Computation at the speed of light: metamaterials for all-optical calculations and neural networks

Advanced Photonics, Journal Year: 2022, Volume and Issue: 4(06)

Published: Dec. 21, 2022

The explosion in the amount of information that is being processed prompting need for new computing systems beyond existing electronic computers. Photonic emerging as an attractive alternative due to performing calculations at speed light, change massive parallelism, and also extremely low energy consumption. We review physical implementation basic optical calculations, such differentiation integration, using metamaterials, introduce realization all-optical artificial neural networks. start with concise introductions mathematical principles behind computation methods present advantages, current problems be overcome, potential future directions field. expect our will useful both novice experienced researchers field platforms metamaterials.

Language: Английский

---

Artificial Intelligence and Advanced Materials

Advanced Materials, Journal Year: 2022, Volume and Issue: 35(23)

Published: Dec. 23, 2022

Abstract Artificial intelligence (AI) is gaining strength, and materials science can both contribute to profit from it. In a simultaneous progress race, new materials, systems, processes be devised optimized thanks machine learning (ML) techniques, such turned into innovative computing platforms. Future scientists will understanding how ML boost the conception of advanced materials. This review covers aspects computation fundamentals directions taken repercussions produced by account for origins, procedures, applications AI. its methods are reviewed provide basic knowledge implementation potential. The systems used implement AI with electric charges finding serious competition other information‐carrying processing agents. impact these techniques have on inception so deep that paradigm developing where implicit being mined conceive functions instead found How far this trend carried hard fathom, as exemplified power discover unheard or physical laws buried in data.

Language: Английский

---

Hardware-algorithm collaborative computing with photonic spiking neuron chip based on an integrated Fabry–Perot laser with a saturable absorber

Optica, Journal Year: 2022, Volume and Issue: 10(2), P. 162 - 162

Published: Dec. 12, 2022

Photonic neuromorphic computing has emerged as a promising approach to building low-latency and energy-efficient non-von Neuman system. A photonic spiking neural network (PSNN) exploits brain-like spatiotemporal processing realize high-performance computing. However, the nonlinear computation of PSNN remains significant challenge. Here, we propose fabricate neuron chip based on an integrated Fabry–Perot laser with saturable absorber (FP-SA). The neuron-like dynamics including temporal integration, threshold spike generation, refractory period, inhibitory behavior cascadability are experimentally demonstrated, which offers indispensable fundamental block construct hardware. Furthermore, time-multiplexed encoding functional far beyond hardware integration scale limit. PSNNs single/cascaded neurons demonstrated hardware-algorithm collaborative computing, showing capability perform classification tasks supervised learning algorithm, paves way for multilayer that can handle complex tasks.

Language: Английский

---

Electrically programmable phase-change photonic memory for optical neural networks with nanoseconds in situ training capability

Advanced Photonics, Journal Year: 2023, Volume and Issue: 5(04)

Published: July 18, 2023

Optical neural networks (ONNs), enabling low latency and high parallel data processing without electromagnetic interference, have become a viable player for fast energy-efficient calculation to meet the increasing demand hash rate. Photonic memories employing nonvolatile phase-change materials could achieve zero static power consumption, thermal cross talk, large-scale, high-energy-efficient photonic networks. Nevertheless, switching speed dynamic energy consumption of material-based make them inapplicable in situ training. Here, by integrating patch phase change thin film with PIN-diode-embedded microring resonator, bifunctional memory both 5-bit storage nanoseconds volatile modulation was demonstrated. For first time, concept is presented electrically programmable material-driven integrated nanosecond allow training ONNs. ONNs an optical convolution kernel constructed our theoretically achieved accuracy predictions higher than 95% when tested MNIST handwritten digit database. This provides feasible solution constructing large-scale high-speed capability.

Language: Английский

---

Silicon-based optoelectronics for general-purpose matrix computation: a review

Advanced Photonics, Journal Year: 2022, Volume and Issue: 4(04)

Published: July 6, 2022

Conventional electronic processors, which are the mainstream and almost invincible hardware for computation, approaching their limits in both computational power energy efficiency, especially large-scale matrix computation. By combining electronic, photonic, optoelectronic devices circuits together, silicon-based computation has been demonstrating great capabilities feasibilities. Matrix is one of few general-purpose computations that have potential to exceed performance digital logic power, latency. Moreover, processors also suffer from tremendous consumption transceiver during high-capacity data interconnections. We review recent progress photonic including matrix-vector multiplication, convolution, multiply–accumulate operations artificial neural networks, quantum information processing, combinatorial optimization, compressed sensing, with particular attention paid consumption. summarize advantages interconnections photonic-electronic integration over conventional optical computing processors. Looking toward future computations, we believe optoelectronics a promising comprehensive platform disruptively improving post-Moore’s law era.

Language: Английский

---