MgO-Tunneling-Barrier Crystallinity Influence on Resistance Sensing Margin for Spiking Neuron Having Perpendicular-Spin-Transfer-Torque Integrate Characteristic

The application fields of artificial intelligence (AI) has been widely expanded; security, finance, health care, etc. A conventional neuromorphic chips for AI has been fabricated with synapse array and neuron integration using C-MOSFET technology, having a scaling-down limit of neuromorphic chips because of a large neuron size and producing a high power consumption. In particular, as an alternative solution for scaling-down neuron, 2-terminal perpendicular-spin-transfer-torque (p-STT) neuron has been proposed [1], [2]. In this neuron, the face –centered-cubic (f.c.c) crystallinity of a MgO tunneling barrier is an extremely critical parameter to determine the integrate characteristic of a neuron. In our study, we investigated how the f.c.c crystallinity of MgO tunneling barrier influenced the resistance sensing margin for integrate characteristic in 2-terminal p-STT neuron. The f.c.c crystallinity of MgO tunneling barrier strongly depended on the RF sputtering power to sputter 1.15-nm-thick MgO tunneling barrier, as shown in Fig.1(a). The resistance difference between anti-parallel (AP) state and parallel (P) state increased with the RF sputtering power as shown in Fig.1(b). The better sequence of the f.c.c crystallinity of the MgO tunneling barrier was followed by 320, 350, 290, and 260 Watt, as shown in Fig. 1(c). The dependency of the f.c.c crystalliny of the MgO tunneling barrier RF sputtering power will be explained by crystalline defect distribution using spectroscopy ellipsometer. In addition, we examined the f.c.c crystallnitiy of the MgO tunneling barrier influence on the resistance sensing margin for integrate characteristic of neuron, as shown in Figs.1(d)-(g). Note that, in our experiment, the integrate characteristics of p-STT neuron was measured after a partially reset. The resistance sensing margin clearly increased with the f.c.c crystallinity of the MgO tunneling barrier in p-STT neuron and evidently increased with the input spike amplitude. In particular, an assure clarification of a resistance margin depending on the input spike amplitude could be achieved above the f.c.c crystallinity of the MgO tunneling barrier sputtered at 320 and 350 Watt. In out presentation, we will demonstrate the mechanism why the integrate characteristic of the neuron mainly depended on the f.c.c crystallinity of the MgO tunneling barrier by reviewing p-STT switching at a MgO tunneling barrier having a nanoscale-diameter MgO grain which affects the activation energy barrier for p-STT switching. Reference [1] Kondo, K., Choi, J., Baek, J., and Jun, H. (2018). A two-terminal perpendicular spin-transfer torque based artificial neuron. J. Phys. D Appl. Phys. 51:504002. [2] Dong Won Kim, Woo Seok Yi, Jun Young Choi, Kei Ashiba, Jong Ung Baek, Han Sol Jun, Jae Joon Kim and Jea Gun Park (2020) Acknowledgement This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (No. 2016M3A7B4910249)and the Brain Korea 21 PLUS Program in 2014. . Figure 1