@Manul : The measurement is made on the pins. The oscilloscope shows only noise.
@Andreas : The supply was linear lab supply, now with battery : Same results
@oPossum : With follower, one input is connected at the output, without resistor.
@tggzzz : The first circuit was on breadboard, the second is PCB with ground plane. The +/- rail are decoupled with 100n
@David Hess : A resistor makes the same results.
I made a setup (attached) with better wiring.
2 battery coins -3V 0 +3V
The first amplifier is a follower with input at ground
The ouput is connected to the second non-inverting amplifier, gain=100
The multimeter is connected via coax + 50 ohms load
Output 1 : 500µV with noise, but without oscillations (without the 50 ohms load : 4mV)
Output 2 : 40 mV, confirmed with oscilloscope, noise not visible
40mV / 100 = 400 µV = (Voffset1 +/- Voffset2 ) * 100
The datasheet shows a span of Voffset between -200 / +200 µV
So I assume that measuring 100µV without noise requires a lot of skill
If the second amplifier says ~200 µV for each amplifier, that is the thruth
The Voffset is in the specs. The follower follows the theory. I will not the next Nobel !
Thanks.
Denis