- consider changing the output stage cascoded NFETs (M4-M13, M12-M14) into a mirror configuration, this will increase gain by a factor of two and also provide other advantages
- the output stage current will be defined by slew rate constraints, which in turn depend on the output capacitance
you can still increase the output resistance by increasing the gm of the cascoded FETs - M18, M17, M1, M2
- for the diff pair you can increase the W/L to push them towards weak inversion
- you have really high gm on the PFET M1 M2, which suggests also high gds; high gm on M1-M2 does not buy much so sacrifice it for a lower gds, lower the gds at least until it matches the gds of the symmetric pair M14-M12
Let us know how it goes...