Op-Amp Selection Guide: Key Parameters for Analog Design

The op-amp market spans from a 30-cent LM358 to a 15-euro precision zero-drift amplifier. The difference is not just the price tag. It is input offset voltage, noise density, GBW, slew rate, supply range and whether the output swings close enough to the rail. Pick the wrong one and your signal chain loses bits, clips prematurely or oscillates.

Input offset voltage and drift

Input offset voltage (Vos) is the differential DC voltage that appears at the output when both inputs are at the same potential. A general-purpose LM358 specifies 2 mV typical, 7 mV max at 25°C. That is fine for a 0-5 V signal with 8-bit ADC resolution (19.5 mV per LSB). It is a disaster for a 3.3 V, 16-bit system where 1 LSB equals 50 µV: the offset alone swamps the first 6-7 bits.

Drift is how Vos changes with temperature. The LM358 drifts roughly 7 µV/°C. A precision part like the OPA333 (zero-drift, chopper-stabilized) specifies 0.02 µV/°C typical. If your product spans -20°C to +60°C, the LM358 adds 560 µV of drift on top of the initial offset.

For thermocouple amplifiers (K-type: 41 µV/°C), a 10 µV offset is a 0.25°C error. A 2 mV offset reads 49°C instead of 25°C. Use an auto-zero or chopper op-amp or calibrate in firmware with a periodic shorted-input measurement.

Gain-bandwidth product

GBW sets the closed-loop bandwidth. An LM358 with 0.7 MHz GBW configured for a gain of 100 delivers only 7 kHz of bandwidth. If your signal is a 10 kHz sinusoidal current-sense waveform, you need at least 500 kHz GBW for gain of 20 (25 kHz bandwidth) plus margin.

A common mistake: designing for DC gain and forgetting that the op-amp's open-loop gain rolls off at 20 dB/decade. At 10 kHz, an LM358 has 37 dB of open-loop gain left. Configured for a closed-loop gain of 100 (40 dB), there is no loop gain left for linearity or distortion reduction. The output will be distorted even if the bandwidth number says it should work.

Rule of thumb: GBW should be at least 10× the required closed-loop bandwidth times the closed-loop gain. For a gain of 20 at 20 kHz, GBW ≥ 10 × 20 × 20 kHz = 4 MHz.

Slew rate

Slew rate limits how fast the output can change, independent of bandwidth. If you need a 5 Vpp sine wave at 100 kHz, the maximum slope is 2π × 100 kHz × 2.5 V = 1.57 V/µs. An LM358 with 0.3 V/µs slew rate cannot reproduce that waveform. It will produce a triangle wave instead. A TL072 (13 V/µs) or OPA1656 (24 V/µs) handles it easily.

For step-response applications (pulse detection, comparator replacement), slew rate matters more than GBW. A fast edge hitting a slow op-amp produces slew-induced distortion long before bandwidth becomes the limiting factor.

Rail-to-rail: input and output

"Rail-to-rail" is not one spec. Check input common-mode range and output swing separately.

Many op-amps marketed as rail-to-rail output still cannot reach the positive rail under load. The MCP6002 (RRIO, 1.8-6 V, 1 MHz) specifies output swing to within 25 mV of each rail at no load, but degrades to 100 mV with a 5 mA load. At 1.8 V supply with a 5 mA load, losing 100 mV from the top rail leaves only 1.6 V of usable swing. A 500 mVpp signal at mid-supply (0.65 V to 1.15 V) fits; a 1.5 Vpp signal does not.

Input common-mode range is trickier. Some op-amps (LM324, LM358) have an input range that extends to the negative rail but stops 1.5-2 V below the positive rail. At a 5 V supply, the valid input range is 0 to 3.5 V. A non-inverting buffer with a 4.2 V input from a voltage reference will clip or behave unpredictably.

Noise: voltage and current

Op-amp noise has two components: voltage noise (nV/√Hz) and current noise (pA/√Hz). Voltage noise dominates in low-impedance circuits. If your source impedance is 1 kΩ, a 10 nV/√Hz op-amp contributes 10 nV/√Hz of voltage noise. The 1 kΩ resistor adds its own Johnson noise (4 nV/√Hz at 25°C). Total: roughly 10.8 nV/√Hz RMS-summed.

In a photodiode transimpedance amplifier with 1 MΩ feedback and 10 pA/√Hz current noise, current noise dominates: 10 pA/√Hz × 1 MΩ = 10 nV/√Hz from current noise alone. At high impedance, a FET-input op-amp (fA/√Hz current noise) like the OPA656 becomes essential.

For a 10 kHz bandwidth, a 10 nV/√Hz noise density integrates to 1.0 µV RMS, roughly 6.6 µV peak-to-peak. At 16-bit resolution with 3.3 V reference (50 µV/LSB), noise eats about 0.13 LSB RMS. Acceptable. At 24-bit with a 2.5 V reference (0.15 µV/LSB), the same noise eats 44 LSBs. You need a much quieter op-amp (OPA2210: 2.2 nV/√Hz) or a smaller bandwidth.

FAQ

Q: When should I use a comparator instead of an op-amp?

Op-amps make poor comparators. They are internally compensated for closed-loop stability, which slows recovery from saturation. An open-loop op-amp takes microseconds to come out of saturation; a comparator like the LM393 does it in hundreds of nanoseconds. Use a comparator when switching speed matters and the input differential can be large.

Q: What is the difference between bipolar and CMOS input op-amps?

Bipolar input (LM358, NE5532) typically has lower voltage noise (3-5 nV/√Hz) but higher input bias current (tens to hundreds of nA). CMOS input (MCP6002, OPA320) has higher voltage noise (20-40 nV/√Hz) but negligible bias current (pA range). Use bipolar for low-impedance sources, CMOS for high-impedance sensors and photodiode amplifiers.

Q: Do I need an external compensation capacitor for my op-amp circuit?

Most op-amps are internally compensated for unity-gain stability. However, driving a capacitive load directly (e.g., a long cable or ADC input) can cause oscillation. Add a small series resistor (50-100 Ω) between the op-amp output and the capacitive load to isolate the phase shift.

Find the right op-amp for your signal chain in Novapart's semiconductor catalog with filters for GBW, slew rate, package and supply voltage. Need a precision amplifier not in stock? Request a quote with your target specs and we will source it.