Delocalized photonic deep learning on the internet’s edge

Sludds, Alexander; Bandyopadhyay, S.; Chen, Zaijun; Zhong, Zhizhen; Cochrane, Jared; Bernstein, Liane; Bunandar, Darius; Dixon, Paul; Hamilton, Scott A.; Streshinsky, Matthew; Novack, Ari; Baehr‐Jones, Tom; Hochberg, Michael; Ghobadi, Manya; Hamerly, Ryan; Englund, Dirk

doi:10.1126/science.abq8271

Cited by 59 publications

(24 citation statements)

References 75 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…From Section IV-A, the stochastic multiplication bit-streams generated by OSMs are guided to a PCA, where they are accumulated to generate a binary output value equivalent to the VDP result. Our PCA is inspired from the time integrating receiver (TIR) design from [37] and the photodetector-based optical-pulse accumulator design from [38]. A PCA circuit, shown in Fig 4(c), has two stages: (i) a stochastic-to-analog conversion stage; (ii) an analog-to-binary conversion stage.…”

Section: Photo Charge Accumulator (Pca)mentioning

confidence: 99%

SCONNA: A Stochastic Computing Based Optical Accelerator for Ultra-Fast, Energy-Efficient Inference of Integer-Quantized CNNs

Vatsavai¹,

Karempudi²,

Thakkar³

et al. 2023

Preprint

View full text Add to dashboard Cite

Convolutional Neural Networks (CNNs) are used extensively for artificial intelligence applications due to their record-breaking accuracy. For efficient and swift hardware-based acceleration, CNNs are typically quantized to have integer input/weight parameters. The acceleration of a CNN inference task uses convolution operations that are typically transformed into vector-dot-product (VDP) operations. Several photonic microring resonators (MRRs) based hardware architectures have been proposed to accelerate integer-quantized CNNs with remarkably higher throughput and energy efficiency compared to their electronic counterparts. However, the existing photonic MRRbased analog accelerators exhibit a very strong trade-off between the achievable input/weight precision and VDP operation size, which severely restricts their achievable VDP operation size for the quantized input/weight precision of 4 bits and higher. The restricted VDP operation size ultimately suppresses computing throughput to severely diminish the achievable performance benefits. To address this shortcoming, we for the first time present a merger of stochastic computing and MRR-based CNN accelerators. To leverage the innate precision flexibility of stochastic computing, we invent an MRR-based optical stochastic multiplier (OSM). We employ multiple OSMs in a cascaded manner using dense wavelength division multiplexing, to forge a novel Stochastic Computing based Optical Neural Network Accelerator (SCONNA). SCONNA achieves significantly high throughput and energy efficiency for accelerating inferences of high-precision quantized CNNs. Our evaluation for the inference of four modern CNNs at 8-bit input/weight precision indicates that SCONNA provides improvements of up to 66.5×, 90×, and 91× in frames-per-second (FPS), FPS/W and FPS/W/mm 2 , respectively, on average over two photonic MRR-based analog CNN accelerators from prior work, with Top-1 accuracy drop of only up to 0.4% for large CNNs and up to 1.5% for small CNNs. We developed a transaction-level, event-driven python-based simulator for the evaluation of SCONNA and other accelerators (https://github.com/uky-UCAT/SC ONN SIM.git).

show abstract

Section: Photo Charge Accumulator (Pca)mentioning

confidence: 99%

SCONNA: A Stochastic Computing Based Optical Accelerator for Ultra-Fast, Energy-Efficient Inference of Integer-Quantized CNNs

Vatsavai¹,

Karempudi²,

Thakkar³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Recently, we proposed Netcast, an optical serverclient protocol for edge computing based on wavelength-division multiplexing (WDM), difference detection and integration, and optical weight delivery. 21,22 This protocol splits the computation into two components: a WDM modulator array constituting the "weight server" (Fig. 4(a)), connected via an optical link to a SWaPconstrained client (Fig.…”

Section: Netcast: Wdm-powered Photonic Edge Computingmentioning

confidence: 99%

Multiplexing methods for scaling up photonic logic

Hamerly

Bandyopadhyay

Sludds

et al. 2023

AI and Optical Data Sciences IV

Self Cite

View full text Add to dashboard Cite

Existing photonic matrix processers are too small to tackle relevant problems. Here, I review our group's recent work on scaling up analog photonic platforms. This work includes iterative advances to old approaches (accurate methods to calibrate MZI meshes), experimental demonstrations of recent proposals (a VCSEL array-based coherent detection ONN and a single-shot ONN based on reconfigurable free-space optical fan-out and weighting) and entirely new architectures (WDM-powered and RF-photonic fiber circuits for edge computing). The lessons learned from studying this diverse array of approaches helps inform the future development of photonic hardware for computation.

show abstract

“…For example, the characteristic features of edge devices are low power consumption, high processing power, and large memory capacities. These features are needed to carry out advanced machine learning at the edge, as discussed by Sludds et al [ 40 ]. To counter the lack of the needed edge resources, Sludds et al devised a method (still using the resource-constrained edge devices) to process data across networks and reported teraflops (10 12 floating point operations per second or FLOPS) computing rates capitalizing on energy efficiency and sensitivity of photonics devices.…”

Section: Motivationmentioning

confidence: 99%

The Concept of a Quantum Edge Simulator: Edge Computing and Sensing in the Quantum Era

Passian

Buchs

Seck

et al. 2022

Sensors

View full text Add to dashboard Cite

Sensors, enabling observations across vast spatial, spectral, and temporal scales, are major data generators for information technology (IT). Processing, storing, and communicating this ever-growing amount of data pose challenges for the current IT infrastructure. Edge computing—an emerging paradigm to overcome the shortcomings of cloud-based computing—could address these challenges. Furthermore, emerging technologies such as quantum computing, quantum sensing, and quantum communications have the potential to fill the performance gaps left by their classical counterparts. Here, we present the concept of an edge quantum computing (EQC) simulator—a platform for designing the next generation of edge computing applications. An EQC simulator is envisioned to integrate elements from both quantum technologies and edge computing to allow studies of quantum edge applications. The presented concept is motivated by the increasing demand for more sensitive and precise sensors that can operate faster at lower power consumption, generating both larger and denser datasets. These demands may be fulfilled with edge quantum sensor networks. Envisioning the EQC era, we present our view on how such a scenario may be amenable to quantification and design. Given the cost and complexity of quantum systems, constructing physical prototypes to explore design and optimization spaces is not sustainable, necessitating EQC infrastructure and component simulators to aid in co-design. We discuss what such a simulator may entail and possible use cases that invoke quantum computing at the edge integrated with new sensor infrastructures.

show abstract

Delocalized photonic deep learning on the internet’s edge

Cited by 59 publications

References 75 publications

SCONNA: A Stochastic Computing Based Optical Accelerator for Ultra-Fast, Energy-Efficient Inference of Integer-Quantized CNNs

SCONNA: A Stochastic Computing Based Optical Accelerator for Ultra-Fast, Energy-Efficient Inference of Integer-Quantized CNNs

Multiplexing methods for scaling up photonic logic

The Concept of a Quantum Edge Simulator: Edge Computing and Sensing in the Quantum Era

Contact Info

Product

Resources

About