An Electro-Photonic System for Accelerating Deep Neural Networks

Demirkiran, Cansu; Eris, Furkan; Wang, Gongyu; Elmhurst, Jonathan; Moore, Nick; Harris, Nicholas C.; Basumallik, Ayon; Reddi, Vijay Janapa; Joshi, Ajay; Bunandar, Darius

doi:10.48550/arxiv.2109.01126

Cited by 7 publications

(13 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This blueprint provided us a promising way to fully exploit the advantage of photonics for accelerating computation in system level. Besides, Demirkiran et al proposed an electro-photonic system for accelerating DNNs in system-level perspective [163].…”

Section: Discussion and Outlookmentioning

confidence: 99%

See 1 more Smart Citation

Optoelectronic integrated circuits for analog optical computing: Development and challenge

et al. 2022

View full text Add to dashboard Cite

Over the past 2 decades, researches in artificial neural networks (ANNs) and deep learning have flourished and enabled the applications of artificial intelligence (AI) in image recognition, natural language processing, medical image analysis, molecular and material science, autopilot and so on. As the application scenarios for AI become more complex, massive perceptual data need to be processed in real-time. Thus, the traditional electronic integrated chips for executing the calculation of ANNs and deep learning algorithms are faced with higher requirements for computation speed and energy consumption. However, due to the unsustainability of Moore’s Law and the failure of the Dennard’s scaling rules, the growth of computing power of the traditional electronic integrated chips based on electronic transistors and von Neumann architecture could difficultly match the rapid growth of data volume. Enabled by silicon-based optoelectronics, analog optical computing can support sub-nanosecond delay and ∼fJ energy consumption efficiency, and provide an alternative method to further greatly improve computing resources and to accelerate deep learning tasks. In Chapter 1, the challenges of electronic computing technologies are briefly explained, and potential solutions including analog optical computing are introduced. Then, separated by four photonic platforms, including coherent integration platform, incoherent integration platform, space-propagation optical platform, and optical fiber platform, the recent important research progresses in analog optical computing are outlined in Chapter 2. Then, the nonlinearity and training algorithm for analog optical computing are summarized and discussed in Chapter 3. In Chapter 4, the prospects and challenges of analog optical computing are pointed out.

show abstract

Section: Discussion and Outlookmentioning

confidence: 99%

“…(D) The schematic of the space-efficient optical integrated diffractive neural networks. (B) Reprinted from Ref [163]. with permission from arXiv preprint.…”

mentioning

confidence: 99%

Optoelectronic integrated circuits for analog optical computing: Development and challenge

et al. 2022

View full text Add to dashboard Cite

show abstract

“…A number of companies have designed custom edge computing application-specific integrated circuits (ASICs) with reduced SWaP (7,35), but these ASICs are hampered by the same energy and bandwidth constraints as larger CMOS processors. Analog accelerators, such as memristive crossbar arrays and meshes of photonic interferometers, hold promise for lowering the power consumption of neural networks compared with electronic counterparts, but existing commercial demonstrations still consume watts of power (8,36).…”

Section: Discussionmentioning

confidence: 99%

“…However, the increasing computational demands of the most powerful DNNs limit deployment on low-power devices, such as smartphones and sensorsand this trend is accelerated by the simultaneous move toward Internet of Things (IoT) devices. Numerous efforts are underway to lower power consumption, but a fundamental bottleneck remains because of energy consumption in matrix algebra (5), even for analog approaches including neuromorphic (6), analog memory (7), and photonic meshes (8). In all these approaches, memory access and multiplyaccumulate (MAC) functions remain a stubborn bottleneck near 1 pJ per MAC (5,(9)(10)(11)(12).…”

mentioning

confidence: 99%

Delocalized photonic deep learning on the internet’s edge

Sludds

Bandyopadhyay

Chen

et al. 2022

Science

Self Cite

View full text Add to dashboard Cite

Advanced machine learning models are currently impossible to run on edge devices such as smart sensors and unmanned aerial vehicles owing to constraints on power, processing, and memory. We introduce an approach to machine learning inference based on delocalized analog processing across networks. In this approach, named Netcast, cloud-based “smart transceivers” stream weight data to edge devices, enabling ultraefficient photonic inference. We demonstrate image recognition at ultralow optical energy of 40 attojoules per multiply (<1 photon per multiply) at 98.8% (93%) classification accuracy. We reproduce this performance in a Boston-area field trial over 86 kilometers of deployed optical fiber, wavelength multiplexed over 3 terahertz of optical bandwidth. Netcast allows milliwatt-class edge devices with minimal memory and processing to compute at teraFLOPS rates reserved for high-power (>100 watts) cloud computers.

show abstract

“…On one hand, if the low-precision ADCs are used directly, the results of optical computing will be damaged serverly beacuse ADCs can not convert the signals beyond the precision rightly. On the other hand, the energy consumption of ADCs accounts for about half of the whole electro-photonic computing system [2], and the energy consumption of ADCs increases exponentially with the precision. Therefore, if the high-precision ADCs are used, the loss of optical computing results is acceptable, but the energy consumption of the entire system will increase significantly.…”

Section: Introductionmentioning

confidence: 99%

An energy-efficient quantized inference framework for electro-photonic computing system

Lin

Zhang

et al. 2023

International Conference on Optical and Photonic Engineering (icOPEN 2022)

View full text Add to dashboard Cite

Deep Learning is developing rapidly and has made breakthroughs in various fields. As a result, the number of parameters in deep neural networks(DNNs) is scaling about 5× rate of Moore's Law. To meet the computing power needs of DNNs, new computing architectures have been proposed, and the electro-photonic computing system is one of them. With the physical characteristics of photons, the electro-photonic computing system has demonstrated great potential in improving the efficiency of convolution calculation. However, there are still much spcace for improvement in the energy efficiency of the electro-photonic computing system. And the high energy consumption led by the high-precision analogto-digital converters(ADCs) is an important factor hindering the energy efficiency. ADCs dominate about 50% energy consumption of the entire system, and they will increase exponentially with the precision. But if the low-precision ADCs are used directly, they will destroy the computing accuracy in the convolution calculation and reduce the inference performance of DNNs. In this paper, we propose an energy-efficient quantized inference framework for electro-photonic computing system to balance the energy consumption of ADCs and the DNNs inference performance, which includes a multi-scale quantization scheme based on per-slice granularity and width variable optical matrix. The proposed framework can reduce the energy consumption of the system by using low-precision ADCs, and ensure the inference performance of the DNNs at the same time. The experiments are carried out in MobileNet-v2, ResNet50 and Vgg16 respectively. The experimental results show that the proposed quantized inference framework can reduce the precision of ADCs by 7 bits and compared with using the low-precision ADCs directly, it improves the model accuracy of 40.5%, 70.2% and 65.4% respectively , which is close to the full precision model.

show abstract

An Electro-Photonic System for Accelerating Deep Neural Networks

Cited by 7 publications

References 0 publications

Optoelectronic integrated circuits for analog optical computing: Development and challenge

Optoelectronic integrated circuits for analog optical computing: Development and challenge

Delocalized photonic deep learning on the internet’s edge

An energy-efficient quantized inference framework for electro-photonic computing system

Contact Info

Product

Resources

About