A Survey on Silicon Photonics for Deep Learning

Sunny, Febin; Taheri, Ebadollah; Nikdast, Mahdi; Pasricha, Sudeep

doi:10.1145/3459009

Cited by 62 publications

(19 citation statements)

References 173 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While several high-performance AI accelerators based on coherent SP-NNs have been recently proposed [2,10,13], [14] showed that the inferencing accuracy of SP-NNs can drop by up to 70% due to fabrication-process variations and thermal crosstalk. In addition to these variations, the work in [10] explored the impact of optical loss non-uniformity among MZIs and showed SP-NN performance degradation.…”

Section: Related Prior Workmentioning

confidence: 99%

“…In (4), 𝑁 𝑀𝑍 𝐼 is the total number of MZIs in the OIU in layer 𝐿 𝑚 and 𝑃 is the input optical power. Moreover, 𝑋 𝑚 𝑗 𝑀𝑍 𝐼 (𝜌) can be calculated using (2) and is the coherent crosstalk on the output of layer 𝐿 𝑚 originating in MZI 𝑗 in the OIU. Also, 𝜌 is the optical phase of the crosstalk signal.…”

Section: Layer-level Compact Modelsmentioning

confidence: 99%

“…However, contemporary electronic deep-learning inference accelerators have shown relatively low energy-efficiency and have been unable to keep up with the performance demands from emerging deep learning applications [1]. To overcome these bottlenecks, novel accelerators tailored towards artificial intelligence (AI) applications are on the rise, among which integrated silicon-photonic neural networks (SP-NNs) have attracted much attention with a promise of lightspeed communication and computation using optical interconnects and silicon photonic devices [2]. For instance, for computationally expensive multiply-and-accumulate operations, optical computing can achieve up to a 1000× better energy-efficiency footprint compared to electronic accelerators [3].…”

Section: Introductionmentioning

confidence: 99%

“…SP-NNs use silicon photonic devices-e.g., Mach-Zehnder interferometers (MZIs)-to realize matrix-vector multiplication with a computational complexity of 𝑂 (1) [1]. Among different SP-NN implementations, coherent SP-NNs, which operate on a single wavelength, have an inherent advantage over noncoherent SP-NNs that require power-hungry wavelength-conversion steps and multiple wavelength sources [2]. Fig.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

LoCI: An Analysis of the Impact of Optical Loss and Crosstalk Noise in Integrated Silicon-Photonic Neural Networks

Shafiee¹,

Banerjee²,

Chakrabarty³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Compared to electronic accelerators, integrated silicon-photonic neural networks (SP-NNs) promise higher speed and energy efficiency for emerging artificial-intelligence applications. However, a hitherto overlooked problem in SP-NNs is that the underlying silicon photonic devices suffer from intrinsic optical loss and crosstalk noise, the impact of which accumulates as the network scales up.Leveraging precise device-level models, this paper presents the first comprehensive and systematic optical loss and crosstalk modeling framework for SP-NNs. For an SP-NN case study with two hidden layers and 1380 tunable parameters, we show a catastrophic 84% drop in inferencing accuracy due to optical loss and crosstalk noise.

show abstract

Section: Related Prior Workmentioning

confidence: 99%

Section: Layer-level Compact Modelsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

LoCI: An Analysis of the Impact of Optical Loss and Crosstalk Noise in Integrated Silicon-Photonic Neural Networks

Shafiee¹,

Banerjee²,

Chakrabarty³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…More advanced structures have been shown in the literature where both photonic accelerators (devices specialized to perform a single function and which are interfaced with electronic computer) [178,179] or photonic integrated neural networks [180][181][182][183] with extremely high capacity and speed [184,185] are demonstrated.…”

Section: Neuromorphic Photonicsmentioning

confidence: 99%

Thirty Years in Silicon Photonics: A Personal View

Pavesi

2021

Front. Phys.

View full text Add to dashboard Cite

Silicon Photonics, the technology where optical devices are fabricated by the mainstream microelectronic processing technology, was proposed almost 30 years ago. I joined this research field at its start. Initially, I concentrated on the main issue of the lack of a silicon laser. Room temperature visible emission from porous silicon first, and from silicon nanocrystals then, showed that optical gain is possible in low-dimensional silicon, but it is severely counterbalanced by nonlinear losses due to free carriers. Then, most of my research focus was on systems where photons show novel features such as Zener tunneling or Anderson localization. Here, the game was to engineer suitable dielectric environments (e.g., one-dimensional photonic crystals or waveguide-based microring resonators) to control photon propagation. Applications of low-dimensional silicon raised up in sensing (e.g., gas-sensing or bio-sensing) and photovoltaics. Interestingly, microring resonators emerged as the fundamental device for integrated photonic circuit since they allow studying the hermitian and non-hermitian physics of light propagation as well as demonstrating on-chip heavily integrated optical networks for reconfigurable switching applications or neural networks for optical signal processing. Finally, I witnessed the emergence of quantum photonic devices, where linear and nonlinear optical effects generate quantum states of light. Here, quantum random number generators or heralded single-photon sources are enabled by silicon photonics. All these developments are discussed in this review by following my own research path.

show abstract

Silicon Photonics for Future Computing Systems

Shafiee

Pasricha

Nikdast

2022

Wiley Encyclopedia of Electrical and Electronics Engineering

View full text Add to dashboard Cite

The primary goal of this article is to provide an overview of silicon photonics technology and its applications in the design and improvement of current and future computing systems. We start by reviewing silicon photonics technology and introducing some of its benefits and challenges as well as providing some background on it. Next, we introduce some fundamental silicon photonic components in the design of silicon photonic integrated circuits (PICs) and optical interconnect for computing systems as well as their operating principles and applications. These components can be active, such as photodetectors and optical modulators, or passive, such as silicon‐on‐insulator (SOI) waveguides. Subsequently, we discuss the application of silicon photonics to improve the communication and computation infrastructure in future computing systems, while reviewing the state‐of‐the‐art and some design and implementation challenges. Finally, we discuss several research opportunities to push forward the application of silicon photonics in the design of future computing systems.

show abstract

A Survey on Silicon Photonics for Deep Learning

Cited by 62 publications

References 173 publications

LoCI: An Analysis of the Impact of Optical Loss and Crosstalk Noise in Integrated Silicon-Photonic Neural Networks

LoCI: An Analysis of the Impact of Optical Loss and Crosstalk Noise in Integrated Silicon-Photonic Neural Networks

Thirty Years in Silicon Photonics: A Personal View

Silicon Photonics for Future Computing Systems

Contact Info

Product

Resources

About