The thermoelectric generator (TEG) is typically used as a clean power supply to harvest waste heat energy in applications involving a large thermal gradient, such as industrial heat removal and power electronic equipment systems. However, it is often difficult to achieve the optimal output power in the loop of the array system of the TEG owing to different output loads. This article proposes an improved fast terminal sliding-mode variable-structure control algorithm (FTSMC) to maximize power point tracking. The variable-structure sliding-mode control function used in the nonlinear sliding-mode surface of the algorithm allows us to obtain the characteristics of global stability that can enable it to converge to the sliding-mode surface at any position to reduce chatter. Digital modeling and simulation as well as experimental developmental field programmable gate array platforms were built to verify the effectiveness of the proposed FTSMC. It can attain the nonlinear sliding mode more quickly than the traditional sliding-mode algorithm. The results of experiments show that it can reach a tracking response speed of 0.08 s and a maximum conversion efficiency of 99.91%. The work here provides a new way for the efficient use of the TEG array for waste heat recovery.