I've used the C++ prog I found on the web to generate the standard Sbox with poly 0x11b, and also modified it to use a lookup table of all the multiplicative inverses of the 256 values, which also works fine. I'm now trying to generate alternative Sbox using polynomial 0x11d, as used in the paper by Das, Zaman & Ghosh, but without much success. I've again generated all the multiplicative inverses for the poly and used them in a lookup table, but I do not get the same results as the paper's authors.
The paper's tables in Appendix A are wrong.
For the entries in the first line of Table A1(a) I independently get:
0A 15 3F 2C 90 3B 19 51 FD EF 28 8C 83 A0 A7 36
and the fourth value (for input 03h) matches the OP's computation.
Note: there are fixed points (including 23h), invalidating the list of the valid additive constants, which is about the only attempt at an addition to self-citation  (at least, Appendix 1 table 2b appears correct and is available for free there).
Recommendation: if working on that paper is an assignment in a crypto course, question the competence or/and honesty of the assigner.