Deep learning with coherent VCSEL neural networks

Chen, Zaijun; Sludds, Alexander; Davis, Ronald; Christen, Ian; Bernstein, Liane; Ateshian, Lamia; Heuser, Tobias; Heermeier, Niels; Lott, James A.; Reitzenstein, Stephan; Hamerly, Ryan; Englund, Dirk

doi:10.1038/s41566-023-01233-w

Cited by 52 publications

(7 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…With the rapid advancement of long-distance sensing technology in autonomous driving, advanced AI computations like ChatGPT demanding high data capacity and speed, and the swift growth of intelligent and quantum technology applications such as deep learning based on VCSELs 25 , 26 , these technological domains inevitably face a critical shared challenge: the issue of energy consumption. Whether it’s the battery consumption of mobile terminals or the energy consumption of data centers, VCSELs, as light sources, constitute a significant part of the energy drain.…”

Section: Introductionmentioning

confidence: 99%

Multi-junction cascaded vertical-cavity surface-emitting laser with a high power conversion efficiency of 74%

Xiao,

Wang,

Liu

et al. 2024

Light Sci Appl

View full text Add to dashboard Cite

High electro-optical conversion efficiency is one of the most distinctive features of semiconductor lasers as compared to other types of lasers. Its further increase remains a significant objective. Further enhancing the efficiency of edge-emitting lasers (EEL), which represent the highest efficiency among semiconductor lasers at present, is challenging. The efficiency of vertical cavity surface emitting lasers (VCSELs) has always been relatively low compared to EEL. This paper, combining modeling with experiments, demonstrates the potential of multi-junction cascaded VCSELs to achieve high efficiency beyond that of EELs, our simulations show, that a 20-junction VCSEL can achieve an efficiency of more than 88% at room temperature. We fabricated VCSEL devices with different numbers of junctions and compared their energy efficiency. 15-junction VCSELs achieved a maximum efficiency of 74% at room temperature under nanosecond driving current, the corresponding differential quantum efficiency exceeds 1100%, being the largest electro-optical conversion efficiency and differential quantum efficiency reported until now for VCSELs.

show abstract

Section: Introductionmentioning

confidence: 99%

Multi-junction cascaded vertical-cavity surface-emitting laser with a high power conversion efficiency of 74%

Xiao,

Wang,

Liu

et al. 2024

Light Sci Appl

View full text Add to dashboard Cite

show abstract

“…Nevertheless, multistage cascaded modulation introduces higher loss for the weight elements downstream, limiting the scalability. Recently, the concept of coherent photonic MAC operation with direct input matrix/vector encoding and large-scale fan-out offers a pathway toward flexible and scalable photonic accelerators, ,, yet this concept has only been applied to low-precision ANN inference scenarios. Here, we propose a photonic matrix processing unit (MPU) based on coherent multidimensional analog photonic cores for matrix-vector multiplication, digital electronics for data storage and reconfigurability, and a fixed-point linear algebra library for error management.…”

Section: Introductionmentioning

confidence: 99%

Coherent General-Purpose Photonic Matrix Processor

Zhu,

Fardoost,

Vanani

et al. 2024

ACS Photonics

View full text Add to dashboard Cite

Matrix computations are at the heart of scientific computing, especially in models involving large-scale linear systems. As the scale and complexity of the problems grow, energy-efficient matrix computation becomes critical in these applications. Meanwhile, the advantages of miniaturizing conventional digital electronic processors, predicted by the Dennard scaling, diminish in post-Moore's law era. Analogue photonic devices based on passive and high-throughput interconnects are becoming promising alternatives as next-generation energy-efficient computing units. However, the limited reconfigurability and precision of an analogue photonic computing device make it unsuitable for scientific computing applications. Here, we report a general-purpose analogue photonic matrix processing unit (MPU) based on coherent analogue photonic cores, which perform signed multiplications, with reconfigurability and memory provided by digital electronics. Combined with error management strategies, our photonic MPU can perform tasks conventionally dominated by floating-point digital processors, elevating analog photonic-based platforms toward scientific computing applications. We have experimentally demonstrated its feasibilities in a range of computing tasks, including matrix multiplication and inversion as well as solving finite-difference partial differential equations.

show abstract

“…Similarly, electro-optic technologies such quantum resonant tunnelling structures [12,13], amongst others (see [14] for a review), have now also been proposed as suitable device platforms for brain-inspired computing hardware components. With one of the major benefits of photonics being the highly-reduced energy requirements, the recent demonstrations of photonic artificial neural networks (pANNs) [15][16][17][18][19] and reservoir computing systems [20][21][22][23] make an exciting case for photonics being the platform at the forefront of potentially sustainable neuromorphic computing. Importantly, the highly desirable energy efficiency of pANN systems is believed to be critical to the continued development of the currently booming Artificial Intelligence (AI) revolution.…”

Section: Introductionmentioning

confidence: 99%

VCSEL-based photonic spiking neural networks for ultrafast detection and tracking

Robertson,

Kirkland,

Di Caterina

et al. 2024

Neuromorph. Comput. Eng.

View full text Add to dashboard Cite

Inspired by efficient biological spike-based neural networks, we demonstrate for the first time the detection and tracking of target patterns in image and video inputs at high-speed rates with networks of multiple artificial spiking optical neurons. Using photonic systems of in-parallel spiking Vertical Cavity Surface Emitting Lasers (VCSELs), we demonstrate the implementation of multiple convolutional kernel operators which, in combination with optical spike signalling, enable the detection and tracking of target features in images/video feeds at an ultrafast photonic operation speed of 1 ns per pixel. Alongside a single layer optical Spiking Neural Network (SNN) demonstration, a multi-layer network of photonic (GHz-rate) spike-firing neurons is reported where the photonic system successfully tracks a large complex feature (Handwritten Digit 3). The consecutive photonic layers perform spike-enabled image reduction and convolution operations, and interact with a software-implemented SNN, that learns the feature patterns that best identify the target to provide a high detection efficiency even in the presence of a distractor feature. This work therefore highlights the effectiveness of combining neuromorphic photonic hardware and software SNNs, for efficient learning and ultrafast operation, thanks to the use of spiking light signals, towards tackling complex AI and computer vision problems.

show abstract

Deep learning with coherent VCSEL neural networks

Cited by 52 publications

References 46 publications

Multi-junction cascaded vertical-cavity surface-emitting laser with a high power conversion efficiency of 74%

Multi-junction cascaded vertical-cavity surface-emitting laser with a high power conversion efficiency of 74%

Coherent General-Purpose Photonic Matrix Processor

VCSEL-based photonic spiking neural networks for ultrafast detection and tracking

Contact Info

Product

Resources

About