-
Asymmetrical estimator for training grey-box deep photonic neural networks
Authors:
Yizhi Wang,
Minjia Chen,
Chunhui Yao,
Jie Ma,
Ting Yan,
Richard Penty,
Qixiang Cheng
Abstract:
Physical neural networks (PNNs) are emerging paradigms for neural network acceleration due to their high-bandwidth, in-propagation analogue processing. Despite the advantages of PNN for inference, training remains a challenge. The imperfect information of the physical transformation means the failure of conventional gradient-based updates from backpropagation (BP). Here, we present the asymmetrica…
▽ More
Physical neural networks (PNNs) are emerging paradigms for neural network acceleration due to their high-bandwidth, in-propagation analogue processing. Despite the advantages of PNN for inference, training remains a challenge. The imperfect information of the physical transformation means the failure of conventional gradient-based updates from backpropagation (BP). Here, we present the asymmetrical training (AT) method, which treats the PNN structure as a grey box. AT performs training while only knowing the last layer output and neuron topological connectivity of a deep neural network structure, not requiring information about the physical control-transformation map**. We experimentally demonstrated the AT method on deep grey-box PNNs implemented by uncalibrated photonic integrated circuits (PICs), improving the classification accuracy of Iris flower and modified MNIST hand-written digits from random guessing to near theoretical maximum. We also showcased the consistently enhanced performance of AT over BP for different datasets, including MNIST, fashion-MNIST, and Kuzushiji-MNIST. The AT method demonstrated successful training with minimal hardware overhead and reduced computational overhead, serving as a robust light-weight training alternative to fully explore the advantages of physical computation.
△ Less
Submitted 28 May, 2024;
originally announced May 2024.
-
Unlocking Electro-optic Resonant Phase Shifting for Multi-dimensional, Ultra-dynamic Photonic Switches
Authors:
Lingzhi Luo,
Rui Ma,
Richard V. Penty,
Qixiang Cheng
Abstract:
Optical circuit switching is connection-oriented, being deterministic through the reservation of a complete wavelength channel or spatial path for a certain period. However, this comes at a trade-off against link dynamics, and overall capacity can thus be constrained by the time slot reservations, especially for switches with microsecond- to millisecond-scale reconfiguration times. For data-intens…
▽ More
Optical circuit switching is connection-oriented, being deterministic through the reservation of a complete wavelength channel or spatial path for a certain period. However, this comes at a trade-off against link dynamics, and overall capacity can thus be constrained by the time slot reservations, especially for switches with microsecond- to millisecond-scale reconfiguration times. For data-intensive applications, the communication patterns associated with random data sets typically yield short-lived flows. This situation calls for a new multi-dimensional switching paradigm that fully exploits not only the space and wavelength domains but also with nanosecond-scale reconfigurable capability in the time domain to enable ultra-dynamic links. In this work, we focus on the exploitation of micro-ring resonant phase shifters (RPSs) that are wavelength selective for optical switching in a single plane. By proposing an innovative analytical method with transmission circle chart, we fully unlock the power of RPS with nanosecond-scale reconfigurability and the capability to arbitrarily manipulate its phase and amplitude. Such a compact model offers fresh insights into designs with under and critically coupled RPSs beyond the commonly explored over-coupling condition. This creates not only versatile switch elements but also perfect absorbers for robust multi-wavelength operations. The proposed device can bring about a breakthrough in the optical switching capacity that potentially addresses the challenges faced by modern data center networks, as well as other photonic circuits for high-throughput signal processing.
△ Less
Submitted 5 March, 2024;
originally announced March 2024.
-
Benchmarking reconstructive spectrometer with multi-resonant cavities
Authors:
Chunhui Yao,
Kangning Xu,
Tianhua Lin,
Jie Ma,
Chumeng Yao,
Peng Bao,
Zhitian Shi,
Richard Penty,
Qixiang Cheng
Abstract:
Recent years have seen the rapid development of miniaturized reconstructive spectrometers (RSs), yet they still confront a range of technical challenges, such as bandwidth/resolution ratio, sensing speed, and/or power efficiency. Reported RS designs often suffer from insufficient decorrelation between sampling channels, which results in limited compressive sampling efficiency, in essence, due to i…
▽ More
Recent years have seen the rapid development of miniaturized reconstructive spectrometers (RSs), yet they still confront a range of technical challenges, such as bandwidth/resolution ratio, sensing speed, and/or power efficiency. Reported RS designs often suffer from insufficient decorrelation between sampling channels, which results in limited compressive sampling efficiency, in essence, due to inadequate engineering of sampling responses. This in turn leads to poor spectral-pixel-to-channel ratios (SPCRs), typically restricted at single digits. So far, there lacks a general guideline for manipulating RS sampling responses for the effectiveness of spectral information acquisition. In this study, we shed light on a fundamental parameter from the compressive sensing theory - the average mutual correlation coefficient v - and provide insight into how it serves as a critical benchmark in RS design with regards to the SPCR and reconstruction accuracy. To this end, we propose a novel RS design with multi-resonant cavities, consisting of a series of partial reflective interfaces. Such multi-cavity configuration offers an expansive parameter space, facilitating the superlative optimization of sampling matrices with minimized v. As a proof-of-concept demonstration, a single-shot, dual-band RS is implemented on a SiN platform, tailored for capturing signature spectral shapes across different wavelength regions, with customized photonic crystal nanobeam mirrors. Experimentally, the device demonstrates an overall operation bandwidth of 270 nm and a <0.5 nm resolution with only 15 sampling channels per band, leading to a record high SPCR of 18.0. Moreover, the proposed multi-cavity design can be readily adapted to various photonic platforms. For instance, we showcase that by employing multi-layer coatings, an ultra-broadband RS can be optimized to exhibit a 700 nm bandwidth with an SPCR of over 100.
△ Less
Submitted 1 March, 2024;
originally announced March 2024.
-
Photonic Chiplet Interconnection via 3D-Nanoprinted Interposer
Authors:
Huiyu Huang,
Zhitian Shi,
Giuseppe Talli,
Maxim Kuschnerov,
Richard Penty,
Qixiang Cheng
Abstract:
Photonic integrated circuits utilize various waveguide materials, each excelling in specific metrics like efficient light emission, low propagation loss, high electro-optic efficiency, and potential for mass production. Inherent shortcomings in each platform push exploration of hybrid and heterogeneous integration, which demands specialized designs and extra fabrication processes for each material…
▽ More
Photonic integrated circuits utilize various waveguide materials, each excelling in specific metrics like efficient light emission, low propagation loss, high electro-optic efficiency, and potential for mass production. Inherent shortcomings in each platform push exploration of hybrid and heterogeneous integration, which demands specialized designs and extra fabrication processes for each material combination. Our work introduces a novel hybrid integration scheme employing a 3D-nanoprinted interposer for a photonic chiplet interconnection system. This method represents a generic solution that can readily couple between chips of any material system, with each fabricated on its own technology platform with no change in the established process flow for the individual chips. Mode-size engineering is enhanced by the off-chip parabolic micro-reflectors. The 3D-nanoprinted chip-coupling frame and fiber-guiding funnel enable low-loss, fully passive assembly with a fast-printing process achieving sub-micron accuracy. Mode-field-dimension conversion ratio of 5:2 from fiber to chip is demonstrated with <0.5dB excess loss on top of the 1.7dB inherent coupling loss, marking the largest mode size conversion using non-waveguided components. Additionally, our system demonstrates a 2.5dB die-to-die coupling loss between silicon and InP chips over a 140nm wavelength range (1480nm to 1620nm), showcasing the potential for extensive cross-platform integration by bridging different waveguide materials.
△ Less
Submitted 19 February, 2024;
originally announced February 2024.
-
Redesigning spectroscopic sensors with programmable photonic circuits
Authors:
Chunhui Yao,
Kangning Xu,
Wanlu Zhang,
Minjia Chen,
Qixiang Cheng,
Richard Penty
Abstract:
Optical spectroscopic sensors are a powerful tool to reveal light-matter interactions in many fields, such as physics, biology, chemistry, and astronomy. Miniaturizing the currently bulky spectrometers has become imperative for the wide range of applications that demand in situ or even in vitro characterization systems, a field that is growing rapidly. Benchtop spectrometers are capable of offerin…
▽ More
Optical spectroscopic sensors are a powerful tool to reveal light-matter interactions in many fields, such as physics, biology, chemistry, and astronomy. Miniaturizing the currently bulky spectrometers has become imperative for the wide range of applications that demand in situ or even in vitro characterization systems, a field that is growing rapidly. Benchtop spectrometers are capable of offering superior resolution and spectral range, but at the expense of requiring a large size. In this paper, we propose a novel method that redesigns spectroscopic sensors via the use of programmable photonic circuits. Drawing from compressive sensing theory, we start by investigating the most ideal sampling matrix for a reconstructive spectrometer and reveal that a sufficiently large number of sampling channels is a prerequisite for both fine resolution and low reconstruction error. This number is, however, still considerably smaller than that of the reconstructed spectral pixels, benefitting from the nature of reconstruction algorithms. We then show that the cascading of a few engineered MZI elements can be readily programmed to create an exponentially scalable number of such sampling spectral responses over an ultra-broad bandwidth, allowing for ultra-high resolution down to single-digit picometers without incurring additional hardware costs. Experimentally, we implement an on-chip spectrometer with a fully-programmable 6-stage cascaded MZI structure and demonstrate a < 10 pm resolution with a > 200 nm bandwidth using only 729 sampling channels. This achieves a bandwidth-to-resolution ratio of over 20,000, which is, to our best knowledge, about one order of magnitude greater than any reported miniaturized spectrometers to date. We further illustrate that by employing dispersion-engineered waveguide components, the device bandwidth can be extended to over 400 nm.
△ Less
Submitted 9 June, 2023;
originally announced June 2023.
-
I/O-efficient iterative matrix inversion with photonic integrated circuits
Authors:
Minjia Chen,
Yizhi Wang,
Chunhui Yao,
Adrian Wonfor,
Shuai Yang,
Richard Penty,
Qixiang Cheng
Abstract:
Photonic integrated circuits have been extensively explored for optical processing with the aim of breaking the speed bottleneck of digital electronics. However, the input/output (IO) bottleneck remains one of the key barriers. Here we report a novel photonic iterative processor (PIP) for matrix-inversion-intensive applications. The direct reuse of inputted data in the optical domain unlocks the p…
▽ More
Photonic integrated circuits have been extensively explored for optical processing with the aim of breaking the speed bottleneck of digital electronics. However, the input/output (IO) bottleneck remains one of the key barriers. Here we report a novel photonic iterative processor (PIP) for matrix-inversion-intensive applications. The direct reuse of inputted data in the optical domain unlocks the potential to break the IO bottleneck. We demonstrate notable IO advantages with a lossless PIP for real-valued matrix inversion and integral-differential equation solving, as well as a coherent PIP with optical loops integrated on-chip, enabling complex-valued computation and a net inversion time of 1.2 ns. Furthermore, we estimate at least an order of magnitude enhancement in IO efficiency of a PIP over photonic single-pass processors and the state-of-the-art electronic processors for reservoir training tasks and MIMO precoding tasks, indicating the huge potential of PIP technology in practical applications.
△ Less
Submitted 22 May, 2024; v1 submitted 26 May, 2023;
originally announced May 2023.
-
Dispersion Studies on Multimode Polymer Spiral Waveguides for Board-Level Optical Interconnects
Authors:
Jian Chen,
Nikos Bamiedakis,
Tom J. Edwards,
Christian T. A. Brown,
Richard V. Penty,
Ian H. White
Abstract:
Dispersion studies are conducted on 1m long multimode polymer spiral waveguides with different refractive index profiles. Bandwidth-length products >40GHzxm are obtained from such waveguides under a 50/125 um MMF, indicating the potential of this technology.
Dispersion studies are conducted on 1m long multimode polymer spiral waveguides with different refractive index profiles. Bandwidth-length products >40GHzxm are obtained from such waveguides under a 50/125 um MMF, indicating the potential of this technology.
△ Less
Submitted 1 February, 2017; v1 submitted 6 November, 2016;
originally announced November 2016.
-
Bandwidth and Offset Launch Investigations on a 1.4 m Multimode Polymer Spiral Waveguide
Authors:
Jian Chen,
Nikos Bamiedakis,
Richard V. Penty,
Ian H. White,
Petter Westbergh,
Anders Larsson
Abstract:
Bandwidth measurements are conducted on a 1.4 m long spiral polymer multimode waveguide for a SMF and 50/125 um MMF launch and for different input offsets. The waveguide exhibits a bandwidth of at least 30 GHz for all input types, yielding a bandwidth-length product of at least 42 GHzxm, while no impact is observed on the waveguide performance due to the different spatial input offsets. The result…
▽ More
Bandwidth measurements are conducted on a 1.4 m long spiral polymer multimode waveguide for a SMF and 50/125 um MMF launch and for different input offsets. The waveguide exhibits a bandwidth of at least 30 GHz for all input types, yielding a bandwidth-length product of at least 42 GHzxm, while no impact is observed on the waveguide performance due to the different spatial input offsets. The results indicate that data transmission at data rates even higher than 25 Gb/s can be achieved over such structures, thereby demonstrating the potential of multimode polymer waveguide technologies in short-reach board-level datacommunication links.
△ Less
Submitted 1 February, 2017; v1 submitted 6 November, 2016;
originally announced November 2016.
-
Pulse generation with ultra-superluminal pulse propagation in semiconductor heterostructures by superradiant phase transition enhanced by transient coherent population gratings
Authors:
Peter P. Vasil'ev,
Richard V. Penty,
Ian H. White
Abstract:
This paper reports the observation of ultra-superluminal pulse propagation in GaAs/AlGaAs multiple contact heterostuctures in a superradiant emission regime, and shows definitively that it is a different class of emission from conventional spontaneous or stimulated emission. It is shown that coherent population gratings induced in the semiconductor medium under strong electrical pum** have great…
▽ More
This paper reports the observation of ultra-superluminal pulse propagation in GaAs/AlGaAs multiple contact heterostuctures in a superradiant emission regime, and shows definitively that it is a different class of emission from conventional spontaneous or stimulated emission. It is shown that coherent population gratings induced in the semiconductor medium under strong electrical pum** have great impact in causing a decrease of the group refractive index in the range of 5-40%. This decrease is much greater than that which would be observed due to conventional carrier depletion or chirp mechanisms. The decrease in refractive index in turn causes faster-than-c propagation of femtosecond pulses. The measurement also shows unequivocally the exist of coherent amplification of electromagnetic pulses in semiconductors at room temperature, the coherence being strongly enhanced by interactions of the light with coherent transient gratings locked to carrier gratings. This pulse generation technique can be anticipated to have great potential in applications where highly coherent femtosecond optical pulses must be generated on demand
△ Less
Submitted 24 August, 2015;
originally announced August 2015.
-
Gigacount/second photon detection with InGaAs avalanche photodiodes
Authors:
K. A. Patel,
J. F. Dynes,
A. W. Sharpe,
Z. L. Yuan,
R. V. Penty,
A. J. Shields
Abstract:
We demonstrate high count rate single photon detection at telecom wavelengths using a thermoelectrically-cooled semiconductor diode. Our device consists of a single InGaAs avalanche photodiode driven by a 2 GHz gating frequency signal and coupled to a tuneable self-differencing circuit for enhanced detection sensitivity. We find the count rate is linear with the photon flux in the single photon de…
▽ More
We demonstrate high count rate single photon detection at telecom wavelengths using a thermoelectrically-cooled semiconductor diode. Our device consists of a single InGaAs avalanche photodiode driven by a 2 GHz gating frequency signal and coupled to a tuneable self-differencing circuit for enhanced detection sensitivity. We find the count rate is linear with the photon flux in the single photon detection regime over approximately four orders of magnitude, and saturates at 1 gigacount/s at high photon fluxes. This result highlights promising potential for APDs in high bit rate quantum information applications.
△ Less
Submitted 16 January, 2012;
originally announced January 2012.
