Op amp input offset voltage

[tedbar]

Trying to understand op amp Input Offset Voltage.

If an op amp has an Input Offset Voltage of 1mV does that mean:

1) That it will transition to zero V output when the input voltage differential is somewhere between -1mV and +1mV (typical) i.e the differential value at which it transitions is random between -1mV and +1mV irrespective of its current state.
or
2) That it will transition to 0V from a +ve voltage output when it is -1mV and to 0V from a -ve voltage output when it is +1mV i.e. something similar to hysteresis

?

[TimFox]

A good op amp (as opposed to a comparator with hysteresis to make a Schmitt trigger) should not show hysteresis.
A practical op amp with +1 mV offset will produce 0 V at its output when the + input is +1 mV with respect to the - input.
Since the DC gain of a normal op amp is huge (typically 200,000 V/V for a 741), the differential input voltage for any normal output voltage will be close to that offset:  for -10 to +10V output range, the input changes by only 100 uV for that gain.
It is difficult to measure the offset by changing the input.  A better method is to make a feedback network, say 10k feedback resistor and 10 ohms from - input to ground (+ input grounded) and measure the output voltage, with split power supply.  The low 10 ohms resistor will minimize the effect of the input bias currents.

[tedbar]

So if I understand correctly and op amp with a +1mV offset will work as you describe. An op amp with an offset of -1mV will need the input differential to be -1mV, with respect to the inverting input.

Some data sheets list the offset voltage as a +- figure, some as a simple mV figure.(absolute value?) Am I correct in reading that to mean that the op amp will (typically) transition at some point between +1mV and -1mV (using our 1mV example)? Any design needs to take this limitation into account.

[TimFox]

The specification for input offset is the guaranteed range of offset (measured in a similar way to my example), for the test conditions (supply voltages, temperature, etc.) called out on the data sheet, over the production from unit to unit.
If no polarity (+ or -) is given, that means absolute value.
Any device whose offset falls outside that range should not be labeled as such:  often a manufacturer sorts the production and sells the lowest offset parts with a suffix at a higher price to those who need it.
Many op amps allow a simple connection of a trimpot to manually adjust the offset to zero.

[golden_labels]

The value you see in a datasheet is not the value your particular, single op-amp has. If that was the case, it would mean the manufacturer intentionally makes your op-amp bad. Instead, the datasheet gives you statistical properties of the manufacturing process. Conceptually it’s the same as with resistor tolerance, but instead of getting ±%ohms, you get ±volts. The way you treat it in design is the same, including accepting the value may further change with temperature or drift as the component ages.

If you want a deeper explanation: production is imprecise. The actual parameters all over the range. The manufacturer samples the process and gets a probabilistic distribution of values. Then a datasheet reports what is the shape of that distribution. Usually a normal distribution is assumed with µ = 0 mV, so that is implied. The datasheet then reports variance. What you see as “TYP(ical)” ±value is the range 68.27% units fall into. The “MAX(imum)” ±value is unfortunately manufacturer-specific,⁽¹⁾ but the interpretation is “being outside that range is as rare enough you should count that the same as buying a broken part”.

If a normal distribution doesn’t approximate reality well enough, which may be important for precision parts, you also get a chart depicting the actual sample distribution, so you can judge for yourself what to expect.

To sum it up, if a datasheet reports 1 mV input offset, it means your part has the its individual input offset somewhere between -1 mV and 1 mV. May be -0.8 mV, may be 0.4 mV, bay happen to be 0.007 mV.

⁽¹⁾ May be ±3σ (99.7%), may be wider, may be the threshold at which binning/QA discards items, may be “not seen in a sample”.

[PGPG]

Some data sheets list the offset voltage as a +- figure, some as a simple mV figure.

May be sometimes datasheet author understands 'offset' as distance from 0 (distance is always positive).
Link the exact datasheet you have doubts to someone else than you look at it.

[tedbar]

It was a general question (not a specific op amp) but has been answered, am certain I understand the details now. Thanks.

[golden_labels]

I was answering from the perspective of op-amp’s input offset. While conceptually that holds for everything, the details may vary between component types and parameters.

For example resistors tolerance is always indicating the maximum deviation allowed, not a typical one. So no valid 5% 1 kΩ resistor is going to ever be outside of range [950, 1050] Ω. In fact the actual variation is much tighter than that. If component undergoes binning, the distribution doesn’t follow the bell curve too. Parameters, that can’t have symmetrical errors, may in some cases not be depicted well by a normal distribution. They will have bias towards either lower or higher value. Some parameters are specified over the entire operating range, not merely indicating imprecision in manufacturing. A good example are capacitors, which may have -80%/+20% tolerance for capacitance. Often inspecting charts and footnotes in datasheets gives hints.

[TimFox]

TI's app note about op amp input offset:  https://www.ti.com/lit/an/sloa059b/sloa059b.pdf?ts=1730106363656
In section 5.3, they state
  "The first entry is for V_OS under static temperature conditions, where the maximum and typical values are listed
for a temperature of 25°C. This specification is listed on virtually all op amp data sheets and is expressed in
millivolts (mV) or microvolts (μV). It is possible that natural variations or future changes in the process used to
create a device result in a V_OS that is much different from the typical value. Every device is tested at the factory
to ensure that it does not exceed the maximum specified value before it is shipped to the customer, so a robust
design may require a designer to look at maximum instead of typical values."
(emphasis added).
Since that important parameter is directly measured at production, it is truly a maximum, and not a statistical parameter.
The app note is careful to discuss the difference between "typ" and "max" values, and reminds the user to use max values intstead of typ for critical applications.
In general, actual specifications are those values and parameters that can be measured with automatic test equipment on the production line, other values may be "guaranteed by design" or "typical", especially if they are hard to measure quickly.

[golden_labels]

It’s confusing that in SLOA038: Understanding Op Amp Parameters (PDF) they claim to use ±6σ for extrema (p.11-1). Something is amiss, because the values in datasheets are clearly more restrictive than that. I suspect the paragraph may be mixing measuring a parameter (σ would refer to that) and process control charts (that would be 6σ). In this sense 6σ would mean ensuring manufacturing has no worse than 3.4 ppm failure rate. Which makes much more sense. But that’s only my guess.

While the document I linked is for Texas Instruments only and contains that unclear part, it still sheds some light on the meaning behind many parameters.

[ArdWar]

It's often a bit tricky to figure out if the min-typ-max performance values in a datasheet are manufacturing variations or operational variations. The two are distinct, and depending on your application can completely change the part's suitability. I often wish they make the distinction clearer.

One easy example is transistor or diode reverse leakage. They're often specced at something like "1nA typ, 1uA max", whoa 3 orders of magnitude. Does that mean you may find one diode with 1.2nA leakage and another one with 456nA leakage? Or if you measure one with 3.4nA leakage today and 567nA leakage tomorrow after a power down? Fortunately it usually not the case, the wide range are due to wide operating temperature, unit-to-unit variations are much tighter. But how'd you know that if you don't know the underlying mechanism?

That unfortunately isn't as easy to figure out with complex part like an opamp. As an example, some design have tight range of initial offset voltage until you exceed certain input voltage. Some even have input bias current that change direction halfway through operating range, and while technically the part offset voltage isn't changing, the system/circuit offset will certainly be. They'll spec the worst case in the datasheet (if they're honest, it's often hidden behind weird qualifier) and that may lead you to the wrong conclusion w.r.t part performance.