Two-dimensional Layered HfS₂-based Transistors And The Neural Synapse Simulation

The development of integrated circuits has crossed into the post-Moore era.Due to the perfect lattice structure,ultra-thin thickness and strong gate control ability of two-dimensional materials,indicating that the two-dimensional materials are the most potential materials to further develop Moore's Law under the background of scaling transistors'feature size.Furthermore,because of the development of information society,efficient computers with higher data processing is required.Therefore,research on synaptic devices based on two-dimensional material transistors to simulate human brain synaptic function and realize neuromorphic computing,which can not only continue Moore's Law,but also realize high-efficiency and low-power data processing.It is also an important development direction of modern information computing science.Among many 2D materials,hafnium sulfide is favored due to its ultrafast photoelectric response,ultrahigh theoretical mobility,and good compatibility with the high-k material Hf O₂.In this paper,based on two-dimensional layered hafnium sulfide,Hf S₂transistors and their 2×2 arrays,Hf S₂/h-BN/Graphene floating gate transistors were fabricated,their basic electrical properties were tested,and then the the synaptic plasticity was mimicked.Using the electrical parameters of synaptic devices based on two-dimensional transistors to construct artificial neural networks for machine learning.The work content and results of this paper are as follows:(1)The two-dimensional layered Hf S₂was exfoliated by mechanical exfoliation to prepare Hf S₂transistor.The electrical properties of the Hf S₂transistor was tested by a probe station and a semiconductor parameter analyzer.A large hysteresis(memory)window was observed in its transfer characteristic curve.In order to explore the origin of the hysteresis window,a Hf S₂/h-BN/Si O₂/p⁺⁺-Si structure transistor was designed as a control experiment,which proved that the hysteresis window originated from the interface trap between Hf S₂and Si O₂.Subsequently,in order to prove the repeatability of the device,the electrical parameters of 20 devices were statistically analyzed:the value of I_on/I_offreached 10~5,the carrier mobility was between 1?3 cm~2V^-1s^-1,the threshold voltage was around 20 V,and the hysteresis window reaches 30 V,indicating that Hf S₂transistors not only have memory effect,but also has excellent electrical properties and experimental repeatability.And the hysteresis window of the transfer characteristic curve provides a physical theoretical basis for the next visual synapses.(2)Under 405 nm wavelength laser,using the interface defect engineering of Hf S₂transistors to simulate the plasticity of human visual synapses,including optical pulse intensity-dependent properties,optical pulse-width-dependent properties,paired pulse facilitation,pulse number-dependent properties,and pulse frequency-dependent properties.Then prepared a 2×2 array of Hf S₂transistors to explore the short-term memory(STM)and long-term memory(LTM)effects of synaptic devices.The synapse plasticity of human visual synapses is well modeled by synaptic devices based on Hf S₂transistors.Next,the long-term potentiation(LTP)and long-term depression(LTD)of each synaptic device in the 2×2 array were studied,and they all showed good linearity,large conductance gain.Finally,the change of conductance value in the LTP and LTD of the Hf S₂synaptic transistor is used as the basis for updating the weight of the machine learning artificial neural network.The handwritten numbers are learned and recognized,and the recognition rate reaches88.5%.It is demonstrated that the synaptic devices based on Hf S₂transistor are promising for new generation artificial vision systems.(3)Graphene/h-BN/Hf S₂structure floating gate transistors were prepared.In order to verify the repeatability of the device,the electrical performance data of 5floating gate transistors were counted.The test results showed that the current switching ratio reached 10~5,the electron mobility was 8?15 cm~2v^-1s^-1,and a storage window of 100 V was observed in the transfer characteristic curve.By designing Hf S₂/Si O₂/p⁺⁺-Si and Hf S₂/h-BN/Si O₂/p⁺⁺-Si structure transistors as comparative experiments.It is verified that the storage window is caused by the floating gate structure.Then,the data retention capability and endurance of the Graphene/h-BN/Hf S₂floating gate transistor were tested,showing excellent charge retention capability(10~4s)and endurance(10~3).Then,based on Hf S₂floating gate transistors,biological synapses were simulated,including short-term memory and long-term memory under electrical stimulation and light stimulation.The synaptic plasticity of biological synapses is well simulated by testing the current changes of the synaptic devices based on Hf S₂floating gate transistor under stimulation with different electrical(optical)pulse amplitudes,different pulse widths,and different pulse numbers.When studying the long-term plasticity of synaptic device based on Hf S₂floating gate transistor,the G_max/G_minvalue of LTP/LTD reached 34.7,the nonlinearity was around 0.3.It shows high conductance gain and good linearity,and the energy consumption per pulse can be as low as 0.2 p J.Therefore,based on the change of the conductance value of LTP and LTD as the basis for updating the weight of the machine learning neural network,an artificial neural network is constructed to learn and recognize the handwritten digits,and the recognition accuracy reaches91.5%.