Because of the summing of 2 currents from 2 pairs with slightly different - yet controlled - offset voltages (due to the K ratio). Works correctly only with BJTs or MOSFETs in weak inversion. See the visualization in the PDF below: View attachment multi-tanh-linearization.pdf
Q10/Q11 isn't a differential pair. Please notice that they are operating on two separate current sources. I think, the explanation in the linked paper is rather graphic.
Q10/Q11 isn't a differential pair. Please notice that they are operating on two separate current sources. I think, the explanation in the linked paper is rather graphic.