Voltage offset test

[Kingcosmos]

I am trying to perform a simple voltage offset test in Spice for an
OPA227. I have both inputs shorted to ground, +/-15V supply, and I am
getting 48.9mV on the output. The datasheet shows +/-100 uV for the
OPA227P and +/-200 uV for the OPA227PU. Either way, my test simulation
is off by a few magnitudes. Is this a limitation of the simulation or
is there a better testing circuit configuration for op-amps?

 

[John Popelish]

I am trying to perform a simple voltage offset test in Spice for an
OPA227. I have both inputs shorted to ground,

So they are at the same voltage.

... +/-15V supply, and I am
getting 48.9mV on the output.

So the input offset is being amplified by the open loop gain of the opamp.

The datasheet shows +/-100 uV for the
OPA227P and +/-200 uV for the OPA227PU. Either way, my test simulation
is off by a few magnitudes. Is this a limitation of the simulation or
is there a better testing circuit configuration for op-amps?

Since Spice can measure very small voltages, why not connect the opamp
as a unity gain follower, with zero volts on the + input. The output
will be at the offset voltage. Or connect the amplifier up as a
non-inverting gain of 100 (1k between inverting input and ground, 99k
resistor between non-inverting input and output, ground non-inverting
input), and the output will be at 100 times the offset voltage.

 

[Kingcosmos]

So they are at the same voltage.

Yes, they would be at ground potential.

This bothers me a little. The typical open loop gain of the OPA227 is
160dB. If the output I am getting is the product of the input offset
and open loop again, then that would make the input offset incredibly
small. Not that it is a horrible thing in fact it would be great. I
wouldn't have thought that the input offset would be THAT small and I
would expect the OPA227 to saturate to one of the rails. But I noticed
another result. If I make the supplies smaller, +/-5V, then the output
is 10mV. Why would changing the supplies make this drastic change?
as a unity gain follower, with zero volts on the + input. The output
will be at the offset voltage. Or connect the amplifier up as a
non-inverting gain of 100 (1k between inverting input and ground, 99k
resistor between non-inverting input and output, ground non-inverting
input), and the output will be at 100 times the offset voltage.

Thank you, I will try this and see what I get.

 

[John Popelish]

Yes, they would be at ground potential.




This bothers me a little. The typical open loop gain of the OPA227 is
160dB. If the output I am getting is the product of the input offset
and open loop again, then that would make the input offset incredibly
small.

The offset spec is either the typical value or the worst case. But
actual offset can be anything smaller than the worst case for the
given conditions, an I have no idea what offset is intentionally built
into your model.

Not that it is a horrible thing in fact it would be great. I
wouldn't have thought that the input offset would be THAT small and I
would expect the OPA227 to saturate to one of the rails. But I noticed
another result. If I make the supplies smaller, +/-5V, then the output
is 10mV. Why would changing the supplies make this drastic change?

For a real opamp, many things affect the offset voltage, including
temperature and supply voltages (which change the internal temperature
gradients). I doubt that all the things that alter the actual offset
voltage are included in the model you are using.

 

[Kingcosmos]

The offset spec is either the typical value or the worst case. But
actual offset can be anything smaller than the worst case for the
given conditions, an I have no idea what offset is intentionally built
into your model.
temperature and supply voltages (which change the internal temperature
gradients). I doubt that all the things that alter the actual offset
voltage are included in the model you are using.

Good show. At least this helps confirm my fears that it is the acutal
Spice model. I will have to test the silicon itself to get some 'real
world' values. BTW, I have cracked open the model and have no idea how
to read them. Is there a tutorial that explains the syntax or teaches
one to write a macro model? I am under the assumption that the
'language' used in Spice has a structure like C (things like functions,
variables etc) because in all honesty, it looks like meaningless
sentences.

 

[John Popelish]

Kingcosmos wrote:
(snip)

Good show. At least this helps confirm my fears that it is the acutal
Spice model. I will have to test the silicon itself to get some 'real
world' values. BTW, I have cracked open the model and have no idea how
to read them. Is there a tutorial that explains the syntax or teaches
one to write a macro model? I am under the assumption that the
'language' used in Spice has a structure like C (things like functions,
variables etc) because in all honesty, it looks like meaningless
sentences.

The choices available vary, somewhat, depending on what Spice you are
using. There are models that are component level (sub circuits that
use the standard component models and the schematic of the internal
circuit of the chip, but these are slow, and to be accurate must give
away all the design secrets of the chip. Some models are based on
polynomials that curve match various internal nonlinearities. Some
models are based on program secrets that are more accurate than the
polynomial versions, but still hide the design secrets of the chip
(LTspice does a lot of this for their chips). Some are based a
combination of basic components in combination with all kinds of
mathematical functions to simplify a full schematic model.

Reverse engineering a difficult model can be a real challenge.

Are you a member of the LTspice discussion on Yahoo groups? They have
some very knowledgeable members.