How to Choose an Op Amp for a Real Circuit

Most op-amp selection starts in a parametric search tool, sorted by price, and stops at the first part whose bandwidth looks big enough. That works often enough to be a habit and fails in exactly the ways the previous nine lessons described.

This lesson is the opposite approach. Start from what the circuit has to do, turn each requirement into a number, and apply the numbers in an order that eliminates candidates as fast as possible.

Quick answer. Work in three passes. First apply the constraints that a part either meets or does not: supply voltage, input common-mode range, output swing and current, minimum stable gain, temperature range. Those eliminate without any trade. Then apply the performance requirements you can calculate: gain-bandwidth, slew rate, offset and drift, bias current, noise. Only then trade among what is left, on quiescent current, package and cost.

Why there is no best op amp

A ranking needs a weighting, and the weighting is a property of your circuit, not of the parts. A part that is excellent for a photodiode amplifier is wrong for a motor current sense, and the reason is not that one is better made.

That is not a disclaimer, it is the method. If you cannot say which parameter matters most in your circuit, no list of recommendations will help you, because you will not be able to tell a good recommendation from a bad one.

So the output of this lesson is not a part number. It is a shortlist you can defend, and a record of which requirement eliminated what.

Pass one: the constraints with no trade

These are binary. The part meets them or it is out, and no amount of being good elsewhere compensates.

• Supply voltage. Does it operate at your rail, at the low end of its

tolerance, and is your rail within its absolute maximum at the high end? This eliminates more candidates than anything else and takes ten seconds.

• Input common-mode range at your common-mode voltage, over temperature.

Work out where the inputs actually sit across the whole signal range, then check the specification at the temperature extremes rather than at 25 °C. Lesson 7 is this requirement in detail.

• Output swing at your load current. Not the headline figure. The figure

at the current your load actually draws, which lesson 8 shows can be several times worse.

• Output current. Including whatever the feedback network draws, which is

easy to forget in a low-impedance divider.

• Minimum stable gain. Decompensated parts give more bandwidth in exchange

for not being usable below a stated closed-loop gain. If your noise gain is below it, the part is not a candidate.

• Temperature range, qualification and package. Including whether you can

actually buy it in the quantity and timeframe you need.

Pass two: the numbers you can calculate

Each of these comes from an equation in an earlier lesson, applied to your circuit.

The discipline that makes this useful is allocating the error budget before looking at any parts. Decide how many converter codes of error each mechanism is allowed, then convert each allowance into a specification limit. Doing it the other way round, by finding a part and then checking whether its numbers seem acceptable, is how designs acquire one parameter that is ten times better than it needs to be and another that is marginal.

Working it through on our front end

Requirements, gathered from the previous lessons:

• Single 3.3 V supply.
• Sensor output 0 to 20 mV, referenced to ground, source resistance about

1 kΩ.

• Non-inverting gain of 119, noise gain 119.
• Output 0.02 V to 2.38 V into a 12-bit converter input, negligible DC load.
• Signal bandwidth 1 kHz.
• Operating range 25 °C to 85 °C, single-point calibration at 25 °C.
• Error budget: 10 codes for initial offset, 10 codes for drift, half a code

of noise.

Converting those into limits:

GBW ≥ 1 000 × 119 = 119 kHz
slew rate ≥ 2π × 1 000 × 2.38 = 0.015 V/µs
max offset = 10 × 806 µV / 119 = 67.7 µV
max drift = 67.7 µV / 60 °C = 1.13 µV/°C
bias current error = IB × 1 kΩ, so 10 nA gives only 10 µV
noise density ≤ (6.8 µV / 2) / √1 000 Hz ≈ 107 nV/√Hz

Now read which of those did any work.

Binding: the input common-mode range has to include ground, because the signal sits there. The output has to get within about 20 mV of ground for the same reason. Offset must be under 68 µV and drift under about 1.1 µV/°C, which together are the real constraint and point at a precision part rather than a general-purpose one.

Not binding, by a wide margin: 119 kHz of gain-bandwidth is undemanding. A slew rate of 0.015 V/µs is met by anything. 107 nV/√Hz of noise is relaxed. A bias current of 10 nA costs 10 µV, barely more than a code, because the source resistance is low. Four of the six calculations eliminated nothing.

That result is the point of doing the arithmetic. Without it, the obvious instinct is to buy bandwidth and low noise, neither of which this circuit needs, while possibly accepting an offset specification that quietly consumes the entire error budget.

Notice also how much the answer depends on the circuit rather than the amplifier. Swap the 1 kΩ sensor for the 1 MΩ one from lesson 6 and the bias current line jumps from 10 µV to 10 mV, moving from irrelevant to dominant, and the shortlist changes from precision bipolar parts to CMOS ones. Nothing about the amplifier changed. One resistor did.

The parameters people over-specify

• Bandwidth. More gain-bandwidth generally costs quiescent current, and a

faster amplifier is more willing to amplify things you did not want, so a circuit that needs 8 kHz does not benefit from a 50 MHz part.

• Low noise. Low voltage noise often comes with higher input bias

current, because it tends to mean a bipolar input stage running at a higher current. With a high-impedance source that is a bad trade.

• Precision. If a two-point calibration is already in the production

process, initial offset is largely free and the money belongs in the drift specification instead.

• Rail-to-rail everything. As lesson 8 showed, it is paid for somewhere,

and a stage with gain and a mid-supply signal needs neither end of it.

Two things the datasheet front page will not tell you

Power supply rejection falls with frequency. The headline figure is a DC or low-frequency number, and rejection degrades as frequency rises, which matters when the supply is a switching regulator whose ripple sits at hundreds of kilohertz. The amplifier's ability to ignore that ripple at the switching frequency is far worse than the front page suggests, and the remedy is usually at the supply rather than in the amplifier. Where that ripple and its harmonics come from is covered in the switch node lesson of the switching regulator course.

Typical is not maximum. Several of the parameters in this course are commonly given as typical values with no guaranteed limit. A design whose correctness depends on a typical figure is a design that works on most units.

Where this stops

This course has been about a single amplifier in a feedback loop. Several closely related problems are deliberately outside it: amplifying a small differential signal riding on a large common-mode voltage, which is the job of an instrumentation amplifier; budgeting noise and filtering across a complete signal chain; and driving a converter input so that it settles within one sample period, which is a different problem from everything here because the load is switched rather than static.

Each of those is a subject in its own right, and each assumes what this course has covered.

Sources and further reading

• Texas Instruments: Understanding Operational Amplifier Specifications (SLOA011) — application report, what each datasheet parameter means and under what conditions it is guaranteed.
• Texas Instruments: Trade-offs Between CMOS, JFET, and Bipolar Input Stage Technology (SBOA355) — application report, how input technology couples noise, offset and bias current into one choice.
• Microchip: AN722 Operational Amplifier Topologies and DC Specifications — application note, turning DC specifications into a circuit error budget.
• Microchip: AN723 Operational Amplifier AC Specifications and Applications — application note, the AC side of the same exercise.
• Texas Instruments: When to Use an Amplifier with Rail-to-rail Inputs, and What to Watch Out for (SSZTB56) — technical article, deciding whether a rail-to-rail input is a requirement or a habit.