Chip最新文献

In-memory computing has carried out calculations in situ within each memory unit and its main power consumption comes from data writing and erasing. Further improvements in the energy efficiency of in-memory computing require memory devices with sub-femto-Joule energy consumption. Floating gate memory devices based on two-dimensional (2D) material heterostructures have outstanding characteristics such as non-volatility, multi-bit storage, and low operation energy, suitable for application in in-memory computing chips. Here, we report a floating gate memory device based on a WSe₂/h-BN/Multilayer-graphene/h-BN heterostructure, the energy consumption of which is in sub-femto Joule (0.6 fJ) per operation for program/erase, and the read power consumption is in the tens of femto Watt (60 fW) range. We show a Hopfield neural network composed of WSe₂/h-BN/Multilayer-graphene/h-BN heterostructure floating gate memory devices, which can recall the original patterns from incorrect patterns. These results shed light on the development of future compact and energy-efficient hardware for in-memory computing systems.

Conventional von Neumann architecture faces many challenges in dealing with data-intensive artificial intelligence tasks efficiently due to huge amounts of data movement between physically separated data computing and storage units. Novel computing-in-memory (CIM) architecture implements data processing and storage in the same place, and thus can be much more energy-efficient than state-of-the-art von Neumann architecture. Compared with their counterparts, resistive random-access memory (RRAM)-based CIM systems could consume much less power and area when processing the same amount of data. In this paper, we first introduce the principles and challenges related to RRAM-based CIM systems. Then, recent works on the circuit and macro levels of RRAM-CIM systems will be reviewed to highlight the trends and challenges in this field.

Graphene on top of semiconductor builds an emerging highly sensitive photodetector with internal gain. Owing to the graphene/semiconductor interface junction, one kind of photo-excited carriers are drifted to graphene and the other carriers remain in the semiconductor. The decisive factor for the gain is the localization extent of the non-transporting carriers. Several localization strategies such as Schottky barrier regulation, introducing localized states, quantum dot confinement, and double heterojunction design are reviewed. Despite the high sensitivity, the accompanying persistent photocurrent limits the response speed. The long-wavelength light acceleration and the back-gate voltage acceleration methods are utilized to effectively eliminate the persistent photocurrent.

Quantum entanglement, as the strictly non-classical phenomenon, is the kernel of quantum computing and quantum simulation, and has been widely applied ranging from fundamental tests of quantum physics to quantum information processing. Meanwhile, the topological phase is found inherently capable of protecting physical fields from unavoidable fabrication-induced disorder, which inspires the potential application of topological protection to quantum states. Here, we present the experimental demonstration of topologically protected quantum entangled states on a photonic chip. The process tomography shows that quantum entanglement can be well preserved by the topological states even when the chip material introduces disorder and relative polarization rotation in phase space. Our work links the fields of materials, topological science and quantum physics, opening the door to wide applications of topological enhancement in quantum regime.

A silicon quantum photonic circuit was proposed and realized for the generation and the dynamic manipulation of telecom-band frequency-degenerate polarization entangled Bell states. Frequency degenerate biphoton states were generated in four silicon waveguides by spontaneous four wave mixing. They were transformed to polarization entangled Bell states through on-chip quantum interference and quantum superposition, and then coupled to optical fibers. The property of polarization entanglement in generated photon pairs was demonstrated by two-photon interference under two non-orthogonal polarization bases. The output state could be dynamically switched between two Bell states, which was demonstrated by the simplified Bell state measurement. The experiment results indicated that the manipulation speed supported a modulation rate of several tens kHz, showing its potential on applications of quantum communication and quantum information processing requiring Bell state encoding and dynamic control.

Avalanche-photodiode-based near-infrared single-photon detectors have seen rapid development in the last two decades because of their enormous internal gain, high sensitivity, fast response, small volume, and ease of integration. The InGaAs/InP near-infrared single-photon detector is the most widely used avalanche diode at present. Its device performance is still being continuously improved through the optimization of device structure and external quenching circuits. This paper analyzes the latest development and application of these InGaAs/InP photodiodes, then briefly reviews other near-infrared single-photon detection technologies based on new materials and new mechanisms.

Optical neural network (ONNs) are emerging as attractive proposals for machine-learning applications. However, the stability of ONNs decreases with the circuit depth, limiting the scalability of ONNs for practical uses. Here we demonstrate how to compress the circuit depth to scale only logarithmically in terms of the dimension of the data, leading to an exponential gain in terms of noise robustness. Our low-depth (LD)-ONN is based on an architecture, called Optical CompuTing Of dot-Product UnitS (OCTOPUS), which can also be applied individually as a linear perceptron for solving classification problems. We present both numerical and theoretical evidence showing that LD-ONN can exhibit a significant improvement on robustness, compared with previous ONN proposals based on singular-value decomposition.