. This makes the topology especially valuable for pulse amplification, for driving the fast edges of video and display signals, and for buffering high-speed converters, where rapid, clean settling after a large transition is essential.

Feedback-Resistor Constraints

The freedoms of the current-feedback amplifier come with a discipline that has no exact counterpart in voltage-feedback design: the feedback resistor cannot be chosen freely, and in particular it must not be replaced by a direct connection. Understanding the constraints on this resistor is the key to using the topology successfully.

Why a Direct Connection Is Forbidden

Because the feedback resistor sets the loop gain and participates in the dominant pole that establishes stability, shorting the output to the inverting input, as one would to make a voltage-feedback unity-gain buffer, removes the element that controls the loop and almost always provokes oscillation. A current-feedback amplifier configured as a follower therefore still requires a feedback resistor of the recommended value between the output and the inverting input, even though that resistor carries no signal current in the unity-gain case. The manufacturer specifies an optimum feedback resistance, and the recommended value, rather than a wire, must be used for the follower.

Choosing the Value

Lowering the feedback resistance increases the loop gain and widens the bandwidth, but if it is made too small the loop gain becomes large enough, in combination with internal and parasitic phase shift, to cause peaking in the frequency response and ultimately oscillation. Raising the feedback resistance reduces the bandwidth and improves stability margin but eventually slows the amplifier unnecessarily. The data sheet's recommended value represents the manufacturer's balance of these effects for a stated supply and gain, and it is the sound starting point. Adjusting bandwidth by changing the feedback resistor, and adjusting gain by changing the gain-setting resistor, is the natural design procedure.

Capacitance and Layout

Stray capacitance at the inverting node and across the feedback resistor is particularly harmful in current-feedback amplifiers, because that node is the sensitive current-summing point. Excess capacitance there introduces phase shift that erodes stability margin and produces peaking or ringing. For this reason a capacitor must not be placed directly in the feedback path as one might in a voltage-feedback integrator, the inverting-node trace must be kept short and its area small, and the ground plane is often relieved beneath that node to minimize stray capacitance. Disciplined high-frequency layout is not optional with these devices; it is part of meeting the stability constraints.

Applications

The current-feedback amplifier is selected wherever wide bandwidth, high slew rate, and fast settling are paramount and where its few limitations can be tolerated. Several application areas are characteristic.

Video and Display Drivers

Video signal paths demand flat gain across a wide bandwidth and clean reproduction of fast edges with minimal differential gain and phase error. The current-feedback amplifier's wide, gain-independent bandwidth and high slew rate suit it to video line drivers, distribution amplifiers, and the buffering of display and graphics signals, where it can drive cables and loads with the necessary speed and fidelity.

High-Speed Data Acquisition and Converter Buffers

Driving and buffering fast analog-to-digital converters, and buffering the outputs of fast digital-to-analog converters, requires an amplifier that settles quickly and accurately after large steps. The high slew rate and rapid settling of the current-feedback amplifier make it a natural converter driver, preserving the speed of the data-conversion path rather than limiting it.

Pulse and Transimpedance Amplification

Pulse amplifiers, used in instrumentation, radar, time-of-flight measurement, and optical receivers, must reproduce sharp edges with minimal distortion, a task to which the high slew rate is ideally matched. Because the inverting input is a current node, the topology is also well suited to transimpedance amplification, the conversion of a photodiode or sensor current into a voltage, where the low-impedance summing node and wide bandwidth combine to give fast, low-distortion current-to-voltage conversion.

Wideband Gain Blocks and Active Filters

Wherever a fixed gain must be delivered over a wide bandwidth, the current-feedback amplifier offers gain without the bandwidth penalty of a voltage-feedback part. It serves as a general wideband gain stage and can implement high-frequency active filters, with the caveat that the prohibition on capacitance in the direct feedback path constrains the filter topologies that may be used and steers the designer toward arrangements that keep the inverting node clean.

Limitations and Trade-Offs

The current-feedback amplifier is not a universal replacement for the voltage-feedback op-amp, and its strengths are paid for in other coin. Its direct-current precision is generally inferior: the input offset voltage, the input bias currents at its two dissimilar inputs, and the resulting offset and drift are typically larger than those of precision voltage-feedback amplifiers, so it is a poor choice for high-accuracy, low-frequency measurement. Its inputs are not symmetric, which complicates fully differential and certain inverting arrangements and forbids the casual interchange of inputs. The mandatory, value-constrained feedback resistor removes the freedom to short the feedback path or to place capacitance in it, ruling out the simplest unity-gain buffer wiring and the textbook integrator. And its sensitivity to stray capacitance at the inverting node demands careful layout. Where wideband, high-slew-rate performance is the goal, these are acceptable costs; where dc precision dominates, the voltage-feedback amplifier remains the better instrument.

Summary

The current-feedback amplifier is a high-speed operational amplifier whose inverting input is a low-impedance, current-sensing node driven by an internal input buffer, in contrast with the symmetric high-impedance inputs and voltage-difference sensing of the voltage-feedback amplifier. Internally it is a buffer, a set of current mirrors, and an output buffer, an arrangement equivalent to a positive current conveyor with an output stage, and it senses an error current that a large transimpedance converts into the output voltage.

From this architecture flow its hallmark advantages. The closed-loop bandwidth is set chiefly by the feedback resistor and is largely independent of the closed-loop gain, so high gain and wide bandwidth coexist. The slew rate is very high because the charging current at the internal node grows with the size of the transient rather than being clamped to a fixed bias, which yields fast, clean settling of large signals. These benefits are governed by a strict discipline on the feedback resistor, which must take a value near the manufacturer's recommendation, must never be replaced by a direct connection even for unity gain, and must be kept free of stray and deliberate capacitance at the sensitive inverting node. With those constraints respected, the current-feedback amplifier excels in video and display driving, high-speed data conversion, pulse and transimpedance amplification, and wideband gain, while the voltage-feedback amplifier remains preferable wherever direct-current precision is the overriding requirement.