Electronic systems are dominant in information processing, display, and storage. However, optical systems are needed for broad-band long distance communication, antenna remoting, filtering, sensing, and imaging. Conventional high-performance electro-optic modulators, which place electronic information on optical waves, are the bottleneck of optical communication links and photonic time-stretch analog to digital converters, suffering a trade-off between low-voltage operation and large microwave bandwidths.

Unfortunately, the development of EO modulators has not kept up with the tremendous pace of electronics over the past decades. To be sure, access to broadband optical channels is curtailed by the limited bandwidth of EO modulators which are currently available in the 35 GHz range. Furthermore, the required drive voltage is currently in the 2-5 V range, a value that increases with speed. While EO modulators with 100Gbps capacity (~70GHz analog bandwidth) are being developed, their bandwidth is approaching the fundamental limit imposed by velocity mismatch between electrical and optical waves in the traveling wave electrode. This is in direct conflict with the trend in electronics in which the voltage swing continues to decrease with increasing transistor speed. It is then clear that the current EO modulator technology is unable to meet the requirements of future systems.

This dissertation focuses on an exciting new concept that exploits modulation instability, a nonlinear optical process, to compensate for the intrinsically weak chi^(2) electro-optic effect. The technique, known as Modulation Index Boosting or MiBo, can be exploited to boost the weak electrooptic effect, one of the most fundamental and pressing predicaments in optical communication. In the approach, modulation sidebands stimulate MI in a third order chi^(3) nonlinear optical material placed after the electrooptic device. This effect enhances the of the electrooptic modulation material and also compensates for the device related high frequency roll-off. Boosting the chi^(2) of one material with the chi^(3)-induced MI of another material is an intriguing concept that enables low voltage electrooptic modulation at ultrahigh frequencies. Such a technology could provide a tremendous capability enhancement to high-speed digital, analog, SIGINT, and radar systems.

Traditional algorithms prevalent in computational imaging and signal processingare hand-crafted empirical rules synthesized to achieve a desired goal. In contrast, our approach is to craft qualitatively new algorithms by emulating laws of physics. Here, we show that Non-Linear Schrodinger Equation (NLSE), the master equation in optical physics can be exploited to invent a new class of computational imaging algorithms with best-in-class performance. We demonstrate a new contrast enhancement algorithm that is computationally efficient, achieves superior color gamut performance, and is able to support real-time video enhancement at 4K and 8K resolutions. We also show how the NLSE operator becomes an edge detection algorithm with exceptional performance in low light levels. In certain cases, these algorithms have the potential to be implemented in physical optics. We demonstrate efficacy of these algorithms in solving a variety of problems for different real-world applications. Specially, we have developed CytoLive, an award-winning real-time live cell tracking tool utilizing our NLSE-guided algorithms to analyze time-lapse microscopy videos acquired under low light conditions. This tool preserves inherent cell behavior by overcoming phototoxicity and photobleaching and has the potential for accelerating research in drug discovery. Next, we discuss CytoEye, a cancer immunotherapy toolbox that mitigates the computational overload of analyzing giga-pixel sized pathology images of tumor microenvironment. Quantitative features extracted by this tool have the capability to predict whether or not patients respond to therapy { an important step toward personalized cancer immunotherapy.

Understanding electrical activity of neural network is of great interest. Although electrodes and optical voltage sensors have been evolved over the past few decades, there is no such probe satisfying not only large enough sensitivity but also fast enough time response in large scale yet. Therefore, the reverse engineering on brain circuitry such that `every action potential in every neuron in large field of view' is still far-off.

In this thesis, I introduce the inorganic voltage sensor for neural electricity imaging, based on semiconductor nanoparticles (NPs) via quantum-confined Stark effect (QCSE). Firstly, I validate NPs' QCSE at room temperature (RT) at single molecule level. Based on the experimental results and theoretical investigation, I optimize the NP structure displaying the largest QCSE. Besides, undiscovered physical phenomena including wavelength blue-shift, linear energy-field dependency, and field dependent Auger recombination were revealed from the QCSE experiment for the first time. Secondly, I predict the performance of membrane

inserted NPs by self-consistent Schr�dinger-Poisson calculation, and propose voltage imaging strategies.

As a delivery method, peptide-based surface coating is developed which successfully guides NPs to the membrane. Lastly, delivered NPs are tested under voltage oscillating HEK

cells and they generate voltage dependent emission fluorescence. This voltage information is finally captured and imaged by charge coupled device and analyzed. This result demonstrates the inorganic voltage sensor's high throughput simultaneous multisite voltage imaging.

The remarkable success of physics in explaining nature and engineering machines is rooted in low-dimensional deterministic models that accurately describe a wide range of natural phenomena. In contrast, Artificial Intelligence (AI) led by neural networks has introduced an alternate data-driven computational framework, achieving astonishing performance in various fields. However, this impressive feat comes at the cost of challenges, including the need for massive datasets, high computational cost, and limited interpretability. To address these issues, there is a growing interest in hybrid approaches that merge the laws of physics and physical systems with AI. This dissertation focuses on developing efficient and interpretable algorithms for vision enhancement and AI applications by drawing inspiration from the principles of optical physics. Unlike traditional algorithms that rely on hand-crafted empirical rules or neural networks that are data-driven and computationally heavy, physics-inspired algorithms leverage laws of nature as blueprints, resulting in low dimensionality, high efficiency, and full interpretability.

The dissertation begins with the principles of Photonic Time Stretch, a hardware technique grounded in optical physics for ultrafast and single-shot data acquisition, which seeded the ideas herein. A unified physical and mathematical framework is derived based on the governing Nonlinear Schrödinger Equation (NLSE). Inspired by this physical principle, a novel class of algorithms called PhyCV (Physics-inspired Computer Vision) is developed for image and video processing tasks, including edge detection, low-light enhancement, and motion detection. These algorithms demonstrate superior performance in vision enhancement and are well-suited for applications on mobile and edge devices with constrained resources. Furthermore, a novel physics-based AI model called the "Nonlinear Schrödinger Network" is proposed, which treats the NLSE as a trainable model in the numerical domain to learn complex patterns and nonlinear mappings from data. This physics-inspired approach provides a more interpretable and parameter-efficient alternative compared to black-box neural networks, achieving comparable or better accuracy on nonlinear classification tasks while significantly reducing the required number of parameters.

Photonic time-stretch is a novel technology that enables capturing of fast, rare and non-repetitive events. Therefore, it operates in real-time with ability to record over long period of time while having fine temporal resolution. The powerful property of photonic time-stretch has already been employed in various fields of application such as analog-to-digital conversion, spectroscopy, laser scanner and microscopy. Further expanding the scope, we fully exploit the time-stretch technology to demonstrate a high throughput web inspection system. Web inspection, namely surface inspection is a nondestructive evaluation method which is crucial for semiconductor wafer and thin film production. We successfully report a dark-field web inspection system with line scan speed of 90.9 MHz which is up to 1000 times faster than conventional inspection instruments. The manufacturing of high quality semiconductor wafer and thin film may directly benefit from this technology as it can easily locate defects with area of less than 10 µm × 10 µm where it allows maximum web flow speed of 1.8 km/s.

The thesis provides an overview of our web inspection technique, followed by description of the photonic time-stretch technique which is the keystone in our system. A detailed explanation of each component is covered to provide quantitative understanding of the system. Finally, imaging results from a hard-disk sample and flexible films are presented along with performance analysis of the system. This project was the first application of time-stretch to industrial inspection, and was conducted under financial support and with close involvement by Hitachi, Ltd.

Label-free classification of microstructures is a valuable approach in a variety of fields including cytometry and atmospheric science. The sensitive classification of microscopic cells and organisms is especially an important outstanding problem in biology. Flow cytometry is a routine method for cell classification. Current flow cytometers use light scattering at two fixed angles to infer information about size and internal complexity of cells at rates of more than 10,000 cells per second. However, this approach limits the precision and information that can be deduced by the cell population from the light scattering patterns. Capturing the full angular scattering spectrum of cells and particles would enable classification of cells with higher resolution and specificity. By capturing the angular dependence of scattering intensity we will be able to extract information about the scattering particle, hence providing a label-free method for particle classification. Current systems that provide angular scattering patterns do not have the throughput required to be implemented in state-of-the-art flow cytometers.

Inverse scattering has been one of the more difficult problems to solve in electromagnetic wave interaction problems. Yet there have been many solutions obtained by analytical and computational modeling for various cases. The angular light scattering profile of particles is dependent on their morphological parameters, such as size, shape, and their internal structure. One of the most interesting applications of modeling these scattering profiles is in characterizing cells to identify abnormalities. Methods to take advantage of this angular dependent information have been demonstrated, however, these methods have various limitations such as low speed and precision. Here I present a new high throughput multi-angle resolved light scattering measurement technique that is able to capture the full angular scattering profile of particles in a single shot using a single detector. Termed Radiofrequency Encoded Angular-resolved Light Scattering (REALS), this technique uses one-to-one radiofrequency-to-angle mapping to measure angular dependence of light scattered from particles in a single shot using a photomultiplier tube. Using this technique it is possible to capture the continuous scattering profile over a wide dynamic range without mechanical scanning. This information allows us to characterize particle morphology and size with increased accuracy and high throughput, enabling label-free and high speed flow cytometry. As a proof of concept we distinguish the radius of tapered silica fiber over a range of radii.

High-throughput real-time optical sensing and imaging instruments for capture and analysis of fast phenomena are among the most essential tools for scientific, industrial, military, and most importantly biomedical applications. The key challenge in these instruments is the fundamental trade-off between speed and sensitivity of the measurement system due to the limited signal energy collected in each measurement window. Based on two enabling technologies, namely photonic time-stretch dispersive Fourier transform and optical amplification, we developed several novel high-throughput optical measurement tools for applications such as flow cytometry, vibrometry, and volumetric scanning.

We demonstrated optical Raman amplification at about 800 nm wavelength for the first time and extended time-stretch dispersive Fourier transform to this region of electromagnetic spectrum. We used this enabling technology to make an ultrafast three-dimensional laser scanner with about hundred thousand scans per second and an imaging vibrometer with nanometer-scale axial resolution.

We also employed our high-speed laser scanner to perform label-free cell screening in flow. One of the fundamental challenges in cell analysis is the undesirable impact of cell labeling on cellular behavior. To eliminate the need for these labels, while keeping the cell classification accuracy high, additional label-free parameters such as precise measurement of the cell protein concentration is required. We introduced a high-accuracy label-free imaging flow cytometer based on simultaneous measurement of morphology and optical path length through the cell at flow speeds as high as a few meters per second.

Finally, the ultimate challenge in ultra-high-throughput instrumentation is the storage and analysis of the torrent of generated data. As an example, our imaging flow cytometer generates about ten terabytes of cell images over a course of one hour acquisition, which captures images of every single cell in more than two milliliters of sample e.g. blood. We enabled practical use of these big data volumes by efficient combination of analog preprocessing techniques such as quadrature demodulation with parallel storage and digital post-processing.

We discuss a new paradigm wherein physical phenomena coded as algorithms perform computational imaging tasks. Vision Enhancement via Virtual diffraction and coherent Detection (VEViD) is a high-performance low-light-level and color enhancement tool that emerges from this paradigm. VEViD reimagines a digital image as a spatially varying metaphoric “lightfield” and subjects this field to physical processes akin to diffraction and coherent detection. The term “Virtual” captures the deviation from the physical world. The lightfield is pixelated and the propagation imparts a phase with dependence on spatial frequency. This phase, as opposed to intensity, represents the output image. The algorithm is extremely fast, interpretable, and reduces to a compact and intuitively-appealing mathematical expression. We demonstrate image enhancement of 4K resolution video at over 200 frames per second and show the utility of this physical algorithm in improving the accuracy of object detection in low-light conditions by neural networks.

Photonic time-stretch, invented at UCLA, has established world’s fastest real-time spectrometers and cameras with applications in biological cell screening, tomography, microfluidics, velocimetry and vibrometry. Time stretch instruments have led to several scientific breakthroughs including the discoveries of optical rogue waves, relativistic electron bunching in synchrotrons, and first ever observations of the birth of laser mode-locking and internal motion of soliton molecules. In time-stretch imaging, the target’s spatial information is encoded in the spectrum of the ultrafast laser pulses, which is stretched in time and then detected by a single-pixel detector and digitized by a real-time ADC, and processed by a CPU or a dedicated FPGA or GPU. Various methods have been proposed to realize time stretch, including single mode fibers, dispersion compensating fibers, chirped Bragg grating, and chromo-modal dispersion. However, none of those methods provide chirp with a large time-bandwidth product, which limits the time/depth range the pulse can measure.

In this study, we demonstrate a discrete time-stretch method that can generate the giant time-bandwidth product with arbitrary nonlinear chirp for operating wavelength from the visible to the infrared. We show its application in warped-stretch (foveated) imaging and a time-of-flight LIDAR with ∼MHz refresh rate.

Most optical sensing and measurement techniques suffer from the limited dynamic range. In

a second and related project, we have proposed the concept of optical dynamic range compression. This powerful technique provides a mean to match the dynamic range of the signal to that of the detector and data converter, leading to improved signal to noise ratio and a wider dynamic range. We outline various methods to implement optical dynamic range compression using nonlinear optics, silicon photonics and saturated amplifications.

Phase Stretch Transform (PST) is a physics-inspired computational approach developed in Jalali lab for feature enhancement in images. Here, it is applied to medical images. The results of its application to X-rays leads to development of an assistance tool for diagnosis of pneumothorax in X-ray images. The tool, which is first-of-its-kind, helps in locating the boundary of a collapsed lung, a life-critical clinical examination which is otherwise difficult for a radiologist to locate with a naked eye.

Additionally, PST is applied to other medical images, such as histology and mammograms, to demonstrate feature enhancement. The resulting edge detection map offers promising application in segmentation and analysis of medical images which is explored here. Further, texture segmentation using PST is also demonstrated.