Magnetization switching by the interaction between spins and charges has greatly brightened the future of spintronic memories. [1][2][3][4][5][6] This has been evident in the rapid development of spin transfer torque-magnetic random-access memory (STT-MRAM) as a mainstream non-volatile memory technology, in which a spin-polarized current is injected into magnetic tunnel junctions (MTJs) for cell programming. 7-18 However, as cell areas scale down to meet density and power demands, conventional STT-MRAM suffers from serious endurance and reliability issues due to the aging of the ultrathin MgO barrier and read disturbance. The challenge of lowering STT switching current densities to further reduce power consumption is still yet to be met. [19][20][21] The discovery of spin-orbit torque (SOT) switching in heavy metal/ferromagnetic metal/oxide heterostructures by applying an in-plane charge current to three-terminal devices provides a promising alternative mechanism. 22-28 It shows the potential to enhance the endurance and reliability of MRAM, while improving speed and reducing power consumption. [29][30][31][32] Thus, considerable research has been triggered to further elucidate the mechanism of SOT switching, which is currently described as magnetic reversal via two vector components, the damping-like (DL) and field-like (FL) torques. 33,34 Since the demonstration of perpendicular-anisotropy MgO/CoFeB MTJs (p-MTJs), the switching of perpendicular magnetization by SOT has become of particular interest. [33][34][35][36][37][38] However, an external magnetic field collinear with the charge current is required to execute deterministic switching of p-MTJs. This intrinsic constraint, combined with the three-terminal device configuration, is limiting the practical application of SOT-MRAM. [26][27][28]35 Great efforts have been made to eliminate the need