A5020714759- Photonic and Optical Devices
- Neural Networks and Reservoir Computing
- Optical Network Technologies
- Photonic Crystals and Applications
- Advanced Fiber Laser Technologies
- Optical Coatings and Gratings
- Laser-Plasma Interactions and Diagnostics
- Particle Accelerators and Free-Electron Lasers
- Pulsed Power Technology Applications
- Gyrotron and Vacuum Electronics Research
- Random lasers and scattering media
- Terahertz technology and applications
- Laser Design and Applications
- Magneto-Optical Properties and Applications
- Bacterial Identification and Susceptibility Testing
- Thin-Film Transistor Technologies
- Electromagnetic Simulation and Numerical Methods
- Quantum optics and atomic interactions
- Photorefractive and Nonlinear Optics
- Advanced Surface Polishing Techniques
- Optical Polarization and Ellipsometry
- Laser Material Processing Techniques
- Particle Detector Development and Performance
- Plasmonic and Surface Plasmon Research
- Micro and Nano Robotics
Flex (United States)
2021-2024
Stanford University
1993-2023
University College London
2023
Faculty (United Kingdom)
2023
University of Iowa
2022
California State University, Bakersfield
2019
California State University System
2018
Behavioral Tech Research, Inc.
2018
University of Michigan–Ann Arbor
2014
Lawrence Livermore National Laboratory
1999
Recently, integrated optics has gained interest as a hardware platform for implementing machine learning algorithms. Of particular are artificial neural networks, since matrix-vector multi- plications, which used heavily in can be done efficiently photonic circuits. The training of an network is crucial step its application. However, currently on the photonics there no efficient protocol these networks. In this work, we introduce method that enables highly efficient, situ network. We use...
Analog machine learning hardware platforms promise to be faster and more energy-efficient than their digital counterparts. Wave physics, as found in acoustics optics, is a natural candidate for building analog processors time-varying signals. Here we identify mapping between the dynamics of wave computation recurrent neural networks. This indicates that physical systems can trained learn complex features temporal data, using standard training techniques As demonstration, show an...
We introduce an electro-optic hardware platform for nonlinear activation functions in optical neural networks. The optical-to-optical nonlinearity operates by converting a small portion of the input signal into analog electric signal, which is used to intensity-modulate original with no reduction processing speed. Our scheme allows complete on-off contrast transmission at relatively low power thresholds and eliminates requirement having additional sources between each layer network....
The development of inverse design, where computational optimization techniques are used to design devices based on certain specifications, has led the discovery many compact, non-intuitive structures with superior performance. Among various methods, large-scale, gradient-based have been one most important ways a structure containing vast number degrees freedom. These made possible by adjoint method, in which gradient an objective function respect all freedom can be computed using only two...
Gradient-based inverse design in photonics has already achieved remarkable results designing small-footprint, high-performance optical devices. The adjoint variable method, which allows for the efficient computation of gradients, played a major role this success. However, gradient-based optimization not yet been applied to mode-expansion methods that are most common approaches studying periodic structures such as photonic crystals. This is because, simulations, method cannot be defined...
Neural networks are widely deployed models across many scientific disciplines and commercial endeavors ranging from edge computing sensing to large-scale signal processing in data centers. The most efficient well-entrenched method train such is backpropagation, or reverse-mode automatic differentiation. To counter an exponentially increasing energy budget the artificial intelligence sector, there has been recent interest analog implementations of neural networks, specifically nanophotonic...
We experimentally demonstrate an on-chip electro-optic circuit for realizing arbitrary nonlinear activation functions optical neural networks (ONNs). The operates by converting a small portion of the input signal into electrical and modulating intensity remaining signal. Electrical processing allows function to realize any optical-to-optical nonlinearity that does not require amplification. Such line shapes are constrained those conventional nonlinearities. Through numerical simulations, we...
Reconfigurable photonic mesh networks of tunable beamsplitter nodes can linearly transform $N$-dimensional vectors representing input modal amplitudes light for applications such as energy-efficient machine learning hardware, quantum information processing, and mode demultiplexing. Such meshes are typically programmed and/or calibrated by tuning or characterizing each beam splitter one-by-one, which be time-consuming limit scaling to larger meshes. Here we introduce a graph-topological...
Compound semiconductors are the basis for many of highest performance optical and electronic devices in use today. Their widespread commercial application has, however, been limited due to high cost substrates. Device costs can be significantly reduced if substrate is reused a simple, totally non‐destructive rapid process. Here, method that allows indefinite reuse recycling wafers demonstrated, employing combination epitaxial “protection layers”, plasma cleaning techniques return their...
We present a previously unexplored forward-mode differentiation method for Maxwell's equations, with applications in the field of sensitivity analysis. This approach yields exact gradients and is similar to popular adjoint variable method, but provides significant improvement both memory speed scaling problems involving several output parameters, as we analyze context finite-difference time-domain (FDTD) simulations. Furthermore, it an alternative numerical derivative methods, based on...
Metalenses for optical beam manipulation have a significant impact in many exciting applications due their compact, planar geometry and ease of fabrication. However, the enormous physical size metalenses relative to wavelength provides barrier performing accurate simulations reasonable time frame. In principle, full-wave simulation techniques, such as finite-difference time-domain (FDTD) method, are ideal metalens modeling they give an picture device performance. when applied using...
We show that the adjoint variable method can be combined with multi-frequency finite-difference frequency-domain for efficient sensitivity calculations, enabling systematic optimization of active nanophotonic devices. As a proof principle demonstration, we have optimized dynamic isolator structure in two-dimensions, resulting reduction length modulated regions by factor two, while retaining good performance isolation ratio and insertion loss.
We propose an on-chip optical power delivery system for dielectric laser accelerators based on a fractal 'tree-branch' waveguide network. This replaces experimentally demanding free-space manipulations of the driving beam with chip-integrated techniques precise nano-fabrication, enabling access to orders magnitude increases in interaction length and total energy gain these miniature accelerators. Based computational modeling, relativistic regime, our is estimated provide 21 keV over...
Dielectric laser accelerators (DLAs) are fundamentally based on the interaction of photons with free electrons, where energy and momentum conservation satisfied by mediation a nanostructure. In this scheme, photonic nanostructure induces near-fields which transfer from photon to electron, similar inverse-Smith–Purcell effect described in metallic gratings. This, turn, may provide ground-breaking applications, as it is technology promising miniaturize particle down chip scale. This...
Dielectric microstructures have generated much interest in recent years as a means of accelerating charged particles when powered by solid state lasers. The acceleration gradient (or particle energy gain per unit length) is an important figure merit. To design structures with high gradients, we explore the adjoint variable method, highly efficient technique used to compute sensitivity objective respect large number parameters. With this formalism, dielectric structure its entire spatial...
Machine Learning (ML) techniques now impact a wide variety of domains. Highly regulated industries such as healthcare and finance have stringent compliance data governance policies around sharing. Advances in secure multiparty computation (SMC) for privacy-preserving machine learning (PPML) can help transform these by allowing ML computations over encrypted with personally identifiable information (PII). Yet very little SMC-based PPML has been put into practice so far. In this paper we...
This document provides detailed information on the status of Advanced and Novel Accelerators techniques describes steps that need to be envisaged for their implementation in future accelerators, particular high energy physics applications. It complements overview prepared update European Strategy particle physics, a description field. The scientific priorities community are described each technique acceleration able achieve accelerating gradient GeV~range or above. ALEGRO working group...
We created a glucose oxidase (GOx) working electrode on silicon-on-insulator (SOI) wafer for sensing. The SOI was electrically connected to copper wire, and the GOx immobilized onto hydrophilized surface via silanization with aminopropyltriethoxysilane glutaraldehyde. Electrochemical analysis (i.e., cyclic voltammetry) employed identify sensing mechanism evaluate performance of these SOI-GOx sensors. response significantly higher in presence oxygen than that without oxygen, indicating...
To be useful for most scientific and medical applications, compact particle accelerators will require much higher average current than enabled by architectures. For this purpose, we propose a photonic crystal architecture dielectric laser accelerator, referred to as multi-input multi-output silicon accelerator (MIMOSA), that enables simultaneous acceleration of multiple electron beams, increasing the total throughput at least 1 order magnitude. achieve this, show must support mode <mml:math...
We propose a dielectric laser accelerator design based on tapered slot waveguide structure for sub-relativistic electron acceleration. This tapering scheme allows straightforward tuning of the phase velocity accelerating field along propagation direction, which is necessary maintaining synchronization with electrons as their velocities increase. Furthermore, non-resonant nature this better tolerance to experimental errors. also introduce method continuously eikonal approximation, and give...
We study the weakly guided silicon nitride waveguide as an on-chip power delivery solution for dielectric laser accelerators (DLAs). focus on two main limiting factors network DLAs: optical damage and nonlinear characteristics. The typical delivered fluence at onset of is measured to be ∼0.19 J/cm2 a 2 μm central wavelength 250 fs pulse width. This lower than that bulk Si3N4 (∼0.65 J/cm2), but higher (∼0.17 J/cm2). also report nonlinearity-induced spectrum phase variance output this...
Abstract Particle acceleration in dielectric microstructures powered by infrared lasers, or “dielectric laser acceleration” (DLA), is a promising area of advanced accelerator research with the potential to enable more affordable and higher-gradient accelerators for energy frontier science variety other applications. DLA leverages well-established industrial fabrication capabilities commercial availability tabletop lasers reduce cost, axial accelerating fields GV/m range. Desirable...
We describe the integration of a flexible UHF antenna with an epitaxial lift-off thin-film III-V solar cell array used for power generation and wireless communication in flapping-wing robotic platform. The is configured to utilize cells their interconnections as part radiating element, leading compact multifunctional antenna/solar surface. high junction capacitance individual allows RF current conduction through without associated penalty. Metallic interconnects carry both DC well forming...
We present a theoretical framework, based on plasmonic circuit models, for generating multiresonant field intensity enhancement spectrum at single "hot spot" in device. introduce model, consisting of an array coupled LC resonators, that directs current asymmetrically the array, and we show this can funnel energy efficiently from each resonance to element. implement model nanostructure series metal bars differing length, with nearest neighbor strongly electromagnetically through air gaps. The...