In recent years, research on neuromorphic chips of spin-based devices capable of low-power, ultra-high-speed switching operation has been actively conducted. There were reports of spin-based neuromorphic devices using spin-transfer-torque (STT), which are assisted by an external magnetic field [1] and spin-orbit-torque (SOT) spin neurons via heavy metal that have a high spin Hall angle. [2-4] Since these spin neurons operate with pulse bias, they are suitable for calculating spiking neural network (SNN), which is a low-power artificial neural network (ANN) that operates similar to the human brain. However, the previously announced devices required an external magnetic field or implemented in a three-terminal method (16F2), making fabrication impossible for high density and low-power devices. In this study, a 300 nm-scale 2-terminal (4F2) perpendicular STT-based (p-STT) spin neuron device capable of switching operation without an external magnetic field was fabricated to demonstrate stochastic neuron nature. Furthermore, an SNN was constructed using fitting data and an MNIST pattern recognition test was conducted.As shown in Fig. 1-(a), a 300 nm pattern process was processed via E-beam lithography of a double MgO p-STT magnetic tunnel junction (MTJ) with a tunnel-magnetoresistance (TMR) ratio of 158.13%. Figure 1-(b) is a transmission electron microscopy (TEM) image of a 300 nm scale p-STT spin based neuron. Figure 1-(c) is a DC sweep R-V graph of a spin neuron. It was confirmed that the resistance was 985 Ω in parallel state (LRS) and 2137 Ω in anti-parallel state (HRS) relatively, and the difference of resistance was 1114 Ω. To measure the stochastic characteristics depending on the set voltage, the set voltage pulse amplitude was varied from -0.61 mV to -0.695 mV with pulse width of 250 μs. It was confirmed that switching is impossible below -0.61 V, and switching occurs with one pulse above -0.695 V. In addition, as a result of the measurement the switching probability depending on the set voltage pulse was fitted with a sigmoid function. Furthermore, we performed a pattern recognition simulation assuming that p-STT neurons were applied in a single-layer SNN consisting of 784 input layers and 100 output layers. In this simulation, the images used for pattern recognition were MNIST handwritten digit images with a size of 28X28 pixels, 60000 images were used for training, and 10000 images not used for training were used for testing. In our presentation, we will review in detail the stochastic nature of the p-STT neuron depending on the spike amplitude and SNN pattern recognition.This research was supported by National R&D Program through the National Research Foundation of Korea(NRF) funded by Ministry of Science and ICT(2021M3F3A2A01037733) Reference [1] Kondo, Kei, et al. "A two-terminal perpendicular spin-transfer torque based artificial neuron." Journal of Physics D: Applied Physics 51.50 (2018): 504002.[2] Kurenkov, Aleksandr, et al. "Artificial neuron and synapse realized in an antiferromagnet/ferromagnet heterostructure using dynamics of spin–orbit torque switching." Advanced Materials 31.23 (2019): 1900636.[3] Ostwal, Vaibhav, et al. "Spin-torque devices with hard axis initialization as Stochastic Binary Neurons." Scientific reports 8.1 (2018): 1-8.[4] Reidy, Brendan, and Ramtin Zand. "SOT-MRAM based Sigmoidal Neuron for Neuromorphic Architectures." arXiv preprint arXiv:2006.01238 (2020). Figure 1