Table of Contents
AD8620, AD8622 and AD8628 solve different analog problems. Their similar names can hide a more consequential distinction than accuracy grade: one is a dual JFET-input amplifier, another is a dual precision low-power amplifier, and the third is a single zero-drift amplifier.
A replacement search should first reject devices that cannot operate on the existing supply or support the required number of channels. Only the survivors deserve a detailed noise and error comparison. The dual AD8620ARZ and single AD8628ARTZ-REEL7 illustrate why the full device and package identity must survive that screening. The AD862x directory keeps the exact ordering options visible without treating them as approved substitutes.
Which devices pass the supply and channel screen?
AD8620, AD8622 and AD8628 have different specified operating ranges and channel counts. Match the circuit’s actual rail extremes and signal swing to each datasheet before evaluating offset or sourcing cost.
| Device | Architecture or task distinction | Supply range described by the manufacturer | Channels |
|---|---|---|---|
| AD8620 | JFET-input, wide-bandwidth precision amplification | ±5 V to ±13 V | 2 |
| AD8622 | Precision low-power amplification | ±2.5 V to ±15 V | 2 |
| AD8628 | Auto-zero/zero-drift amplification | 2.7 V to 5 V single supply | 1 |
These ranges come from the AD8620, AD8622 and AD8628 documentation. An operating-range screen does not replace the electrical tables: input common-mode range, output loading, gain and temperature still matter.
For example, an existing ±12 V circuit excludes AD8628 without a power and signal-interface redesign. A 3.3 V sensor circuit points toward a different starting set. Neither case is resolved by choosing the amplifier with the smallest headline offset.
How does source impedance change the error comparison?
Source impedance converts input current into voltage error and adds resistor noise. Compare the amplifier with the source it will actually see, including impedance imbalance between the inputs.
A first-order bias-current estimate is Verror = Ib × Rsource. With an illustrative 1 nA and 100 kΩ, the contribution is 100 µV. At 1 kΩ it becomes 1 µV. These inputs are not specifications for the three devices.
For two inputs, include both bias currents, their mismatch and the resistance seen by each input. Bias-current compensation that works at one temperature may not retain the same residual error across the operating range. Use guaranteed limits where the design requires a guaranteed bound; typical values serve exploration, not worst-case acceptance.
Voltage-noise density alone is also insufficient. Include current noise multiplied by source impedance, resistor noise, the relevant bandwidth and the amplifier’s behavior at the chosen gain. A JFET input can be useful for a high-impedance source, but the full error budget decides whether it is suitable.
When should zero drift influence the selection?
Zero drift is useful when low-frequency offset stability limits the measurement and the circuit can accommodate the amplifier’s switching behavior. It does not make bandwidth, source impedance or sampled-system noise irrelevant.
For a slow bridge measurement, evaluate low-frequency noise and drift with the intended gain and filter. For a sampled signal chain, inspect whether switching artifacts can enter or alias into the measurement band. The zero-drift source-impedance guide develops that diagnostic task separately.
A faster waveform path may instead be limited by settling, slew rate, capacitive loading or recovery from overload. Compare those conditions at the required output swing. A part can produce excellent DC readings and still distort the transient that the application must measure.
Turn the shortlist into a controlled replacement trial
Keep supply range, input range, output load, gain, source impedance, signal bandwidth and temperature in the same comparison record. Add the required package pinout and channel use before any production purchase.
The evaluation should answer a specific acceptance question for each candidate: Does the DC error remain inside the allocated budget? Does the output settle before sampling? Is the circuit stable with the real cable or converter load? Does recovery from overload fit the system’s required response?
The op-amp replacement validation pillar provides the full validation sequence. Use the Analog Devices catalog only after the engineering shortlist preserves the complete device identity. A quote for an available AD862x option is a commercial input to that process, not evidence that the circuit will accept it. A candidate that fails an acceptance question stays outside the approved BOM until the circuit or the candidate changes.
Frequently Asked Questions (FAQ)
Is AD8628 the single-channel version of AD8620?
No. AD8628 belongs to a different zero-drift amplifier family. Similar part-number prefixes do not establish shared architecture or electrical behavior.
Can rail-to-rail output be treated as rail-to-rail input?
No. Input common-mode range and output swing are separate specifications, each with supply and loading conditions.
What should happen to an unused channel in a dual amplifier?
Follow the manufacturer’s guidance for a stable, valid input and output configuration. Leaving inputs floating can create unwanted behavior and should not be the default implementation.