AD623_error output under stress

Q 

One of our customers meets an issue of AD623, Could you give any suggestion?
          The power is 3.3V single supply, REF pin is 720mV, Gain resistor is
470Ohm, Gain is about 210.
          When he twists the board, the AD623's output will also change. If he
twists counter-clockwise, the output will increase 0~8mV. if he twists
clockwise the output will decrease roughly 0~8mV. The input is from a bridge,
which share the power supply with AD623.  I tested the REF and Power when the
output changes, they are quite stable. 
          
           I also did the following test,

          1. Used 4 resistor to build a fixed output bridge, issue is still
there. So the problem is not from the sensor.
          2. Moved the Gain resistor and the Bridge to another board. AD623 was
the only component on the board, twist the board, problem was same. So problem
is not from the Gain resistor or bridge.
          3.Replaced AD623 with AD8223, the problem was same.

 

A 

The AD623 and AD8223 have trimmed resistors in the IC. When it is stressed, it
can act like a strain gauge. So when you twist one way, the AD623 strains in
one direction and outputs an error voltage. If you stress it the other way, it
will strain in the other direction and output a negative error voltage. 8mV is
high. If you remove the gain resistor, i.e use G=1, is the error as large?

We call this problem package and board stress. When the package or board
stresses the die, it strains the laser trimmed resistors on the IC causing an
error. We laser trimmed the resistors to be close to perfect. When we put it in
the package, it gets stressed and we end up with CMR of 70dB (or 80dB for the
AD8221). When the board stresses the part, the IC is stressed even more and we
get more CMR error and some offset error.

Larger packages have less stress. The SOIC is better than the MSOP package. The
DIP is even better.