| With the development of technology,the optical,electrical and mechanical properties of two-dimensional materials are significantly better than those of bulk materials,which make them important focus in the research field today.With the development of Moore’s Law approaching physical limits,researchers urgently need to develop new materials and devices as candidates for silicon-based materials in the postMoore era.At present,memory devices are widely used.Floating gate memory based on two-dimensional materials have the characteristics of high sensitivity,small area and fast response speed.Among them,tungsten diselenide(WSe2)is a representative twodimensional material,which is often used as conductive channel due to its bipolarity,high mobility,high ON/OFF ratio and high optoelectric response characteristics.In this paper,the application of two-dimensional materials based on WSe2 heterostructure floating gate device is studied,especially a new non-volatile memory device is studied in depth.The main contents of the paper include:We first studied tungsten diselenide/hexagonal lattice boron nitride/multilayer graphene heterostructure floating gate device.Among them,tungsten diselenide is used as channel,hexagonal lattice boron nitride as tunneling layer,and multilayer graphene as floating gate.The device exhibits a large memory window(～11.5 V)and a large ON/OFF ratio(～104),but the ON/OFF ratio decays to 10 within 2000 s,showing poor data retention characteristics.To solve this problem,by studying the mechanism of data attenuation and fitting the data with the theoretical model,we find that the data attenuation stems from charge leakage in the floating gate.Due to the poor data retention characteristics of the devices mentioned above,we introduced a layer of hexagonal lattice boron nitride between the silicon oxide substrate and the floating gate,and proposed a novel tungsten diselenide/hexagonal lattice boron nitride/multilayer graphene/hexagonal lattice boron nitride heterostructure floating gate structure.The device exhibits relatively high data retention stability,with an ON/OFF ratio of 15 for 1000 s,and a very low erase energy consumption(～0.6 fJ).The Hopfield neural network computing architecture built by this device can be used to accelerate neural network computing,and the whole system shows very low energy efficiency(～274 TOPS/W).In order to further reduce the energy consumption of the device,we chose a P-type tungsten-N-type molybdenum disulfide diode with higher sensitivity as channel,and propose a new p-n diode floating gate structure.The device exhibits a rectifier ratio of 200,a storage window of 14 V,an ON/OFF ratio of 104,data retention characteristics of 5000 s,and a durability of 1500.The performance of the device in simulating artificial synapses is demonstrated,featuring polymorphic memory(10/15)characteristics,high linearity(-2.0/-2.5)and very low excitation/inhibition energy consumption(9/15 aJ).These results demonstrate the potential of the diode floating gate device in neuromorphic computing.In order to realize the perception and storage of optical information,and expand it to the field of ultraviolet light,we combined gallium nitride as the substrate with floating gate device,and proposed a new gallium nitride/tungsten diselenide/hexagonal lattice boron nitride/multilayer graphene floating gate structure.In addition to characterizing the basic storage electrical properties,we also describe the synergistic interaction mechanism between optical and electrical impulses and demonstrate the heterostructure’s ability to act as photoelectric synaptic device.We demonstrate pulse time-dependent plasticity,paired pulse promotion,spike number-dependent plasticity,and spike rate-dependent plasticity through electrical and optical stimuli.In addition,we show the optical response of the heterostructure in the spectrum range of 265～750 nm,demonstrating its capabilities as a photodetector.This provides a possibility for future development in phototunable neuromorphic computing. |
PDF Full Text Request