Magnetic memory driven by topological insulators

Giant spin-orbit torque (SOT) from topological insulators (TIs) provides an energy efficient writing method for magnetic memory, which, however, is still premature for practical applications due to the challenge of the integration with magnetic tunnel junctions (MTJs). Here, we demonstrate a functional TI-MTJ device that could become the core element of the future energy-efficient spintronic devices, such as SOT-based magnetic random-access memory (SOT-MRAM). The state-of-the-art tunneling magnetoresistance (TMR) ratio of 102% and the ultralow switching current density of 1.2 × 105 A cm−2 have been simultaneously achieved in the TI-MTJ device at room temperature, laying down the foundation for TI-driven SOT-MRAM. The charge-spin conversion efficiency θSH in TIs is quantified by both the SOT-induced shift of the magnetic switching field (θSH = 1.59) and the SOT-induced ferromagnetic resonance (ST-FMR) (θSH = 1.02), which is one order of magnitude larger than that in conventional heavy metals. These results inspire a revolution of SOT-MRAM from classical to quantum materials, with great potential to further reduce the energy consumption.


Supplementary Note 3: Transport characterization of (BiSb)2Te3
The temperature dependence of the sheet resistance Rs = ρxx /t of (BiSb)2Te3 in Supplementary Fig. 3a shows the semiconducting properties, and the surface states dominate at lower temperature. From the Hall signals in Supplementary Fig. 3b, we can obtain the 2-dimeantional (2-D) carrier density of 2.1 × 10 12 cm −2 at 5 K and 8.6 × 10 12 cm −2 at 300 K, respectively.
Supplementary Fig. 3. a, Temperature dependence of the sheet resistance Rs in (BiSb)2Te3. b, The Hall resistance Rxy as a function of the magnetic field H at 5 K and 300 K, respectively. 5 We perform the current-driven SOT switching measurement in the MTJ device (Ru/CoFeB/MgO/CoFeB/Ta/Ru) without the topological insulator of (BiSb)2Te3, as shown in Supplementary Fig. 4a. In this case, there is no switching even the current density reaches 4 × 10 7 A cm −2 , indicating the negligible SOT contribution from Ru 4 .

Supplementary Note 4: Possible SOT contribution from Ru
Also, the SOT-induced ferromagnetic resonance (ST-FMR) is measured in the Ru/CoFeB/MgO stack ( Supplementary Fig. 4b) without TIs, where the tiny symmetric/antisymmetric contribution also indicates that the SOT from Ru is negligible.
Supplementary Fig. 4. a, The current-driven SOT switching result for a MTJ device GHz, with the magnetic field-driven FMR.

Supplementary Note 7: Pulse width dependence of SOT switching
We perform the SOT switching of the TI-MTJ device with varied pulse widths of writing current from 10 ms to 10 ns, as shown in Supplementary Fig. 7a and 7b. The switching current density Jc is gradually increased at a much shorter pulse width (around 2 times from 10 ms to 10 ns), which is a typical feature of the thermal activation range.
Supplementary Fig. 7. a, SOT switching with varied pulse widths of writing current from 10 ns to 10 ms. b, Switching current density as a function of the pulse width.

Supplementary Note 8: Thermal stability test of TI-MTJ device
In order to measure the thermal stability factor Δ in TI-MTJ device, we perform the switching probability P as a function of the applied magnetic field H, by using the 8 relation based on Stoner-Wohlfarth model 5 : Where τ is the pulse duration time (1 s) of H, τ0 is the inverse of attempt frequency (1 ns), Hk eff is the effective anisotropy field.
For experiment, the minor loops of TMR-H are measured for 50 times, as shown in Supplementary Fig. 8a, where only the free layer is switched; and thus, to obtain the switching probability P as a function of the magnetic field H, as shown in Supplementary Fig. 8b. After fitting the P-H curve with above equation, we can obtain a thermal stability factor Δ = 61.
Supplementary Fig. 8. a, Minor TMR-H loops measured for 50 times. b, Switching probability P of the free layer as a function of the applied magnetic field H.

Supplementary Note 9: Current distribution estimation
The  Figure 9 shows the thickness dependent transport data of (BiSb)2Te3 (BST) with the same growth recipe. With BST thickness t increasing from 4 nm to 12 nm, the sheet resistance Rs decreases monotonically, indicating the increasing portion of bulk conduction involvement, as shown in Supplementary Fig. 9a. For the thickness dependence (t) of resistivity (ρ) in Supplementary Fig. 9b, it shows a similar trend with  Table 2. It should be noted that the bulk band gap of BST is small ~ 0.2 eV, so there would still be a considerable portion of bulk states contribution at room temperature (300 K).