The idea is to apply a temperature-dependent voltage to the input of A1 to counter the observed drift.  Thus, if A1 is observed to have an input drift of say 1 µV/ºC we can, for example, apply an equal drift to the opposite input.   

The temperature dependent voltage source can easily be made by connecting a diode from ground to VCC or VEE through a large resistor.  The temperature dependence of the voltage across the diode can be measured and suitably attenuated down to be fed into an input of A1.  Remember, the dependence of the diode voltage on the temperature is dependent on the current flowing through the diode, so this must be kept constant, i.e. do not change the current through the diode once the voltage dependence has been measured. The attenuation methods are shown below in Figs. 9(a) and 9(b).

Calculation of the resistor values can easily be done and leads to the results:

For corrections applied to the inverting input:  If the output of A1 requires a correction of
V_corr /ºC, and V_diode is the voltage change of the diode/ºC, then R_c =V_diode*R_f/ V_corr, where R_f is the feedback resistance.

Once R_c is selected, R_t will have to be changed to restore the original current through the diode, since some current will flow through R_c, and this has to be accounted for. If V₀ was the voltage on the diode before R_c was connected, and I_c is the current through R_c, R_t should be changed to R_t¹, where R_t¹ is given by the formula R_t¹ = I₀ x R_t/(I_c+I₀), where I_c=V₀/R_c and I₀=(Vcc-V₀)/R_t

If the correction is applied to the non-inverting input, then R_c=R_in x V_diode/V_corr, where V_corr is the correction needed at the input of A1. The value of R_t will have to be changed here as well to account for I_c. The same correction formula for R_t used in the last paragraph can be applied here as well. It is not exact, but the differences from the exact correction formula are negligible.