ic_engr wrote on May 13th, 2014, 7:56am:RobG,
Thanks for the comments, if any of the characteristics e.g. beta vs. IC and delta VBE vs. currents were an issue I should have seen in simulation.
The whole problem is that simulation is not predicting the tempco seen in silicon...
Yup... and if you aren't on the flat part of the curve the bipolar models can be suspect. (Bipolar models are always suspect anyway depending on the person running the curve fitter.) I'm reaching though... you have quite a bit more error than I'd expect from any modeling error or stress, but maybe that curvature compensation circuit adds extra sensitivity.
I had a 10% mismatch in a mirror in 0.18um that didn't use dummy devices.
Good luck and please let us know what you find out.