1. Why the LM358 is everywhere

The LM358 is a dual operational amplifier that works from a single supply, costs cents, and tolerates inputs down to the negative rail. That combination made it the default amplifier for sensor conditioning, current sensing and general signal work in cost-sensitive products.

Its limitations are equally well known: a 1 MHz gain-bandwidth, a slow slew rate, a few millivolts of offset, and an output that stops about 1.5 V below the positive rail. None of that matters if the design was written around those limits.

Substitution problems start when a designer assumes "dual op-amp, 8 pins" means interchangeable. In this category, the input and output ranges matter more than the pinout, and the differences are invisible on a schematic.

2. What the part actually does

The input stage is a PNP pair, which allows the common-mode range to extend to the negative supply — often below it by a few hundred millivolts. That is why the LM358 is popular for low-side current sensing: the inputs sit near ground, and the amplifier still works.

The output stage is a class-AB emitter follower. It can sink and source modest current, but it cannot pull the output closer than about 1.5 V to the positive rail, and it has a crossover region near ground that adds distortion.

That crossover region is why audio designs avoid the part and why precision DC designs sometimes see a small nonlinearity at low output levels. It is a real characteristic, not a defect.

One practical detail: the LM358 is specified for single-supply operation, but it also works on split supplies. Designs that move from ±5 V to a single 5 V rail often keep the same part and discover that the input range no longer covers the signal, because the negative rail is now ground and the signal swings below it.

3. The family and its neighbours

The LM358B is the most useful drop-in: same pinout, improved offset and bandwidth, and the same input and output limitations, so the circuit around it does not need to change. The LV and MCP6002 parts are rail-to-rail but are different silicon with different trade-offs.

Input bias current is a fifth number worth checking for high-impedance sources. The LM358's bipolar input draws nanoamps, which is small but not zero; across a 1 MΩ source resistance it becomes millivolts of offset. A CMOS part with picoamp bias current removes that error, which is why sensor front-ends often move to CMOS.

The four numbers that decide a substitution

Most of the datasheet can be ignored for a swap decision. Four numbers carry almost all of the risk.

Input common-mode range. Expressed relative to the supplies, often as "V− to V+ − 1.5 V". If the circuit operates near the positive rail, a part that stops short will saturate the input stage.

Output voltage swing. Specified at a load current. A part that swings to within 50 mV of the rails at 1 mA may only reach 500 mV at 20 mA.

Gain-bandwidth product. Sets the closed-loop bandwidth for a given gain. Doubling it changes stability margins in an existing layout.

Quiescent current per amplifier. Bipolar parts draw milliamps; CMOS parts draw microamps. In a battery product this can dominate the budget.

If those four match, the substitution is usually safe. If any differs, the circuit needs a review before the part is ordered.

4. What to check before substituting

The first two rows are the ones that cause failures. A rail-to-rail alternative sounds strictly better, but its input stage may not work at the negative rail, which is exactly where a low-side current-sense design sits.

Single-supply biasing and virtual ground

An op-amp on a single supply cannot amplify a bipolar signal without a reference. Designs create a virtual ground at half the supply with a divider and a capacitor, then reference the signal to it.

The LM358's input range helps here: the divider can sit near ground, and the amplifier still works. Its output range hurts: the amplified signal must stay below V+ minus about 1.5 V, which limits the usable swing.

A rail-to-rail part removes that ceiling but changes the bias point behaviour, because its input stage may not tolerate the negative rail. Swapping parts without recalculating the bias network is a common source of "it works but the readings are off".

5. The low-side current-sense case

A typical circuit puts a shunt resistor between the load and ground, and amplifies the small voltage across it. The amplifier input sits within millivolts of ground, and the output drives an ADC.

The LM358 handles the input side well: its PNP input stage tolerates the negative rail. The output side is the problem: at low currents the output is near ground, in the crossover region, where gain is reduced and distortion appears. The result is a nonlinear reading at the bottom of the range.

Two fixes exist. Add a small offset so the output never sits at the crossover point, or use a CMOS rail-to-rail part with a linear output near ground. The first keeps the LM358 and costs a resistor; the second changes the part.

Comparator misuse

The LM358 is occasionally used as a comparator because it is cheap and available. It works, with caveats: no hysteresis unless added externally, slow response, and a phase inversion risk when the input goes below the negative rail.

For a slow threshold with a clean signal, it is acceptable. For anything fast or noisy, a dedicated comparator with hysteresis and push-pull output is the correct part, and it costs little more.

The substitution angle is important: replacing an op-amp with a comparator in a socket is not a drop-in. Pinouts differ, output stages differ, and the circuit around it changes.

Package variants are not interchangeable in the other direction: a DIP-8 part will not fit a SOIC-8 land pattern, and the dual-quad difference (LM324) changes the pin count. Check the package before the electricals, because a part that does not fit never reaches the bench.

6. Rail-to-rail is not one thing

"Rail-to-rail" describes the input, the output or both, and vendors are not always precise. Three combinations matter.

Input to the negative rail only. The LM358 class. Good for low-side sensing, poor for high-side sensing near V+.

Rail-to-rail output. The output can approach both rails, but the input range may still stop short of one of them.

True rail-to-rail I/O. Both sides reach the rails, usually with a CMOS input stage. The trade-off is higher noise and, sometimes, a lower supply limit.

Read the conditions attached to the rail-to-rail claim: it is often specified at a particular supply voltage and load current, not across the whole range.

Slew rate and large-signal bandwidth

The datasheet's gain-bandwidth figure applies to small signals. A 1 MHz part asked to swing several volts at 100 kHz will be limited by slew rate long before bandwidth: the output simply cannot move fast enough.

The LM358's slew rate is modest, which makes it fine for DC and low-frequency work and unsuitable for fast control loops or video. A rail-to-rail CMOS part often has a higher slew rate at the same bandwidth, which is one reason it can be a better choice even when the bandwidth matches.

When substituting, compare slew rate at the actual load capacitance. A faster amplifier driving a large capacitive load can oscillate, so the trade-off is not one-directional.

Driving an ADC or a cable

An op-amp that drives an ADC sees a switched sampling capacitor; one that drives a cable sees capacitance and possible radio-frequency pickup. Both loads can destabilise a part that was stable on the bench with a scope probe.

The standard fixes are a small series resistor at the output and a capacitor to ground, forming a low-pass filter that isolates the amplifier from the load. The values depend on the ADC's sampling time and the required bandwidth.

If the substituted part has a different output impedance or a different phase margin, the same resistor and capacitor may no longer be enough. Re-check the step response with the real load, not a 10 MΩ probe.

7. Test plan

When to leave the LM358 family

Three applications justify a different part rather than a substitute.

Precision DC. Offset in the millivolt range is too large for a shunt measurement without calibration. A chopper-stabilised amplifier is the right tool.

Audio. The crossover distortion of a class-AB output is audible. An audio op-amp with a different output stage is required.

High-side sensing. The input range does not reach the positive rail. A dedicated current-sense amplifier with a matched input divider is the correct choice.

In all three cases the answer is a different circuit, not a pin-compatible part. Recognising that early saves a qualification cycle.

Second-source quality in this family varies in offset voltage and input bias current, not in function. Two parts can be pin-identical and differ by a factor of three in offset, which matters for a DC measurement and is invisible in an AC test.

Measure offset on a sample from each lot if the circuit depends on it, and keep the measurement setup simple: a unity-gain buffer with the input grounded will show the offset directly.

Output current has a limit that is easy to forget. The LM358 can sink and source tens of milliamps, but the output stage heats and the swing collapses at higher currents. Driving an LED, a relay or a long cable directly is usually a mistake; add a transistor or a buffer.

Decoupling matters as much as in any analog part: 100 nF close to the supply pins, and a ground plane under the package. An op-amp that oscillates at 100 MHz is usually a layout problem, not a part problem, and changing the part rarely fixes it in the field.

8. Cost and supply

The bipolar part is one of the cheapest amplifiers available, with many sources. The improved LM358B costs little more and removes some of the original's weaker specifications; for new designs it is the better default.

CMOS rail-to-rail parts cost more and draw far less current. For battery products they are usually worth it; for mains-powered products the difference rarely matters.

Buy from a traceable channel. This is a high-volume part with a long history of remarked devices, and an op-amp with the wrong offset is difficult to detect in a functional test, and a lot number is the only practical way to isolate a bad batch.

A worked design

A 12 V single-supply system needs to amplify a 0–100 mV shunt signal to 0–3 V for an ADC. Gain is 30, so a 1 MHz part gives about 33 kHz of closed-loop bandwidth — ample for a current loop.

The input sits near ground, which the LM358 handles. The output must reach 3 V, well below V+ minus 1.5 V, so the swing is fine. At 0 mA the output sits at the crossover region, which is the one weakness; a 10 mV offset at the input keeps the amplifier out of it.

Replacing the LM358 with a rail-to-rail part would allow a larger output swing, but the design does not need it. Replacing it with a CMOS part would remove the bias current error, which matters only if the shunt network is high impedance. Neither change is required — and that is the point: substitution should follow the requirement, not the datasheet's best number.

9. FAQ

Is LM358 rail-to-rail?

No. Its input reaches the negative rail, but its output stops about 1.5 V below the positive rail.

Can I replace LM358 with MCP6002?

Pin-compatible and rail-to-rail, but the input stage, bandwidth and supply range differ. Validate the circuit, especially at the negative rail.

What is the difference between LM358 and LM2904?

Mostly temperature grade and qualification; the LM2904 is the automotive-oriented variant.

Why does my current-sense reading go non-linear at low current?

The output is in the crossover region near ground. Add an offset or use a rail-to-rail output stage.

Can LM358 drive an ADC directly?

It can, but check the output impedance and the ADC sampling capacitor. A series resistor and a small capacitor are often needed.

Is a faster op-amp a safe upgrade?

Not automatically. Higher bandwidth can oscillate in a layout designed for 1 MHz. Check stability with the actual load.

Can LM358 be used for audio?

It can pass audio, but the class-AB output adds crossover distortion and the bandwidth is limited. For a product where sound quality matters, use an audio op-amp designed for the job.

10. Sources

• TI LM358 and LM358B documentation: ti.com
• Microchip MCP6002 documentation: microchip.com
• ST op-amp portfolio: st.com