The current-feedback op-amp architecture has emerged to become a dominant solution for many applications. Possessing a number of strengths, this amplifier architecture can be used in nearly any application that calls for an op amp. Current-feedback amplifiers do not have a fundamental gain bandwidth product limitation.

This is often shown by a very small loss in bandwidth as signal amplitudes increase. Since large signals can be accommodated with minimal distortion, these amplifiers have good linearity at very high frequencies.

While voltage-feedback amplifiers lose bandwidth with increasing gain, current-feedback amplifiers keep most of their bandwidth over a wide range of gains. Thus, it's accurate to say that current feedback amplifiers have no gain bandwidth product limitation.

Of course, current-feedback amplifiers are not infinitely fast either. The slew rate is limited not by the internal bias currents, but by the speed limitations of the transistors themselves. This allows larger slew rates for a given bias current without using positive feedback or other slew-boosting techniques that can often hamper stability.

So how do we build one of these wonder circuits?
Current feedback op amps have an input buffer as opposed to a differential pair. The input buffer is most often an emitter follower or something very similar. The non-inverting input is high impedance, while the buffer's output is low impedance. In contrast, both inputs of a voltage-feedback amplifier are high impedance.

The output of a current-feedback op amp is a voltage; it is related to the current that flows out of or into the inverting input of the op amp by a complex function called transimpedance, Z(s), which is a very high number at DC. Like a voltage-feedback op amp, it has a single pole roll off with increasing frequency (Figure 1, below).

Figure 1. Voltage is related to the current that flows out of or into the inverting input of the op amp by a complex function called transimpedance Z(s).

The current-feedback op amp's key flexibility is adjustable bandwidth and stability. Because the feedback resistor value actually changes the AC loop dynamics of the amplifier, it can impact both the bandwidth and stability.

Coupled with very high slew rates and adjustable bandwidth based on the feedback resistor, you can get large signal bandwidths that are very close to the small signal bandwidth of the device. Moreover, this bandwidth is largely preserved over a wide range of gains. And because of the inherent linearity, you can get low distortion with large signals at high frequencies too.

Finding the best RF
Having the AC characteristics of the amplifier partially dependent on the feedback resistor enables us to tailor the amplifier for each unique application. Lowering the value of the feedback resistor boosts the loop gain.

At low gains, the feedback resistor is set to a higher value to preserve stability and maximum bandwidth. As gain goes up, the loop gain is naturally reduced. This loop gain can be partially recovered by using a smaller feedback resistor when a high gain is required. Figure 2 below illustrates what happens to the bandwidth as you change the feedback resistor.

Figure 2. The RF = 300 curve has excellent flatness and gain, and still good bandwidth comparable to the peaked frequency response.

At the far right-hand curve where the RF = 147 ohm, the frequency response has peaked quite substantially. This curve also has the highest bandwidth. Decreasing the resistor too far below 147 ohms results in ringing on your pulse response, and it will actually oscillate.

The RF = 300 curve has excellent flatness and gain, and still good bandwidth comparable to the peaked frequency response. Thus, we have gained a great deal of stability without giving up much of the bandwidth.

With a feedback resistor of 600 ohms, you can scale back your frequency response. If an application only needs 50MHz or 60MHz of bandwidth and anything over that contributes to noise, you can tailor the frequency response of your part with the feedback resistor. The reason to use a fast amplifier with such a limited bandwidth is the excellent signal fidelity it provides.

Figure 3 below is from the datasheet of the same device and shows the suggested feedback resistor for a given non-inverting gain.

Figure 3. The recommendation for a gain of two is the 300 ohm resistor, with the best combination of gain flatness, settling time and speed.

As might be expected, the recommendation for a gain of two is the 300 ohm resistor, with the best combination of gain flatness, settling time and speed. Moreover, for a gain of one, you need a 600 ohm feedback resistor to get optimum performance. This is because the loop gain is very high, and a bigger resistor value is required for stability.

This is a key difference from the voltage-feedback architecture. A current-feedback amplifier cannot be used with the output shorted to the inverting input.

The most common resistor specified on datasheets is for the gain of two. However, there is a good deal of flexibility in the actual value you could end up using (Figure 2 above). The value recommended in a datasheet is the value used to generate the specifications published in the performance tables and curves. For a gain of five, RF is down to 200 ohms (Figure 3 above).

The gain set resistor is only 50 ohm now, so we are getting to the point where the input buffer resistance and gain set resistance are of similar values. This lowers the closed loop transimpedance of the op amp and will begin to limit the bandwidth as gain increases.

At a gain of eight, we are back up to a 275 ohms feedback resistor. Once the feedback resistor cannot be decreased for higher gain, the bandwidth is compromised, and the amplifier begins to behave like a voltage-feedback amplifier.

Board layout
One of the things to carefully consider with current-feedback op amps (or high-speed parts, in general) is the board layout. The surface-mount, ceramic supply bypassing capacitors need to be very close to the device, typically less than 3 millimeters.

If needed, the larger electrolytic capacitor can be placed farther on the board. Often, there is an on-board voltage regulator. In this case, there is no need for additional electrolytic capacitors beyond those recommended by the voltage-regulator vendor.

The small ceramic bypass capacitors that are placed near the amplifier provide energy for the high-frequency response of the amplifier. Depending on the amplifier's speed and the signal being amplified, it may be desirable to use two ceramic capacitors with values about a decade apart. For example, a 400MHz amplifier could use 0.01 microfarad and 1 nanofarad capacitors installed in parallel.

When purchasing capacitors, it is important to check the self-resonant frequency. The capacitors provide no benefit at frequencies near or above this frequency.

Ground and power planes help provide low impedance paths for both the ground and the power currents. Both the ground and power planes should be removed from underneath the I/O pins of the amplifier and the feedback resistor. This helps maintain amplifier stability by reducing undesired parasitic capacitance.

Always try to use surface mount components where possible. These provide the best performance and take up minimal board space as well. PCB traces should be kept as short as possible and should be sized to minimize parasitic effects.

On power-supply traces, the worst parasitic characteristic is DC resistance and inductance. Thus, power traces should be as wide as possible. On the other hand, I/O traces often carry very little current, so capacitive parasitics are the most harmful. For signal paths of over 1 centimeter, it is best to use controlled impedance and doubly terminated transmission lines.

Since small amounts of parasitic loading are unavoidable, the feedback resistor of a current- feedback amplifier provides flexibility for tuning amplifier performance for the specific application.

Figure 4. By including a resistor (Rout), nearly any amount of capacitance on the output of an amplifier can be driven without stability problems.

With a really challenging board layout, though, even a very large feedback resistor may not be sufficient. In this case, another option is available. By including a resistor (Rout), nearly any amount of capacitance on the output of an amplifier can be driven without stability problems (Figure 4, above).

This is a common technique with op amps that work on both voltage- and current-feedback amplifiers. This technique is especially useful when driving a high-speed ADC. The Rout resistor is placed between the op amp and the capacitive load (the ADC). The resistor should be as close to the amplifier as board space allows. The curve in Figure 5, below shows the suggested resistor values of Rout based on the size of the capacitor.

Figure 5. If RL is smaller in value, Rout can be smaller too.

It is based on a 1k-ohm resistive load. If RL is smaller in value, Rout can be smaller too. Another option is to put the Rout inside the feedback loop. Instead of connecting RF between Rout and the amplifier, you can connect RF to the output side of the isolation resistor.

This will preserve gain accuracy, but you will still lose the same amount of voltage swing in the isolation resistor as in the other example. It should be realized, however, that this technique has a drawback. Because the resistor and capacitor form a low pass filter, there is some bandwidth penalty for using this circuit.

Figure 6 below shows the response for the LMH6738. Regardless of the resistor value, higher capacitance is harder to drive and reduces bandwidth accordingly.

Figure 6. Regardless of the resistor value, higher capacitance is harder to drive and reduces bandwidth accordingly.

Reducing system noise
Minimizing noise is particularly important if you are building an IF amplifier or a low-frequency RF amplifier. With current-feedback amplifiers (although it may seem counter-intuitive), increasing the feedback resistor can often reduce system noise. This is because the frequency response falls off faster than the resistor noise goes up.

To reduce noise to the portions of the circuit following the amplifier, it is very important to have only the bandwidth necessary and no more. Besides using the best value of feedback resistor, you can add additional filtering to the circuit.

The filter can often be incorporated right into the amplifier feedback network using a Sallen-Key filter topology. If possible, AC coupling will help remove low-frequency noise that is often called 1/F noise. The goal is to cut out all the noise except in the band that you are amplifying. System-level considerations call for placing the lowest noise and highest gain blocks early in the circuit.

The earlier you add gain, the less noise impacts your signal later on. If possible, avoid large source resistances. Resistances add thermal noise proportional to the value of the resistor.

When you compare a current feedback op amp with a voltage feedback op amp, there are a few areas where the voltage-feedback amplifier may offer an advantage.

With the current-feedback topology, the input bias currents are not systematically matched. The non-inverting input—being a higher impedance input than the inverting input—typically has a lower input bias current.

The inverting input bias current will usually be larger, which can lead to input voltage offsets if the bias currents have to flow through large valued resistors.

The offset voltage on a current feedback part can be matched and made quite small, but they are not systematically zero. Thus, while the typical offset voltage of a current-feedback op amp can be made very good, it will tend to vary more with normal process lot variations and with temperature variations. If very high precision is required for the input offset voltage, a voltage-feedback amplifier is usually a better choice.

The buffer configuration of a current-feedback amplifier needs a feedback resistor, while a voltage- feedback amplifier can use a direct short. This is not normally a problem unless replacing an existing voltage-feedback amplifier in an existing design.

And finally, capacitance in the feedback loop of a current-feedback amplifier creates instability. Some common circuit topologies will not accommodate a current feedback amplifier. For most of these circuits, there are substitute layouts that will work for a current feedback amplifier.

Applications that benefit from the unique qualities of a current- feedback amplifier include: presentation-quality video line drivers and routers, ADC drivers, IF amplifiers and clock buffers. Current-feedback amplifiers shine anywhere signal fidelity and high speed are the primary goals.

Loren Siebert is an Applications Engineer at National Semiconductor Corp.