Recently, the world-renowned comprehensive academic journal Advanced Science reported the latest achievements of Professor Wang Xingsheng and Miao Xiangshui from the School of Optical and Electronic Information at Huazhong University of Science and Technology (HUST) in the research of high-performance memristive synaptic devices, entitled "HfOx/AlOy Superlattice-Like Memristive Synapse." HUST is the first and corresponding affiliation of the paper, Professor Wang Xingsheng is the corresponding author of the paper, and doctoral students Wang Chengxu and Mao Ge-Qi are the co-first authors of the paper. This is also the second important research achievement in the field of memristors published in Advanced Science by Professor Wang's group in the past two months.
Brain-inspired neuromorphic computing is regarded to be a promising computation architecture for breaking the Von Neumann bottleneck, and has been used for artificial intelligence (AI). Because of its simple construction, low power consumption, scalability, and process compatibility, memristive synaptic devices owning gradual conductance adjustment capabilities are viewed as the more attractive alternatives for synaptic devices than CMOS circuit-based electronic synapses. Kirchhoff’s law and Ohm’s law state that a memristor crossbar array can perform vector-matrix multiplication of voltage input vector and conductance matrix itself in a single step.
The key prerequisite for memristors to be employed as efficient synaptic devices is that they have analog switching behavior rather than the binary switching process. Filamentary memristors, on the other hand, are never easy to achieve analog switching behavior because their conductive filament (CF) usually develops or breaks abruptly.
Figure caption: The Superlattice-like functional switching layer design and the corresponding memristor synaptic behavior.
Aiming at the problem of poor resistance gradient tuning ability, the research group designed and fabricated a HfOx/AlOy superlattice-like (SLL) memristor. By utilizing the higher migration barrier of VO in Al2O3 atomic layer, several barrier layers were periodically arranged in HfOx switching layer to gracefully control the VO migration and the formation and rupture of CFs. The devices with optimized SLL switching layer were measured and analyzed, showing bidirectional analog-type switching behavior. The SET and RESET processes exhibited 160-level and 62-level resistance states respectively under the tuning of DC bias without abrupt switching behavior. The device exhibited excellent synaptic properties, which can be gradually regulated in 100-level conductance states between 20 ~120 μS under fixed pulses of 100 ns, with great weight update linearity and long-term synaptic plasticity. The advanced synaptic performance of the SLL memristor was verified in a convolutional neural network, where the handwritten number recognition rate based on the SLL memristor can be up to 94.95%. In addition, the device showed fast operation speed, which can achieve high linearity continuous gradient regulation of conductance states and stable cycling, under fixed pulses of 30 ns. According to the fitting of the conductive mechanism and the results of the first principle calculation, because of the blocking effect of AlOy layers on VO and CFs, in the low resistance state, multiple weakness CFs will be formed in the SLL switching layer. There are three elements of CF that can be gradually regulated under the pulse regulation, including CF`s number, width, and gap, which can provide more conductance states for modulation.
This work was supported in part by the National Key Research and Development Program of MOST of China, and in part by Hubei Yangtze Memory Laboratories and Hubei Key Laboratory of Advanced Memories.
Article information: http://doi.org/10.1002/advs.202201446
Source: School of Optical and Electronic Information
Edited by: Luo Xin, Jiang Jing