Search

In this work, a memristor using Si nano-tip bottom electrode has been fabricated and evaluated. Compared with the control device fabricated in a planar structure, the invented Si nano-tip device statistically demonstrates 400 times of area shrinkage and 3 and 8 times of current reductions in the low- and high-resistance states, respectively, owing to the effects of highly confined switching area on the operations of multiple-weak-filament (MWF)-type memristor. The results reveal the validity of structural scaling on the MWF-type memristors as effective as a material approach. Moreover, the fabricated Si memristor has been evaluated as a synaptic device via pattern recognition toward neuromorphic applications in consideration of interconnect resistance. The proposed Si memristor shows substantially reduced inference power consumption by an increase in resistances and improved accuracy by 5% compared with the control device. [ABSTRACT FROM AUTHOR]

As interest in artificial intelligence (AI) and relevant hardware technologies has been developed rapidly, algorithms and network structures have become significantly complicated, causing serious power consumption issues because an enormous amount of computation is required. Neuromorphic computing, a hardware AI technology with memory devices, has emerged to solve this problem. For this application, multilevel operations of synaptic devices are important to imitate floating point weight values in software AI technologies. Furthermore, weight transfer methods to desired weight targets must be arranged for off-chip training. From this point of view, we fabricate 32 × 32 memristor crossbar array and verify the 3-bit multilevel operations. The programming accuracy is verified for 3-bit quantized levels by applying a reset-voltage-control programming scheme to the fabricated TiOx/Al2O3-based memristor array. After that, a synapse composed of two differential memristors and a fully-connected neural network for modified national institute of standards and technology (MNIST) pattern recognition are constructed. The trained weights are post-training quantized in consideration of the 3-bit characteristics of the memristor. Finally, the effect of programming error on classification accuracy is verified based on the measured data, and we obtained 98.12% classification accuracy for MNIST data with the programming accuracy of 1.79% root-mean-square-error. These results imply that the proposed reset-voltage-control programming scheme can be utilized for a precise tuning, and expected to contribute for the development of a neuromorphic system capable of highly precise weight transfer. [ABSTRACT FROM AUTHOR]

In this study, we examine an ITO/NiO X /WO X /Pt p-n heterojunction memristor for neuromorphic applications as a synaptic crossbar array. The transition in the depletion region at the interface between p-type NiO X and n-type WO X crucially influences the self-rectifying characteristics of the device depending on the voltage polarity. Furthermore, long- and short-term memory coexist depending on the switching voltage condition, thus enabling various neuromorphic applications, such as reservoir computing and the use of the synaptic device at the off-chip trained network. The reliable operational characteristics are confirmed by obtaining an average memory window (>9) and rectification ratio (>58) between device-to-device and cycle-to-cycle. Furthermore, synaptic functions were successfully implemented, such as spike-rate-dependent plasticity, spike-number-dependent plasticity, paired-pulse facilitation, post-tetanic potentiation, paired-pulse depression, and post-tetanic depression, in conjunction with repeatability. Ultimately, reservoir computing is accomplished based on the I–V nonlinearity and short-term memory characteristics. A high pattern recognition rate (>96.2 %) is achieved in MNIST tasks using 16 reservoir states, thus affirming the trustworthiness of the reservoir computing system. With its comprehensive approach to synaptic applications for neuromorphic computing systems, the heterojunction device highlights its considerable potential to advance artificial neural networks in conjunction with novel memristor technology directions. • Memristor devices incorporate the diode junction of the NiO X /WO X. • Linear potentiation and depression are enhanced for neuromorphic systems. • Controlling CC activates both long- and short-term memory. • Time-domain conductance modulation enables reservoir computing. [ABSTRACT FROM AUTHOR]