It could be taken as u_sw=-(1/b)ksgn(s) where k is defined as in Applied nonlinear Control by slotine. Does the definition vary for advanced SMC like non-singular terminal SMC?
The first step is to define a "Sliding Surface" and make your system dynamic reach there and then stay there afterwards. We normally take
surface = S = sum of states = 0 and then define a control law that ensures it. Now one way to approach it is by defining a Lyapunov function as
V=1/2(S^2) so that V'=SS' and then take the control input U such that V' is negative definite (i.e. if V' = F(x) + G(x)U take U=1/G(x) [ -F(x) - Ksign(s) ] and your V' becomes
V' = -Ksign(s) which is negative definite and ensures that the system trajectory reaches the surface S=0 in a finite time and then stays there.
So we take our system dynamics to a predefined surface which ensures that origin of the closed loop system is asymptotically stable.
A very good reference on this topic is the Book "Non-Linear Systems" by Hassan K Khalil Chapter 14 (3rd Ed). I have it in pdf so ask me if you can't find it.
It's not necessry to design a sliding surface and this approch did not work in general if the surface are not lipschitz. The objective is to stabilize a perturbed integrator chain.
You can refer to my proposed approach wich you can stabilize a perturbed integrator chain, based on controller wich stalilize a pure integrator chain, and based on Lyapunov analysis.