The paper presents a global optimization approach to the trajectory planning problem of mechanical manipulators. The purpose is to obtain a minimum-time cubic spline trajectory subject to constraints given by limited joint torques and torque derivatives taking into account the non-linear manipulator dynamics. It is shown how, without conservativeness, a semi-infinite optimization problem emerges. Conditions ensuring that the formalized problem admits a solution are given. The estimated global solution can be actually obtained by means of an hybrid genetic/interval algorithm that guarantees the feasibility of the found solution. The methodology is illustrated with numerical details for a two-link planar arm and a PUMA six-link manipulator; for the former, comparisons with an alternative optimization solver are exposed.