Integrated terahertz (THz) devices are capable of improving system performance in security imaging, high-speed communication, molecular spectroscopy, and other applications[1,2]. Various THz circuits including conventional microstrip lines, coplanar waveguides[4,5], substrate-integrated image guides (SIIGs)[6,7], and spoof surface plasmon polariton (SSPP) waveguides[8–13] have been reported. Metallic waveguides suffer from high ohmic losses at the THz band[14–18]. They are costly and difficult to fabricate. Thus, the all-dielectric chips are preferred. As the all-dielectric chips are based on guided-wave optics, they are reasonably efficient. However, it is not possible to independently operate the THz dielectric functional circuits due to the difficulties in the on-chip generation and detection of THz radiation. Usually, the dielectric functional circuits are connected to the conventional THz rectangular metallic waveguides or antennas. Hence, it is necessary to develop the THz couplers with high efficiency. Recently, different types of THz couplers have been reported, including inverse-taper waveguides, Luneburg lenses, and directly coupled couplers. All of these couplers work on the principle of coupling the THz wave through the chip facet in-plane. However, out-of-plane couplers can be used directly to inject and extract THz waves at arbitrary locations on the chip, which is very important for on-wafer testing. A promising approach is to couple the light to (or from) a chip via a diffraction grating with a spot-size converter formed on the chip surface. This technique is widely used in the optical domain[22–26]. Interestingly, the diffraction grating structures have good beam steering properties[27,28]. However, this coupler will be very large in size in the THz frequency range, probably more than 10 cm in length, due to the large wavelength of the THz wave. Thus, it will be difficult to integrate the coupler with other THz functional devices on a single wafer.