Searcharxiv⌕ Search

arXiv subjects

Taewon Park

Publications and source records attributed to Taewon Park.

15 recordsLinked to original sources
arXiv2026

Practical Limits on Integrated Squeezers

Recent experiments have demonstrated the successful generation and detection of moderately squeezed vacuum states with integrated photonics. However, in order to benefit from the reduced noise of highly squeezed light, many different noise sources must be mitigated. Here, we quantify the fundamental limits these noise sources impose on squeezing measurements and find surprising generality across different platforms and designs. We combine these different limitations into a simple model that provides practical guidance for the design and benchmarking of next-generation integrated squeezed-light systems.

quant-ph↗
arXiv2026

Ultra-wideband electrically-tuned mid-infrared on-chip parametric oscillator

Developing compact, broadband mid-infrared coherent sources for applications in spectroscopy and sensing remains a pressing challenge in photonics. However, material limitations and integration constraints have restricted the accessible wavelengths and operation bandwidths of current mid-infrared lasers. Here, we address these challenges by developing a nonlinear integrated photonic device that converts a fixed-wavelength near-infrared pump laser into broadly tunable mid-infrared light. Our device, an optical parametric oscillator (OPO) integrated on thin-film lithium niobate, generates 22 THz of multi-milliwatt, voltage-tunable radiation from 2.7-3.4 microns, a region typically difficult to access but vital for environmental, chemical, and biological sensing. By introducing an on-chip-tunable OPO architecture taking advantage of the Vernier effect, we obtain electrical control of the emission wavelengths from coarse, multi-THz scales down to continuous, sub-100-GHz mode-hop-free tuning ranges. This work establishes a robust platform for a new class of compact, widely tunable mid-infrared sources with potential for future scaling.

physics.optics↗
arXiv2026

Ultrabroadband Passive Laser Noise Suppression to Quantum Noise Limit through on-chip Second Harmonic Generation

Laser intensity noise limits performance in quantum sensing, metrology, and computing. Existing stabilization methods face a trade-off between bandwidth and complexity: electronic feedback loops are speed-limited, while optical resonators are constrained by narrow linewidths and locking requirements. Here, we demonstrate an all-optical "noise eater" that passively suppresses intensity fluctuations from DC to >10 gigahertz. By leveraging high-efficiency second-harmonic generation in nanophotonic lithium niobate waveguides, we operate at a pump-depletion stationary point where input fluctuations are decoupled from the output to first order. This passive and nonresonant nanophotonic device suppresses relative intensity noise by 25 to 60 dB over the full measurement bandwidth and stabilizes a noisy fiber amplifier output to the shot-noise limit. Our results establish a scalable, wide-bandwidth paradigm for laser stabilization essential for high-throughput quantum technologies and deployable photonic sensing systems.

physics.optics↗
arXiv2025

Low-power integrated optical parametric amplification via second-harmonic resonance

Optical amplifiers are fundamental to modern photonics, enabling long-distance communications, precision sensing, and quantum information processing. Erbium-doped amplifiers dominate telecommunications but are restricted to specific wavelength bands, while semiconductor amplifiers offer broader coverage but suffer from high noise and nonlinear distortions. Optical parametric amplifiers (OPAs) promise broadband, quantum-limited amplification across arbitrary wavelengths. However, their miniaturization and deployment has been hampered by watt-level power requirements. Here we demonstrate an integrated OPA on thin-film lithium niobate that achieves >17 dB gain with <200 mW input power -- an order of magnitude improvement over previous demonstrations. Our second-harmonic-resonant design enhances both pump generation efficiency (95% conversion) and pump power utilization through recirculation, without sacrificing bandwidth. The resonant architecture increases the effective pump power by nearly an order of magnitude compared to conventional single-pass designs, while also multiplexing the signal and pump. We demonstrate flat near-quantum-limited noise performance over 110 nm. Our low-power architecture enables practical on-chip OPAs for next generation quantum and classical photonics.

physics.optics↗
arXiv2025

Low-loss, fabrication-tolerant, and highly-tunable Sagnac loop reflectors and Fabry-Pérot cavities on thin-film lithium niobate

We present low-loss ($<1.5\%$) and power-efficient Mach-Zehnder interferometers (MZIs) on thin-film lithium niobate. To accurately measure low MZI losses, we develop a self-calibrated method using tunable Sagnac loop reflectors (SLRs) to build cavities. Fabry-Pérot cavities constructed from these fabrication-tolerant SLRs achieve an intrinsic quality factor of $2 \times 10^6$. By implementing thermal isolation trenches, we also demonstrate a $>10\times$ reduction in power consumption for thermo-optic phase shifters, achieving a $π$-phase shift ($P_π$) with just 2.5 mW. These tunable and efficient components are key for scaling up to complex photonic integrated circuits.

physics.optics↗
arXiv2024

Discrete Dictionary-based Decomposition Layer for Structured Representation Learning

Neuro-symbolic neural networks have been extensively studied to integrate symbolic operations with neural networks, thereby improving systematic generalization. Specifically, Tensor Product Representation (TPR) framework enables neural networks to perform differentiable symbolic operations by encoding the symbolic structure of data within vector spaces. However, TPR-based neural networks often struggle to decompose unseen data into structured TPR representations, undermining their symbolic operations. To address this decomposition problem, we propose a Discrete Dictionary-based Decomposition (D3) layer designed to enhance the decomposition capabilities of TPR-based models. D3 employs discrete, learnable key-value dictionaries trained to capture symbolic features essential for decomposition operations. It leverages the prior knowledge acquired during training to generate structured TPR representations by mapping input data to pre-learned symbolic features within these dictionaries. D3 is a straightforward drop-in layer that can be seamlessly integrated into any TPR-based model without modifications. Our experimental results demonstrate that D3 significantly improves the systematic generalization of various TPR-based models while requiring fewer additional parameters. Notably, D3 outperforms baseline models on the synthetic task that demands the systematic decomposition of unseen combinatorial data.

cs.LG↗
arXiv2024

Attention-based Iterative Decomposition for Tensor Product Representation

In recent research, Tensor Product Representation (TPR) is applied for the systematic generalization task of deep neural networks by learning the compositional structure of data. However, such prior works show limited performance in discovering and representing the symbolic structure from unseen test data because their decomposition to the structural representations was incomplete. In this work, we propose an Attention-based Iterative Decomposition (AID) module designed to enhance the decomposition operations for the structured representations encoded from the sequential input data with TPR. Our AID can be easily adapted to any TPR-based model and provides enhanced systematic decomposition through a competitive attention mechanism between input features and structured representations. In our experiments, AID shows effectiveness by significantly improving the performance of TPR-based prior works on the series of systematic generalization tasks. Moreover, in the quantitative and qualitative evaluations, AID produces more compositional and well-bound structural representations than other works.

cs.LG↗
arXiv2023

Single-Mode Squeezed Light Generation and Tomography with an Integrated Optical Parametric Oscillator

Quantum optical technologies promise advances in sensing, computing, and communication. A key resource is squeezed light, where quantum noise is redistributed between optical quadratures. We introduce a monolithic, chip-scale platform that exploits the $χ^{(2)}$ nonlinearity of a thin-film lithium niobate (TFLN) resonator device to efficiently generate squeezed states of light. Our system integrates all essential components -- except for the laser and two detectors -- on a single chip with an area of one square centimeter, significantly reducing the size, operational complexity, and power consumption associated with conventional setups. Our work addresses challenges that have limited previous integrated nonlinear photonic implementations that rely on either $χ^{(3)}$ nonlinear resonators or on integrated waveguide $χ^{(2)}$ parametric amplifiers. Using the balanced homodyne measurement subsystem that we implemented on the same chip, we measure a squeezing of 0.55 dB and an anti-squeezing of 1.55 dB. We use 20 mW of input power to generate the parametric oscillator pump field by employing second harmonic generation on the same chip. Our work represents a substantial step toward compact and efficient quantum optical systems posed to leverage the rapid advances in integrated nonlinear and quantum photonics.

quant-ph↗
arXiv2023

Mid-infrared spectroscopy with a broadly tunable thin-film lithium niobate optical parametric oscillator

Mid-infrared spectroscopy, an important and widespread technique for sensing molecules, has encountered barriers stemming from sources either limited in tuning range or excessively bulky for practical field use. We present a compact, efficient, and broadly tunable optical parametric oscillator (OPO) device surmounting these challenges. Leveraging a dispersion-engineered singly-resonant OPO implemented in thin-film lithium niobate-on-sapphire, we achieve broad and controlled tuning over an octave, from 1.5 to 3.3 microns by combining laser and temperature tuning. The device generates > 25 mW of mid-infrared light at 3.2 microns, offering a power conversion efficiency of 15% (45% quantum efficiency). We demonstrate the tuning and performance of the device by successfully measuring the spectra of methane and ammonia, verifying our approach's relevance for gas sensing. Our device signifies an important advance in nonlinear photonics miniaturization and brings practical field applications of high-speed and broadband mid-infrared spectroscopy closer to reality.

physics.optics↗
arXiv2023

Integrated frequency-modulated optical parametric oscillator

Optical frequency combs have revolutionized precision measurement, time-keeping, and molecular spectroscopy. A substantial effort has developed around "microcombs": integrating comb-generating technologies into compact, reliable photonic platforms. Current approaches for generating these microcombs involve either the electro-optic (EO) or Kerr mechanisms. Despite rapid progress, maintaining high efficiency and wide bandwidth remains challenging. Here, we introduce a new class of microcomb -- an integrated optical frequency comb generator that combines electro-optics and parametric amplification to yield a frequency-modulated optical parametric oscillator (FM-OPO). In stark contrast to EO and Kerr combs, the FM-OPO microcomb does not form pulses but maintains operational simplicity and highly efficient pump power utilization with an output resembling a frequency-modulated laser. We outline the working principles of FM-OPO and demonstrate them by fabricating the complete optical system in thin-film lithium niobate (LNOI). We measure pump to comb internal conversion efficiency exceeding 93% (34% out-coupled) over a nearly flat-top spectral distribution spanning approximately 1,000 modes (approximately 6 THz). Compared to an EO comb, the cavity dispersion rather than loss determines the FM-OPO bandwidth, enabling broadband combs with a smaller RF modulation power. The FM-OPO microcomb, with its robust operational dynamics, high efficiency, and large bandwidth, contributes a new approach to the field of microcombs and promises to herald an era of miniaturized precision measurement, and spectroscopy tools to accelerate advancements in metrology, spectroscopy, telecommunications, sensing, and computing.

physics.optics↗
arXiv2022

Integrated Quantum Optical Phase Sensor

The quantum noise of light fundamentally limits optical phase sensors. A semiclassical picture attributes this noise to the random arrival time of photons from a coherent light source such as a laser. An engineered source of squeezed states suppresses this noise and allows sensitivity beyond the standard quantum limit (SQL) for phase detection. Advanced gravitational wave detectors like LIGO have already incorporated such sources, and nascent efforts in realizing quantum biological measurements have provided glimpses into new capabilities emerging in quantum measurement. We need ways to engineer and use quantum light within deployable quantum sensors that operate outside the confines of a lab environment. Here we present a photonic integrated circuit fabricated in thin-film lithium niobate that provides a path to meet these requirements. We use the second-order nonlinearity to produce a squeezed state at the same frequency as the pump light and realize circuit control and sensing with electro-optics. Using a 26.2 milliwatts of optical power, we measure (2.7 $\pm$ 0.2 )$\%$ squeezing and apply it to increase the signal-to-noise ratio of phase measurement. We anticipate that on-chip photonic systems like this, which operate with low power and integrate all of the needed functionality on a single die, will open new opportunities for quantum optical sensing.

quant-ph↗
arXiv2022

Experimental evaluation of digitally-verifiable photonic computing for blockchain and cryptocurrency

As blockchain technology and cryptocurrency become increasingly mainstream, ever-increasing energy costs required to maintain the computational power running these decentralized platforms create a market for more energy-efficient hardware. Photonic cryptographic hash functions, which use photonic integrated circuits to accelerate computation, promise energy efficiency for verifying transactions and mining in a cryptonetwork. Like many analog computing approaches, however, current proposals for photonic cryptographic hash functions that promise similar security guarantees as Bitcoin are susceptible to systematic error, so multiple devices may not reach a consensus on computation despite high numerical precision (associated with low photodetector noise). In this paper, we theoretically and experimentally demonstrate that a more general family of robust discrete analog cryptographic hash functions, which we introduce as LightHash, leverages integer matrix-vector operations on photonic mesh networks of interferometers. The difficulty of LightHash can be adjusted to be sufficiently tolerant to systematic error (calibration error, loss error, coupling error, and phase error) and preserve inherent security guarantees present in the Bitcoin protocol. Finally, going beyond our proof-of-concept, we define a ``photonic advantage'' criterion and justify how recent developments in CMOS optoelectronics (including analog-digital conversion) provably achieve such advantage for robust and digitally-verifiable photonic computing and ultimately generate a new market for decentralized photonic technology.

cs.ET↗
arXiv2022

Experimentally realized in situ backpropagation for deep learning in nanophotonic neural networks

Neural networks are widely deployed models across many scientific disciplines and commercial endeavors ranging from edge computing and sensing to large-scale signal processing in data centers. The most efficient and well-entrenched method to train such networks is backpropagation, or reverse-mode automatic differentiation. To counter an exponentially increasing energy budget in the artificial intelligence sector, there has been recent interest in analog implementations of neural networks, specifically nanophotonic neural networks for which no analog backpropagation demonstration exists. We design mass-manufacturable silicon photonic neural networks that alternately cascade our custom designed "photonic mesh" accelerator with digitally implemented nonlinearities. These reconfigurable photonic meshes program computationally intensive arbitrary matrix multiplication by setting physical voltages that tune the interference of optically encoded input data propagating through integrated Mach-Zehnder interferometer networks. Here, using our packaged photonic chip, we demonstrate in situ backpropagation for the first time to solve classification tasks and evaluate a new protocol to keep the entire gradient measurement and update of physical device voltages in the analog domain, improving on past theoretical proposals. Our method is made possible by introducing three changes to typical photonic meshes: (1) measurements at optical "grating tap" monitors, (2) bidirectional optical signal propagation automated by fiber switch, and (3) universal generation and readout of optical amplitude and phase. After training, our classification achieves accuracies similar to digital equivalents even in presence of systematic error. Our findings suggest a new training paradigm for photonics-accelerated artificial intelligence based entirely on a physical analog of the popular backpropagation technique.

cs.ET↗
arXiv2021

Distributed Associative Memory Network with Memory Refreshing Loss

Despite recent progress in memory augmented neural network (MANN) research, associative memory networks with a single external memory still show limited performance on complex relational reasoning tasks. Especially the content-based addressable memory networks often fail to encode input data into rich enough representation for relational reasoning and this limits the relation modeling performance of MANN for long temporal sequence data. To address these problems, here we introduce a novel Distributed Associative Memory architecture (DAM) with Memory Refreshing Loss (MRL) which enhances the relation reasoning performance of MANN. Inspired by how the human brain works, our framework encodes data with distributed representation across multiple memory blocks and repeatedly refreshes the contents for enhanced memorization similar to the rehearsal process of the brain. For this procedure, we replace a single external memory with a set of multiple smaller associative memory blocks and update these sub-memory blocks simultaneously and independently for the distributed representation of input data. Moreover, we propose MRL which assists a task's target objective while learning relational information existing in data. MRL enables MANN to reinforce an association between input data and task objective by reproducing stochastically sampled input data from stored memory contents. With this procedure, MANN further enriches the stored representations with relational information. In experiments, we apply our approaches to Differential Neural Computer (DNC), which is one of the representative content-based addressing memory models and achieves the state-of-the-art performance on both memorization and relational reasoning tasks.

cs.LG↗