Magnetic Random-Access Memory (MRAM) represents a promising candidate for the next generation of memory devices due to its high speed, non-volatility, and impressive storage density. Within the MRAM family, spin-orbit torque MRAM (SOT-MRAM) offers several key features, including the separation of read and write channels, non-destructive read and write operations, and endurance of up to 1015 cycles [1-4]. Notably, SOT-MRAM exhibits robust probabilistic switching characteristics, allowing precise control of the writing probability based on the magnitude of the writing SOT current. Consequently, SOT-MRAM holds significant potential across probability-related fields, including applications in probabilistic binary neural networks (PBNNs) and true random number generators [5-7]. In this paper, we fabricate elliptical magnetic tunnel junctions based on W/CoFeB/MgO/CoFeB on an 8-inch wafer (Fig. 1a). Remarkably, our SOT-MRAM devices achieve reliable switching even with pulses as short as 2ns (Fig. 1b). Furthermore, we measure the tunnel magnetoresistance ratio (TMR) in over 100 randomly selected devices, observing an approximately normal distribution. Notably, the minimum TMR exceeded 69%. To assess endurance, after over 1011 read-write cycles we find that resistance fluctuations remain below 3%. We also perform curve fitting for probabilistic switching, yielding a thermal stability factor of 68.5 ± 5.6 (Fig. 2a). Subsequently, we analyze the probabilistic switching performance by testing 8 randomly selected devices from the wafer (Fig. 2b). These devices consistently exhibit excellent and stable probabilistic switching behavior, with minimal differences among multiple devices. These findings demonstrate the strong performance and consistent probabilistic characteristics of the SOT-MRAM devices fabricated in our study, highlighting their potential for large-scale applications.References: [1]L. Liu, C. Pai and Y. Li, Science., Vol. 336, p.555-558(2012) [2]J. Lu, W. Li and J. Liu, Appl. Phys. Lett., Vol. 122, no.1, Art. no.012402(2023) [3]W. Li, Z. Liu and S. Peng, IEEE Electron Device Lett., Vol. 45, no.5, p.921–924(2024). [4]S. Peng, D. Zhu and W. Li, Nat. Electron., Vol. 3, p.757–764(2020) [5]R. Zhang, X. Li and M. Zhao, Adv. Sci., Vol. 11, no.23, p.2402182(2024) [6]H. Zhang, Y. Sun and K. Hu, J. Semicond., Vol. 43, no.10, p.102501(2022) [7]J. Yin, Y. Liu and B. Zhang, in Proc. IEEE Int. Electron Devices Meeting., p.36.1.1-36.1.4(2022)

Magnetic Random-Access Memory (MRAM) represents a promising candidate for the next generation of memory devices due to its high speed, non-volatility, and impressive storage density. Within the MRAM family, spin-orbit torque MRAM (SOT-MRAM) offers several key features, including the separation of read and write channels, non-destructive read and write operations, and endurance of up to 1015 cycles [1-4]. Notably, SOT-MRAM exhibits robust probabilistic switching characteristics, allowing precise control of the writing probability based on the magnitude of the writing SOT current. Consequently, SOT-MRAM holds significant potential across probability-related fields, including applications in probabilistic binary neural networks (PBNNs) and true random number generators [5-7]. In this paper, we fabricate elliptical magnetic tunnel junctions based on W/CoFeB/MgO/CoFeB on an 8-inch wafer (Fig. 1a). Remarkably, our SOT-MRAM devices achieve reliable switching even with pulses as short as 2ns (Fig. 1b). Furthermore, we measure the tunnel magnetoresistance ratio (TMR) in over 100 randomly selected devices, observing an approximately normal distribution. Notably, the minimum TMR exceeded 69%. To assess endurance, after over 1011 read-write cycles we find that resistance fluctuations remain below 3%. We also perform curve fitting for probabilistic switching, yielding a thermal stability factor of 68.5 ± 5.6 (Fig. 2a). Subsequently, we analyze the probabilistic switching performance by testing 8 randomly selected devices from the wafer (Fig. 2b). These devices consistently exhibit excellent and stable probabilistic switching behavior, with minimal differences among multiple devices. These findings demonstrate the strong performance and consistent probabilistic characteristics of the SOT-MRAM devices fabricated in our study, highlighting their potential for large-scale applications.References: [1]L. Liu, C. Pai and Y. Li, Science., Vol. 336, p.555-558(2012) [2]J. Lu, W. Li and J. Liu, Appl. Phys. Lett., Vol. 122, no.1, Art. no.012402(2023) [3]W. Li, Z. Liu and S. Peng, IEEE Electron Device Lett., Vol. 45, no.5, p.921–924(2024). [4]S. Peng, D. Zhu and W. Li, Nat. Electron., Vol. 3, p.757–764(2020) [5]R. Zhang, X. Li and M. Zhao, Adv. Sci., Vol. 11, no.23, p.2402182(2024) [6]H. Zhang, Y. Sun and K. Hu, J. Semicond., Vol. 43, no.10, p.102501(2022) [7]J. Yin, Y. Liu and B. Zhang, in Proc. IEEE Int. Electron Devices Meeting., p.36.1.1-36.1.4(2022)