Hi Michel,
The sky diffuse factor for non-linear self shading has a relatively small effect on the output of the array, and the discrepancy you have pointed out between SAM's results and values calculated using the equations in Seciton 9.2 of the PV reference manual is also small, so we haven't made sorting this out a priority.
The equations for these calculations are in the sssky_diffuse_table() function, which is defined here in the SSC source code:
github.com/NREL/ssc/blob/f6cf8d33f9f8bc859d174b7c41e8a07eda0d019b/shared/lib_pvshade.cpp#L300
I created an issue for this in the SAM GitHub repository:
github.com/NREL/SAM/issues/1705.
Best regards,
Paul.