3D Artificial Neural Network Research Articles

Novel Three-Dimensional Artificial Neural Network Based on an Eight-Layer Vertical Memristor with an Ultrahigh Rectify Ratio (>107) and an Ultrahigh Nonlinearity (>105) for Neuromorphic Computing.

In recent years, memristors have successfully demonstrated their significant potential in artificial neural networks (ANNs) and neuromorphic computing. Nonetheless, ANNs constructed by crossbar arrays suffer from cross-talk issues and low integration densities. Here, we propose an eight-layer three-dimensional (3D) vertical crossbar memristor with an ultrahigh rectify ratio (RR > 107) and an ultrahigh nonlinearity (>105) to overcome these limitations, which enables it to reach a >1 Tb array size without reading failure. Furthermore, the proposed 3D RRAM shows advanced endurance (>1010 cycles), retention (>104 s), and uniformity. In addition, several synaptic functions observed in the human brain were mimicked. On the basis of the advanced performance, we constructed a novel 3D ANN, whose learning efficiency and recognition accuracy were enhanced significantly compared with those of conventional single-layer ANNs. These findings hold promise for the development of highly efficient, precise, integrated, and stable VLSI neuromorphic computing systems.

(Invited) ALD Based Flexible Memristive Synapses for Neuromorphic Computing Application

With the gradual development of artificial intelligence devices, the demand for data-intensive intelligent computation is increasing. The classical von Neumann computing scheme is suffering a bottleneck of complex information processing and storage[1]. In order to improve computing efficiency, In-memory Computing architectures were proposed to complete computing tasks based on storage units, including neuromorphic computing and logic-in-memory. On one hand, neuromorphic computing has become an attractive candidate due to the high efficiency by emulating the computing mode of human brain[2]. On the other hand, in-situ Boolean logic calculations can combine the arithmetic logic unit (ALU) and the storage unit to form memory-ALU (MALU), which could use the storage states of the memristor as the logic states to implement 16 basic Boolean logic gates[3]. Memristors with continuously adjustable conductance induced by voltage stimulation are suitable for simulating the weights update of bio-synapse[4]. Firstly, the ALD flexible artificial neural network (ANN) based on the integration of three-dimensional (3D) memristors network was proposed. Based on the structure of Pt/HfAlOx/TaN, the 3D ANN not only exhibits excellent memristive switching characteristics, but also achieves ultralow power synaptic characteristic (4.28 aJ/spike) under spike of 50 ns. The operation speed is 5 orders of magnitude higher than that of biological synapses, and the power consumption is 2-4 orders of magnitude lower than that of biological synapses. Besides, the 3D ANN has realized the function of information transmission including single channel and multi-channel, which could successfully transfer the letters “F”, “F” and “U” between the synapses of multilayer nerves. For the "0-9" digital images (28×28 pixels) in the MNIST database, the 3D ANN can complete the recognition processing, and the recognition rate can reach 88.8%. Finally, the in-memory computing device that integrates brain-like analog neuromorphic computing and digital logic computing was prepared, and its functions in low-power storage, brain-like neuromorphic computing, and in-situ Boolean logic computing were studied. In terms of logic calculation, the device can implement Boolean logic gates such as IMP, FALSE, and NAND by the combination and transitions of storage states, laying the foundation for its application in a multi-data type integrated in-memory computing system. This work provides effective guidelines for the development of ALD based devices for low power consumption in-memory computing system.