Showing papers in "IEEE Journal of Selected Topics in Quantum Electronics in 2023"

Petabit-Scale Silicon Photonic Interconnects With Integrated Kerr Frequency Combs

Abstract: Silicon photonics holds significant promise in revolutionizing optical interconnects in data centers and high performance computers to enable scaling into the Pb/s package escape bandwidth regime while consuming orders of magnitude less energy per bit than current solutions. In this work, we review recent progress in silicon photonic interconnects leveraging chip-scale Kerr frequency comb sources and provide a comprehensive overview of massively scalable silicon photonic systems capable of capitalizing on the large number of wavelengths provided by such combs. We first consider the high-level architectural constraints and then proceed to detail the corresponding fundamental device designs supported by both simulated and experimental results. Furthermore, the majority of experimentally measured devices were fabricated in a commercial 300 mm foundry, showing a clear path to volume manufacturing. Finally, we present various system-level experiments which illustrate successful proof-of-principle operation, including flip-chip integration with a co-designed CMOS application-specific integrated circuit (ASIC) to realize a complete Kerr comb-driven electronic-photonic engine. These results provide a viable and appealing path towards future co-packaged silicon photonic interconnects with aggregate per-fiber bandwidth above 1 Tb/s, energy consumption below 1 pJ/bit, and areal bandwidth density greater than 5 Tb/s/mm ² .

Neuromorphic Computing Based on Wavelength-Division Multiplexing

Abstract: Optical neural networks (ONNs), or optical neuromorphic hardware accelerators, have the potential to dramatically enhance the computing power and energy efficiency of mainstream electronic processors, due to their ultralarge bandwidths of up to 10s of terahertz together with their analog architecture that avoids the need for reading and writing data back and forth. Different multiplexing techniques have been employed to demonstrate ONNs, amongst which wavelength division multiplexing (WDM) techniques make sufficient use of the unique advantages of optics in terms of broad bandwidths. Here, we review recent advances in WDM based ONNs, focusing on methods that use integrated microcombs to implement ONNs. We present results for human image processing using an optical convolution accelerator operating at 11 Tera operations per second. The open challenges and limitations of ONNs that need to be addressed for future applications are also discussed.

Universal Linear Optics Revisited: New Perspectives for Neuromorphic Computing With Silicon Photonics

Abstract: Reprogrammable optical meshes comprise a subject of heightened interest for the execution of linear transformations, having a significant impact in numerous applications that extend from the implementation of optical switches up to neuromorphic computing. Herein, we review the state-of-the-art approaches for the realization of unitary transformations and universal linear operators in the photonic domain and present our recent work in the field, that allows for fidelity restorable and low-loss optical circuitry with single-step programmability. These advantages unlock a new framework for matrix-vector multiplications required by neuromorphic silicon photonic circuits, supporting: i) high-speed and high-accuracy neural network (NN) inference, ii) high-speed tiled matrix multiplication, iii) NN training and iv) programmable photonic NNs. This new potential is initially validated through recent experimental results using SiGe EAM technology and static weights and, subsequently, utilized for demonstrating experimentally the first Deep NN (DNN) where optical tiled matrix multiplication up to 50 GHz is realized, allowing optics to execute DNNs with large number of trainable parameters over a limited photonic hardware. Finally, the new performance framework is benchmarked against state-of-the-art NN processors and photonic NN roadmap projections, highlighting its perspectives to turn the energy and area efficiency promise of neuromorphic silicon photonics into a tangible reality.

Coherent Photonic Crossbar Arrays for Large-Scale Matrix-Matrix Multiplication

Abstract: Advances in deep learning research over the past decade have been enabled by an increasingly unsustainable demand for compute power. This trend has dramatically outpaced the slowing growth in the performance and efficiency of electronic computing hardware. Here, we propose a hybrid photonic-electronic computing architecture which leverages a photonic crossbar array and homodyne detection to perform large-scale coherent matrix-matrix multiplication. This approach bypasses the requirements of high-speed electronic readout and frequent reprogramming of photonic weights which significantly reduces energy consumption and latency in the limit of large matrices—two major factors limiting efficiency for many analog computing approaches.

All-Optical Nonlinear Activation Function Based on Germanium Silicon Hybrid Asymmetric Coupler

Abstract: Nonlinear activation functions are crucial for optical neural networks (ONNs) to achieve more various functions. However, the current nonlinear functions suffer from some dilemma, including high power consumption, high loss, and limited bandwidth. Here, we propose and demonstrate an all-optical implementation of a nonlinear activation function based on germanium silicon hybrid integration. The principle lies in the intrinsic absorption and the carrier-induced refractive index change of germanium in C -band. It has a large operating bandwidth and a response frequency of 70 MHz, with a loss of 4.28 dB and a threshold power of 5.1 mW. Adopting it to the MNIST handwriting data set classification, it shows an improvement in accuracy from 91.6% to 96.8%. This proves that our scheme has great potential for advanced ONN applications.

Dynamic Precision Analog Computing for Neural Networks

Abstract: Analog electronic and optical computing exhibit tremendous advantages over digital computing for accelerating deep learning when operations are executed at low precision. Although digital architectures support programmable precision to increase efficiency, analog computing architectures today only support a single, static precision. In this work, we characterize the relationship between the effective number of bits (ENOB) of precision of analog processors, which is limited by noise, and digital bit precision for quantized neural networks. We propose extending analog computing architectures to support dynamic levels of precision by repeating operations and averaging the result, decreasing the impact of noise. To utilize dynamic precision, we propose a method for learning the precision of each layer of a pre-trained model without retraining network weights. We evaluate this method on analog architectures subject to shot noise, thermal noise, and weight noise and find that employing dynamic precision reduces energy consumption by up to 89% for computer vision models such as Resnet50 and by 24% for natural language processing models such as BERT. In one example, we apply dynamic precision to a shot-noise limited homodyne optical neural network and simulate inference at an optical energy consumption of 2.7 aJ/MAC for Resnet50 and 1.6 aJ/MAC for BERT with ${< }2\%$ accuracy degradation, implying that the optical energy consumption is unlikely to be the dominant cost.

Wafer-Scale Hybrid Integration of InP DFB Lasers on Si Photonics by Flip-Chip Bonding With sub-300 nm Alignment Precision

Abstract: InP DFB lasers are flip-chip bonded to 300 mm Si photonic wafers using a pick-and-place tool with an advanced vision system, realizing high-precision and high-throughput passive assembly. By careful co-design of the InP-Si Photonics electrical, optical and mechanical interface, as well as dedicated alignment fiducials, sub-300 nm post-bonding alignment precision is realized in a 25 s cycle time. Optical coupling losses of −1.5±−0.5 dB are achieved at 1550 nm wavelength after epoxy underfill, with up to 40 mW of optical power coupled to the SiN waveguides on the Si photonics wafer. The bonding interface adds less than 10% to the series resistance of the laser diodes and post-bonding thermal resistance is measured to be 76 K/W (or 27 K.mm/W), mostly dominated by heat spreading resistance in the InP lasers as suggested by in-depth thermal modeling. Although the assembled lasers suffer from significant, unintentional optical backreflection from the fiber grating couplers used for optical characterization, laser linewidths well below 1 MHz have been measured under specific drive conditions, as supported by a detailed laser noise analysis. Finally, we demonstrate the ability of bonded laser assemblies to pass early reliability tests.

GHz Rate Neuromorphic Photonic Spiking Neural Network With a Single Vertical-Cavity Surface-Emitting Laser (VCSEL)

Abstract: Vertical-Cavity Surface-Emitting Lasers (VCSELs) are highly promising devices for the construction of neuromorphic photonic information processing systems, due to their numerous desirable properties such as low power consumption, high modulation speed, and compactness. Of particular interest is the ability of VCSELs to exhibit neuron-like spiking responses at ultrafast sub-nanosecond rates; thus offering great prospects for high-speed light-enabled spike-based processors. Recent works have shown spiking VCSELs are capable of tackling pattern recognition and image processing problems, but additionally, VCSELs have been used as nonlinear elements in photonic reservoir computing (RC) implementations, yielding state of the art operation. This work introduces and experimentally demonstrates for the first time a new GHz-rate photonic spiking neural network (SNN) built with a single VCSEL neuron. The reported system effectively implements a photonic VCSEL-based spiking reservoir computer, and demonstrates its successful application to a complex nonlinear classification task. Importantly, the proposed system benefits from a highly hardware-friendly, inexpensive realization (a single VCSEL device and off-the-shelf fibre-optic components), for high-speed (GHz-rate inputs) and low-power (sub-mW optical input power) photonic operation. These results open new pathways towards future neuromorphic photonic spike-based processing systems based upon VCSELs (or other laser types) for novel ultrafast machine learning and AI hardware.

Non-Reciprocal Sub-Micron Waveguide Raman Amplifiers, Towards Loss-Less Silicon Photonics

Abstract: Optical amplification in silicon photonic integrated circuits remains a challenge and stimulated Raman scattering has been proposed as a means of achieving optical gain in silicon. In this work, we experimentally investigate silicon Raman amplifiers optimized for fabrication with open-access foundry sub-micron silicon platform. We discuss and experimentally validate the importance of considering the non-reciprocal Raman-gain by using counter-propagating or co-propagating pumps and probes, different amplifier lengths, input pump powers and nonlinear loss values. We demonstrate a Raman-assisted loss-less optical circuit in a 1.2-cm-length waveguide that reaches zero net-gain with only 60 mW continuous-wave pumping. A 5.5 dB stimulated Raman scattering (SRS) gain yielding 0.5 dB net-gain is also demonstrated with a 115 mW pump in a 3.25-cm waveguide. Furthermore, we examine the nonlinear loss of silicon waveguides to estimate free carrier lifetime (FCL) with different bias voltages applied to the waveguide p-i-n structure. Accounting for non-reciprocal Raman amplification by using the two dataset of co- and counter-propagating SRS gain, we extract the Raman gain coefficient of this 220 nm thick silicon photonic waveguide. We perform the curve fitting over the whole input pump power range using the extracted FCL, and over the low power range where only linear loss is expected. We find a good agreement in the extracted Raman gain coefficients. We use these key parameters as input to a silicon photonic Raman amplifier model to find the optimum performance based on the available footprint and pump power.

High-Power High-Speed MUTC Waveguide Photodiodes Integrated on Si3N4/Si Platform Using Micro-Transfer Printing

Abstract: We demonstrate modified uni-traveling carrier (MUTC) waveguide photodiodes on a silicon nitride/silicon (Si3N4/Si) photonic platform using micro-transfer printing. The photodiodes exhibit high bandwidth of 54 GHz and low dark current of 30 nA at −3 V. The MUTC photodiode (PD) with short InGaAsP waveguide is directly (butt-) coupled to the Si3N4 waveguide and was designed to facilitate efficient light coupling while preserving the favorable high-power capability of InGaAsP/InP evanescently-coupled waveguide PDs. A responsivity of 0.42 A/W at 1310 nm wavelength and an output power as high as +7 dBm at 50 GHz and 22 mA photocurrent were measured.

Micro-Transfer Printing for Heterogeneous Si Photonic Integrated Circuits

Abstract: Silicon photonics (SiPh) is a disruptive technology in the field of integrated photonics and has experienced rapid development over the past two decades. Various high-performance Si and Ge/Si-based components have been developed on this platform that allow for complex photonic integrated circuits (PICs) with small footprint. These PICs have found use in a wide range of applications. Nevertheless, some non-native functions are still desired, despite the versatility of Si, to improve the overall performance of Si PICs and at the same time cut the cost of the eventual Si photonic system-on-chip. Heterogeneous integration is verified as an effective solution to address this issue, e.g. through die-wafer-bonding and flip-chip. In this paper, we discuss another technology, micro-transfer printing, for the integration of non-native material films/opto-electronic components on SiPh-based platforms. This technology allows for efficient use of non-native materials and enables the (co-)integration of a wide range of materials/devices on wafer scale in a massively parallel way. In this paper we review some of the recent developments in the integration of non-native optical functions on Si photonic platforms using micro-transfer printing.

eFIN: Enhanced Fourier Imager Network for Generalizable Autofocusing and Pixel Super-Resolution in Holographic Imaging

Abstract: The application of deep learning techniques has greatly enhanced holographic imaging capabilities, leading to improved phase recovery and image reconstruction. Here, we introduce a deep neural network termed enhanced Fourier Imager Network (eFIN) as a highly generalizable and robust framework for hologram reconstruction with pixel super-resolution and image autofocusing. Through holographic microscopy experiments involving lung, prostate and salivary gland tissue sections and Papanicolau (Pap) smears, we demonstrate that eFIN has a superior image reconstruction quality and exhibits external generalization to new types of samples never seen during the training phase. This network achieves a wide autofocusing axial range of Δz ∼ 350 μm, with the capability to accurately predict the hologram axial distances by physics-informed learning. eFIN enables 3× pixel super-resolution imaging and increases the space-bandwidth product of the reconstructed images by 9-fold with almost no performance loss, which allows for significant time savings in holographic imaging and data processing steps. Our results showcase the advancements of eFIN in pushing the boundaries of holographic imaging for various applications in e.g., quantitative phase imaging and label-free microscopy.

Universal Linear Optics Revisited: New Perspectives for Neuromorphic Computing With Silicon Photonics

Abstract: Reprogrammable optical meshes comprise a subject of heightened interest for the execution of linear transformations, having a significant impact in numerous applications that extend from the implementation of optical switches up to neuromorphic computing. Herein, we review the state-of-the-art approaches for the realization of unitary transformations and universal linear operators in the photonic domain and present our recent work in the field, that allows for fidelity restorable and low-loss optical circuitry with single-step programmability. These advantages unlock a new framework for matrix-vector multiplications required by neuromorphic silicon photonic circuits, supporting: i) high-speed and high-accuracy neural network (NN) inference, ii) high-speed tiled matrix multiplication, iii) NN training and iv) programmable photonic NNs. This new potential is initially validated through recent experimental results using SiGe EAM technology and static weights and, subsequently, utilized for demonstrating experimentally the first Deep NN (DNN) where optical tiled matrix multiplication up to 50 GHz is realized, allowing optics to execute DNNs with large number of trainable parameters over a limited photonic hardware. Finally, the new performance framework is benchmarked against state-of-the-art NN processors and photonic NN roadmap projections, highlighting its perspectives to turn the energy and area efficiency promise of neuromorphic silicon photonics into a tangible reality.

Analysis of Diffractive Neural Networks for Seeing Through Random Diffusers

Abstract: Imaging through diffusive media is a challenging problem, where the existing solutions heavily rely on digital computers to reconstruct distorted images. We provide a detailed analysis of a computer-free, all-optical imaging method for seeing through random, unknown phase diffusers using diffractive neural networks, covering different deep learning-based training strategies. By analyzing various diffractive networks designed to image through random diffusers with different correlation lengths, a trade-off between the image reconstruction fidelity and distortion reduction capability of the diffractive network was observed. During its training, random diffusers with a range of correlation lengths were used to improve the diffractive network's generalization performance. Increasing the number of random diffusers used in each epoch reduced the overfitting of the diffractive network's imaging performance to known diffusers. We also demonstrated that the use of additional diffractive layers improved the generalization capability to see through new, random diffusers. Finally, we introduced deliberate misalignments in training to “vaccinate” the network against random layer-to-layer shifts that might arise due to the imperfect assembly of the diffractive networks. These analyses provide a comprehensive guide in designing diffractive networks to see through random diffusers, which might profoundly impact many fields, such as biomedical imaging, atmospheric physics, and autonomous driving.

Coherent Photonic Crossbar Arrays for Large-Scale Matrix-Matrix Multiplication

Abstract: Advances in deep learning research over the past decade have been enabled by an increasingly unsustainable demand for compute power. This trend has dramatically outpaced the slowing growth in the performance and efficiency of electronic computing hardware. Here, we propose a hybrid photonic-electronic computing architecture which leverages a photonic crossbar array and homodyne detection to perform large-scale coherent matrix-matrix multiplication. This approach bypasses the requirements of high-speed electronic readout and frequent reprogramming of photonic weights which significantly reduces energy consumption and latency in the limit of large matrices—two major factors limiting efficiency for many analog computing approaches.

Ultra-Short Laser Pulses Propagation Through Mouse Head Tissues: Experimental and Computational Study

Abstract: We present a prototype and verification of a multichannel laser system applicable to optogenetic research. In vivo photostimulation of neural cells expressing photoconvertible phytochromes or opsins requires enough light irradiation delivery to the brain that cannot be supported by continues-wave (CW) light sources. The use of ultra-short pulsed (USP) lasers operating in the second near-infrared region (II-NIR) and allowing nonlinear activation and deactivation of the photoactuators is a promising method that allows to increase the penetration depth and provide spatio-temporal localisation of radiation in tissues. This study aimed to investigate the efficiency of USP light propagation in the skin, skull, and brain of the mouse head, as well as to compare it with the corresponding CW radiation propagation in the 750–830 nm and 1086–1183 nm wavelength ranges. The experimental results and computer modelling demonstrate that about 10–12% of the initial laser radiation can reach the brain tissues. These results prove that under certain conditions, the USP laser radiation can reach a penetration depth with required power that will be sufficient for non-linear activation of opsins/phytochromes in the brain of living animals.

All-Optical Nonlinear Activation Function Based on Germanium Silicon Hybrid Asymmetric Coupler

Abstract: Nonlinear activation functions are crucial for optical neural networks (ONNs) to achieve more various functions. However, the current nonlinear functions suffer from some dilemma, including high power consumption, high loss, and limited bandwidth. Here, we propose and demonstrate an all-optical implementation of a nonlinear activation function based on germanium silicon hybrid integration. The principle lies in the intrinsic absorption and the carrier-induced refractive index change of germanium in C -band. It has a large operating bandwidth and a response frequency of 70 MHz, with a loss of 4.28 dB and a threshold power of 5.1 mW. Adopting it to the MNIST handwriting data set classification, it shows an improvement in accuracy from 91.6% to 96.8%. This proves that our scheme has great potential for advanced ONN applications.

Mid-infrared supercontinuum generation in a varying dispersion waveguide for multi-species gas spectroscopy

Abstract: We report the experimental generation of a broadband and flat mid-infrared supercontinuum in a silicon-germanium-on-silicon two-stage waveguide. Our particular design combines a short and narrow waveguide section for efficient supercontinuum generation, and an inverse tapered section that promotes the generation of two spectrally shifted dispersive waves along the propagation direction, leading to an overall broader and flatter supercontinuum. The experimentally generated supercontinuum extended from 2.4 to 5.5 µm, only limited by the long wavelength detection limit of our spectrum analyzer. Numerical simulations predict that the supercontinuum actually extends to 7.8 µm. We exploit the enhanced flatness of our supercontinuum for a proof-of-principle demonstration of free-space multi-species gas spectroscopy of water vapor and carbon dioxide.

Automated Detection of Corneal Edema With Deep Learning-Assisted Second Harmonic Generation Microscopy

Abstract: Second Harmonic Generation Microscopy (SHG) is widely acknowledged as a valuable non-linear optical imaging tool, its contrast mechanism providing the premises to non-invasively identify, characterize, and monitor changes in the collagen architecture of tissues. However, the interpretation of SHG data can pose difficulties even for experts histopathologists, which represents a bottleneck for the translation of SHG-based diagnostic frameworks to clinical settings. The use of artificial intelligence methods for automated SHG analysis is still in an early stage, with only few studies having been reported to date, none addressing ocular tissues yet. In this work we explore the use of three Deep Learning models, the highly popular InceptionV3 and ResNet50, alongside FLIMBA, a custom developed architecture, requiring no pre-training, to automatically detect corneal edema in SHG images of porcine cornea. We observe that Deep Learning models building on different architectures provide complementary results for the classification of cornea SHG images and demonstrate an AU-ROC = 0.98 for their joint use. These results have potential to be extrapolated to other diagnostics scenarios, such as automated extraction of hydration level of cornea, or identification of corneal edema causes, and thus pave the way for novel methods for precision diagnostics of the cornea with Deep-Learning assisted SHG imaging.

Quantifying the Accuracy of Microcomb-Based Photonic RF Transversal Signal Processors

Abstract: Photonic RF transversal signal processors, which are equivalent to reconfigurable electrical digital signal processors but implemented with photonic technologies, have been widely used for modern high-speed information processing. With the capability of generating large numbers of wavelength channels with compact micro-resonators, optical microcombs bring new opportunities for realizing photonic RF transversal signal processors that have greatly reduced size, power consumption, and complexity. Recently, a variety of signal processing functions have been demonstrated using microcomb-based photonic RF transversal signal processors. Here, we provide detailed analysis for quantifying the processing accuracy of microcomb-based photonic RF transversal signal processors. First, we investigate the theoretical limitations of the processing accuracy determined by tap number, signal bandwidth, and pulse waveform. Next, we discuss the practical error sources from different components of the signal processors. Finally, we analyze the contributions of the theoretical limitations and the experimental factors to the overall processing inaccuracy both theoretically and experimentally. These results provide a useful guide for designing microcomb-based photonic RF transversal signal processors to optimize their accuracy.

All-Optical Recurrent Neural Network With Reconfigurable Activation Function

Abstract: Optical Neural Networks (ONNs) can be promising alternatives for conventional electrical neural networks as they offer ultra-fast data processing with low energy consumption. However, lack of suitable nonlinearity is standing in their way of achieving this goal. While this problem can be circumvented in feed-forward neural networks, the performance of the recurrent neural networks (RNNs) depends heavily on their nonlinearity. In this paper, we first propose and numerically demonstrate a novel reconfigurable optical activation function, named ROA, based on adding or subtracting the outputs of two saturable absorbers (SAs). RAO can provide both bounded and unbounded outputs by facilitating an electrically programmable adder/subtractor design. Second, with the help of RAO, which can be altered to resemble Tanh or Sigmoid activation functions, we take a step further and numerically demonstrate OptoRNN, a new design for an all-optical RNN in free-space optics. Moreover, by benefiting from mathematical modeling of the multiplication noise, as well as an altered loss function, we enable vector-matrix multiplication (VMM) six times more parallel than conventional optical VMM method for the linear part of the OptoRNN. Finally, through comprehensive simulation studies, we demonstrate that utilizing the OptoRNN, we can achieve 96.3% train accuracy for sequential-MNIST dataset.

Whole Brain Micro-Vascular Imaging Using Robot Assisted Optical Coherence Tomography Angiography

Abstract: Optical coherence tomography angiography (OCTA) can provide depth resolved image of microvasculature but fail to acquire both high resolution and large field imaging of an irregular sample. The robotic arm can easily scan along the sample surface to ensure that the sample at the same depth is all in the same focal plane, so OCTA can be obtained without the effect of unevenly distributed resolution and fall off as occurred in conventional imaging with linear scanning of irregular sample. So, we integrated a 120 kHz OCT with a robotic arm and proposed a novel calibration method for robotic arm with a repeatability positioning accuracy of ∼60μm. A correlation mapping (CM) masked eigen decomposition OCTA was proposed with CM > 0.6 for binarization to improve the vascular visualization and CM > 0.99 for extraction of big vessels to enhance stitching efficacy. With the aid of robotic arm, micro-vascular OCTA of whole mouse brain was achieved for the first time, to our knowledge, over a volume size of 10 mm × 8.1 mm × 1 mm with a resolution of ∼4μm(axial) x ∼6μm(lateral). We believe robot assisted OCTA system will have great potential in biomedical application like intraoperative navigation and other fields.

Tunable Polarization Encoder Capable of Polarization Conversion and Separation Based on a Layered Photonic Structure

Abstract: In this paper, a polarization encoder with a polarization conversion function is investigated using a layered photonic structure comprising nonlinear materials and indium antimonide. The electric field of the incident linear polarization waves considered are parallel to the x-y plane at an angle of 45° to the x-axis, and can be decomposed into components directed along the x- and y-axes, generating two polarized waves: transverse electric and transverse magnetic waves. Due to the nonlinear effect and the temperature and magnetic flux density tunable properties of indium antimonide, the light intensity, temperature, and magnetic flux density can be used to adjust the reflection phase difference and polarization form. Discussing the influences of these three physical quantities on the phase difference, the incident linear polarization wave and reflected circular polarization wave conversion, the circular polarization conversion of two different rotations and the circular polarization separation of a certain bandwidth can be simultaneously realized with the specific modification of the light intensity or temperature. Moreover, four disparate input logic levels composed of temperature and magnetic flux density achieve specific encoding outputs for all polarization forms. This signifies that polarization selection through control of the logic codes can be easily attained. The consequences show that this form of polarization conversion and encoding is tailored and can be accurately manipulated. Hence, this provides a novel foundation for tunable and diverse polarization splitters and selectors.

All-Optical Recurrent Neural Network With Reconfigurable Activation Function

Abstract: Optical Neural Networks (ONNs) can be promising alternatives for conventional electrical neural networks as they offer ultra-fast data processing with low energy consumption. However, lack of suitable nonlinearity is standing in their way of achieving this goal. While this problem can be circumvented in feed-forward neural networks, the performance of the recurrent neural networks (RNNs) depends heavily on their nonlinearity. In this paper, we first propose and numerically demonstrate a novel reconfigurable optical activation function, named ROA, based on adding or subtracting the outputs of two saturable absorbers (SAs). RAO can provide both bounded and unbounded outputs by facilitating an electrically programmable adder/subtractor design. Second, with the help of RAO, which can be altered to resemble Tanh or Sigmoid activation functions, we take a step further and numerically demonstrate OptoRNN, a new design for an all-optical RNN in free-space optics. Moreover, by benefiting from mathematical modeling of the multiplication noise, as well as an altered loss function, we enable vector-matrix multiplication (VMM) six times more parallel than conventional optical VMM method for the linear part of the OptoRNN. Finally, through comprehensive simulation studies, we demonstrate that utilizing the OptoRNN, we can achieve 96.3% train accuracy for sequential-MNIST dataset.

A Hybrid-Integrated Photonic Spiking Neural Network Framework Based on an MZI Array and VCSELs-SA

Abstract: We propose an inferencing framework of a hybrid-integrated photonic spiking neural network (PSNN) to perform pattern recognition tasks, where the linear computation is realized based on a 4 × 4 silicon photonic Mach-Zehnder interferometer (MZI) array, and the nonlinear computation is performed by an InP-based spiking neuron array based on vertical-cavity surface-emitting lasers with an embedded saturable absorber (VCSELs-SA). With the modified Tempotron-like remote supervised method (ReSuMe) training algorithm, we realize two pattern recognition tasks, the recognition of numbers “0-3” and optical character recognition (OCR). The phase shifts in the MZI array are accurately configured to represent the weight matrix according to the decomposition procedure of a 4 × 4 triangular MZI mesh. Besides, the effects of the phase shift error and quantization precision of phase shifters (PSs) on the recognition performance are analyzed. For the OCR task, the 400 × 10 PSNN is realized by multiplexing the 4 × 4 MZI array based on the matrix blocking and the reconfigurability of the MZI array. This work provides a systematic computational model of the hybrid-integrated PSNN based on the silicon photonics and InP platforms, enabling the co-design and optimization of hardware architecture and algorithm, which contributes one step forward toward the construction of a hybrid-integrated PSNN hardware system.

Raman Spectroscopy and Machine Learning for Microplastics Identification and Classification in Water Environments

Abstract: As emerging pollutants of concern, microplastics (MPs) have been found in different water environments and have an impact on human health through the aquatic food chain. To advance our understanding of the traceability and environmental fate of MPs, reproducible and accurate methods, techniques, and analytical methods are necessary for MP type identification and characterization. In this study, based on Raman spectroscopy technology to extract characteristic peak information of MPs with fingerprint features, coupled to sparse autoencoder (SAE) and softmax classifier framework, the rapid identification and classification of six common MP (PET, PVC, PP, PS, PC, PE) particles in five water (pure water, rain water, lake water, tap water, and sea water) environments was realized. The results show that the average test accuracy of the trained algorithm is as high as 99.1%, which is better than 93.95% and 74.55% of the classical machine learning algorithms support vector machine (SVM) and back propagation (BP) neural network. Success rate indicates that the proposed method can be used to identify the MP samples.

Raman Spectroscopy and Machine Learning for Microplastics Identification and Classification in Water Environments

Abstract: As emerging pollutants of concern, microplastics (MPs) have been found in different water environments and have an impact on human health through the aquatic food chain. To advance our understanding of the traceability and environmental fate of MPs, reproducible and accurate methods, techniques, and analytical methods are necessary for MP type identification and characterization. In this study, based on Raman spectroscopy technology to extract characteristic peak information of MPs with fingerprint features, coupled to sparse autoencoder (SAE) and softmax classifier framework, the rapid identification and classification of six common MP (PET, PVC, PP, PS, PC, PE) particles in five water (pure water, rain water, lake water, tap water, and sea water) environments was realized. The results show that the average test accuracy of the trained algorithm is as high as 99.1%, which is better than 93.95% and 74.55% of the classical machine learning algorithms support vector machine (SVM) and back propagation (BP) neural network. Success rate indicates that the proposed method can be used to identify the MP samples.

The Silicon-Based XOI Wafer: The Most General Electronics-Photonics Platform for Computing, Sensing, and Communications

Abstract: This paper proposes that the 300-mm-diameter silicon wafer coated with a thin insulator layer, which becomes a buried layer, is the most general and most capable platform for high-volume foundry-manufactured, waveguided, photonic integrated circuits (PICs) and for the on-wafer electronics that control and signal-process the photonics. We call this “on insulator” platform an electronic- photonic (or optoelectronic) integrated-circuit wafer. For a few potential applications like “general intelligence” (Shainline et al., 2021), entire wafers would be deployed. However, in almost every case, the wafer will be diced into hundreds of electronic-photonic chips (chips are the real aim of wafer creation). Those chips would be commercial products or custom-made, application-specific PICs. The goal of this paper is to present a detailed vision of the ultimate electronic- photonic wafers that: (1) serve a vast range of applications, (2) operate at any wavelength within the ultraviolet, visible, near-infrared and middle infrared, (3) provide low-loss, well-confined optical waveguiding across the wafer, (4) utilize an optimized or application-specific combination of photonic materials including semiconductors, insulators, ferroelectrics, poled polymers (Xu et al., 2022), phase-change materials (PCMs) (Wuttig et al., 2017), plasmonics (Moor et al., 2021), (Amin et al., 2021), and 2D materials such as graphene (Liu et al., 2020), (5) offer one-or-more practical electro-optical modulation-and-switching mechanisms that are discussed below, (6) offer on-wafer laser diodes, wavelength-multiplexed comb sources, LEDs, optical amplifiers, and photodetectors, (7) provide a full range of CMOS-or-“other” control electronics as well as electronic memories and data converters (analog-to-digital and digital-to-analog), and (8) are manufacturable in volume by proven techniques such as wafer bonding, smart cut, and hetero-epitaxy– or are made by emerging methods. The insulator mentioned above could be silicon dioxide (SiO2) or alumina (Al2O3), or silicon nitride (Si3N4 or SiN). SiO2 is generally preferred, but the Al2O3 and the SiN offer better mid- infrared transparency than the oxide.

Mid-Infrared Supercontinuum Generation in a Varying Dispersion Waveguide for Multi-Species Gas Spectroscopy

Abstract: We report the experimental generation of a broadband and flat mid-infrared supercontinuum in a silicon-germanium-on-silicon two-stage waveguide. Our particular design combines a short and narrow waveguide section for efficient supercontinuum generation, and an inverse tapered section that promotes the generation of two spectrally shifted dispersive waves along the propagation direction, leading to an overall broader and flatter supercontinuum. The experimentally generated supercontinuum extended from 2.4 to 5.5 μm, only limited by the long wavelength detection limit of our spectrum analyzer. Numerical simulations predict that the supercontinuum actually extends to 7.8 μm. We exploit the enhanced flatness of our supercontinuum for a proof-of-principle demonstration of free-space multi-species gas spectroscopy of water vapor and carbon dioxide.