China
Africa
Explore chapters and articles related to this topic
Hybrid Phase-Change Nanophotonic Circuits
Published in Klaus D. Sattler, st Century Nanoscience – A Handbook, 2020
Johannes Feldmann, Wolfram Pernice
A completely different approach to traditional computers are brain-inspired computing techniques called neuromorphic computers implementing the idea of mimicking the highly efficient and parallel processing of huge amounts of data accomplished by biological brains. Individual neurons connected by synapses process the information and are especially suitable for classification tasks, pattern, and speech recognition (Lane et al. 2017). Due to the serial approach used by traditional computers based on the von Neumann architecture that process one command at a time, these tasks cannot be efficiently solved (Preissl et al. 2012). Neuromorphic computers implementing the massively parallel processing of brains can instead efficiently deal with such problems and outperform conventional computers in terms of speed and energy efficiency by orders of magnitude.
Oxide Memristor and Applications
Published in Simon Deleonibus, Emerging Devices for Low-Power and High-Performance Nanosystems, 2018
Mingyi Rao, Rivu Midya, J. Joshua Yang
The paradigm of computing that takes inspiration from the human brain is generally referred to as “neuromorphic computing.” The term was coined by Carver Mead in the 1990s [71]. Ever since that time, researchers have been trying to implement such ideas in hardware. Such efforts were originally limited to the CMOS, which is inefficient in the long run because it is not optimized for this purpose. There has been renewed interest in this field following the advent of memristors [2]. Memristors can be scaled down to form nanometer-level arrays that are very attractive because they can be used to successfully realize the parallelism and the low power consumption of the human brain. Moreover, memristor arrays can be very dense, offering a chance to mimic hardware connectivity that can be rivalled by the connectivity observed in the brain.
Integrated Photonics for Artificial Intelligence Applications
Published in Sandeep Saini, Kusum Lata, G.R. Sinha, VLSI and Hardware Implementations Using Modern Machine Learning Methods, 2021
Ankur Saharia, Kamal Kishor Choure, Nitesh Mudgal, Rahul Pandey, Dinesh Bhatia, Manish Tiwari, Ghanshyam Singh
The human brain is a sophisticated and complex system that constitutes hundreds of billions of interconnected neurons and synapses, respectively. Neurons are the most important and basic element of the human brain, and they are connected through synapses. The information signal between these neurons has the form of a spike. Human brain architecture can provide unmatched high-speed parallel computing, super intelligence, self-learning, and an upgradation system with very low power consumption. The functionality of the human brain can be shown by spiking neural network (SNN) [1]. The signal between the neurons via synapses is in the form of a spike. Drawing inspiration from the architecture of the human brain and its unmatched functionality, the artificial neural network (ANN) was developed. The word “artificial neural” refers to the artificial neuron created with the same functionality as the brain neuron. The computation process performed with help from these brain-inspired artificial neurons is referred to as “neuromorphic computing”. Today, ANNs are needed to process and compute large sets of data [2]. The era of ANNs started back in 1943 when Walter Pitts developed a mathematical model based on the nerve cell of the brain and its information processing system [3]. In recent years, the ANN has been extensively applied in artificial intelligence, machine learning [4], and computing. The ANN is considered to mimic the architecture of the human brain. In von Neumann’s model computer, there is physical separation between computing and signal processing memory for information, which limits computing efficiency while parallel processing a large set of signals. On the other hand, to get high computing efficiency ANNs have massive accumulate computation (MAC) for parallel processing signals [5].
Pseudo-transistors for emerging neuromorphic electronics
Published in Science and Technology of Advanced Materials, 2023
Jingwei Fu, Jie Wang, Xiang He, Jianyu Ming, Le Wang, Yiru Wang, He Shao, Chaoyue Zheng, Linghai Xie, Haifeng Ling
Neuromorphic engineering, also known as neuromorphic computing, is a concept developed by Carver Mead in the late 1980s [1]. This describes the use of electronic analogue circuits to mimic neurobiological architectures present in the nervous system. As the hardware realization of artificial intelligence, neuromorphic electronics are attracting increased attention [2–5]. Neuromorphic electronic devices are the building blocks which perform learning and computing processes by mimicking the functions of biological neurons and synapses. The human brain consists of approximately 1011 neurons and 1015 synapses, and the energy consumption (~20 W) is even far less than supercomputers (~107 W) [6–9]. So far, large-scale neuromorphic chip has been successfully developed with complementary metal-oxide-semiconductor (CMOS) technology [10,11], such as the Loihi announced by Intel Corporation in 2017 [12]. However, silicon neuron requires a large number of MOS transistors [13]. Strategies for building future large-scale neuromorphic circuits rely on the exploration of memory component at the single-device level, such as artificial synapses. Emerging electronic devices with tunable memory property are needed.
Recent progress in neuromorphic and memory devices based on graphdiyne
Published in Science and Technology of Advanced Materials, 2023
Zhi-Cheng Zhang, Xu-Dong Chen, Tong-Bu Lu
Inspired by the human brain, neuromorphic devices, including artificial synapses and neurons, have been proposed to implement neuromorphic computing. Especially, artificial synapses inspired by biological synapses have attracted great attentions in recent years. Various synaptic devices have been proposed with a structure of two-terminal memristors and three-terminal transistors [7,11,42–57]. 2D materials and heterostructures present great advantages for the development of neuromorphic devices due to their atomical thickness, atomically sharp interface and free of surface dangling bonds [15]. Especially, the excellent physical and chemical properties of GDY, including the ordered pore structure, direct bandgap, broadband light absorption, strong charge trapping, low ion diffusion barrier, and excellent flexibility, are favorable for applications in neuromorphic devices. For instance, the abundant defects in GDY introduced during the synthesis process endow GDY with great potential in neuromorphic devices since the long-term trapping and slow release of charges at these charge-trapping sites facilitate the emulation of various synaptic plasticity [119]. The density of defects can also be controlled by defect engineering and post-treatment. In addition, the strong and broadband absorption of GDY make it a promising candidate in optoelectronic synapses, in which the photo-generated charges can be effectively trapped in GDY to emulate the synaptic response. We believe all these features are crucial for applications of GDY in neuromorphic devices.
Hybrid oxide brain-inspired neuromorphic devices for hardware implementation of artificial intelligence
Published in Science and Technology of Advanced Materials, 2021
Jingrui Wang, Xia Zhuge, Fei Zhuge
Although significant progress has been made in the development of memristive synapses, the relevant research field is still in its incipient stage. The device stability and reliability are far from meeting the standard of practical applications in neuromorphic computing. It remains a big challenge to construct a highly effective artificial neural network with relatively low density, e.g. k bits, based on oxide memristors. In most cases, memristors are used to mimic deep neural networks. Spiking neural networks mainly based on STDP, that reflect the actual information processing style in the human brain, are seldom investigated [142]. Anyway, oxide-based hybrid structures provide a promising prospect for the application of oxide memristors to neuromorphic computing and therefore hardware implementation of artificial intelligence.