arXiv subjects

Haixin Liu

Publications and source records attributed to Haixin Liu.

16 recordsLinked to original sources

arXiv2026

Nanophotonic dispersion engineering in tantala-based microresonators with ultralow-loss claddings

Integrated photonics optimized to leverage low loss, nonlinear-optical processes and optoelectronic integration requires robust optical cladding materials and associated fabrication process flows. We develop low-loss silicon dioxide (SiO$_2$) claddings with tantalum pentoxide (Ta$_2$O$_5$, tantala) integrated photonics to realize robust, high intrinsic quality factor ($Q_i$) microresonators and photonic-crystal resonators (PhCRs) for soliton microcomb generation. Specifically, we introduce a full-wafer, low-temperature process compatible with two tantala-based films: pure tantala and an amorphous metal-oxide mixture of titanium dioxide (TiO$_2$, titania) and tantala. Combining tantala with titania enhances the film's resistance to damage from the SiO$_2$ cladding deposition process, offering access to high $Q_i$ above $4 \times 10^6$ in fully cladded microresonators. Our platform supports low-loss edge couplers, group-velocity dispersion engineering by waveguide geometry, access to nanophotonic structures particularly to control phase-matching, and high dimensional tolerance in devices across an entire wafer. Using the platform, we explore the generation of high-efficiency, dark-soliton microcombs with repetition frequency from 50 GHz to 500 GHz, which supports applications such as photonic artificial intelligence acceleration, electronic signaling, and optical data communication.

physics.optics↗

arXiv2025

Amorphous metal oxide mixtures for high-Q integrated nonlinear photonics

Choice of material is ubiquitous in integrated photonics to design device properties, whereas changing material composition is much less common. With phase matching as an additional objective, constraints in depositing and patterning thin films limit the use of integrated nonlinear photonics to only select materials. Here, we explore an amorphous metal oxide mixture of titania (TiO2) and tantala (Ta2O5) in which material composition is a tool to enhance and customize photonics properties. In particular, the inclusion of titania reduces the oxygen-defect density in a tantala film, while maintaining a comparable Kerr nonlinear index. With ion-beam sputtering at room temperature, we deposit a thick, ultralow-loss titania-tantala film, and we nanopattern it to create microresonator frequency combs. Titania-tantala microresonators offer lower loss, higher index of refraction, and reduced optical absorption and photorefractive effects. Specifically, the titania-tantala mixture enables microresonator quality factor up to $10^7$ and a direct factor of 1.7 reduction in optical absorption. Our work demonstrates that composition in metal-oxide mixtures is a design parameter alongside the nanofabrication process and photonics design for integrated nonlinear photonics.

physics.optics↗

arXiv2025

Next-generation soliton frequency combs in photonic-crystal and nanocomposite microresonators

Microresonator frequency combs offer tremendous opportunity to advance applications in fundamental research and technology by linking the optical and microwave frequency domains. Kerr-nonlinear microresonators further enable the generation of portable, integrated optical frequency combs, which are called microcombs. However, the dispersion engineering usually suffers from the small, geometric parameter space, and achieving soliton microcombs is challenging and usually requires complicated experimental techniques and setups. In recent years, the invention of photonic-crystal resonators (PhCRs) provides access to solitons in a convenient and stable way while its mechanism has not been fully understood. In this article, we highlight the perspectives of generating solitons for various applications in PhCRs and give a thorough understanding of the dynamics of soliton formation. We also propose a nanocomposite waveguide structure for the optimization of group velocity dispersion (GVD), lifting the limitation on geometric parameter space. We apply it to the design of the pump-harmonic microcomb, a new concept for f-2f self-referencing in microcombs. Inspired by the pulse-driven microresonator, we provide a scheme of two-microresonator network, retaining the convenience of using a continuous wave laser for microcomb generation. Our work depicts a blueprint of achieving the next generation soliton frequency combs in PhCR and nanocomposite microresonators and highlights their great prospects in optical metrology, precision measurement and optical data transmission.

physics.optics↗

arXiv2025

LogicCat: A Chain-of-Thought Text-to-SQL Benchmark for Complex Reasoning

Text-to-SQL is a critical task in natural language processing that aims to transform natural language questions into accurate and executable SQL queries. In real-world scenarios, these reasoning tasks are often accompanied by complex mathematical computations, domain knowledge, and hypothetical reasoning scenarios. However, existing large-scale Text-to-SQL datasets typically focus on business logic and task logic, neglecting critical factors such as vertical domain knowledge, complex mathematical reasoning, and hypothetical reasoning, which are essential for realistically reflecting the reasoning demands in practical applications and completing data querying and analysis. To bridge this gap, we introduce LogicCat, the first Text-to-SQL benchmark dataset specifically designed for complex reasoning and chain-of-thought parsing, encompassing physics, arithmetic, commonsense, and hypothetical reasoning scenarios. LogicCat comprises 4,038 English questions paired 12,114 detailed chain-of-thought reasoning steps, spanning 45 databases across diverse domains, significantly surpassing existing datasets in complexity. Experimental results demonstrate that LogicCat substantially increases the task difficulty for current state-of-the-art models to at most 33.20% execution accuracy, indicating that this task remains exceptionally challenging. The advancement of LogicCat represents a crucial step toward developing systems suitable for real-world enterprise data analysis and autonomous query generation. We have released our dataset code at https://github.com/Ffunkytao/LogicCat.

cs.CL↗

arXiv2025

Activating high-power parametric oscillation in photonic-crystal resonators

By engineering the mode spectrum of a Kerr microresonator, we selectively activate nonlinear phase matching amongst broadband parametric gain. At threshold, optical parametric oscillators (OPOs) emerge from vacuum fluctuations in the presence of a pump laser, and above threshold, OPOs seed the formation of intraresonator patterns and states, such as chaos and solitons. These competing nonlinear processes hinder an important application of OPOs as wavelength-variable, low-noise sources. Recently, nanopatterned microresonator OPOs have leveraged photonic crystal bandgaps to enable universal phase matching and control of nonlinear interactions. Here, we explore a design paradigm optimized for high-output power that uses geometric dispersion to suppress nonlinear interactions and a photonic crystal bandgap to activate only a single OPO interaction. Our devices convert an input pump laser to output signal and idler waves with powers exceeding 40 mW while maintaining spectral purity and side-mode suppression ratios greater than 40 dB. We show that this approach suits custom wavelengths by measuring four independent oscillators that vary only photonic crystal parameters to select output waves. Our experiments demonstrate that microresonators functionalized by photonic crystals offer a versatile and lossless palette of controls for nonlinear laser conversion.

physics.optics↗

arXiv2025

Implementing photonic-crystal resonator frequency combs in a photonics foundry

We explore an AIM Photonics silicon-nitride platform to fabricate photonic-crystal resonators for generating optical parametric oscillators (OPO) and soliton microcombs. Our approach leverages the scalability and fine feature size of silicon-nitride processing on large-scale silicon wafers to achieve low-loss, high-Q microresonators, functionalized by nano-scale photonic-crystal structures. We demonstrate intrinsic microresonator quality factor up to 1.2*10^7 with complete foundry fabrication on 300 mm silicon, a 700 nm thick silicon-nitride device layer, and inclusion of complex nanophotonics. These features enable a host of nonlinear nanophotonics sources on the platform, including OPOs, microcombs, parametric amplifiers, squeezed-light generators, and single-photon sources. By fine-tuning the photonic-crystal design parameters, we achieve broad tunability in the frequency of the OPO output, spanning a significant portion of the near-infrared. Additionally, we observe the formation of soliton frequency combs, enabled by the precise dispersion engineering of the microresonators. These results highlight the potential of widely accessible, photolithographically patterned, silicon-nitride photonics to enable wide access to and complex integration of frequency-comb sources, with applications in spectroscopy, metrology, and communications.

physics.optics↗

arXiv2024

CMoralEval: A Moral Evaluation Benchmark for Chinese Large Language Models

What a large language model (LLM) would respond in ethically relevant context? In this paper, we curate a large benchmark CMoralEval for morality evaluation of Chinese LLMs. The data sources of CMoralEval are two-fold: 1) a Chinese TV program discussing Chinese moral norms with stories from the society and 2) a collection of Chinese moral anomies from various newspapers and academic papers on morality. With these sources, we aim to create a moral evaluation dataset characterized by diversity and authenticity. We develop a morality taxonomy and a set of fundamental moral principles that are not only rooted in traditional Chinese culture but also consistent with contemporary societal norms. To facilitate efficient construction and annotation of instances in CMoralEval, we establish a platform with AI-assisted instance generation to streamline the annotation process. These help us curate CMoralEval that encompasses both explicit moral scenarios (14,964 instances) and moral dilemma scenarios (15,424 instances), each with instances from different data sources. We conduct extensive experiments with CMoralEval to examine a variety of Chinese LLMs. Experiment results demonstrate that CMoralEval is a challenging benchmark for Chinese LLMs. The dataset is publicly available at \url{https://github.com/tjunlp-lab/CMoralEval}.

cs.CL↗

arXiv2024

Nanophotonic oscillators for laser conversion beyond an octave

Many uses of lasers place the highest importance on access to specific wavelength bands. For example, mobilizing optical-atomic clocks for a leap in sensing requires compact lasers at frequencies spread across the visible and near infrared. Integrated photonics enables high-performance, scalable laser platforms, however, customizing laser-gain media to support wholly new bands is challenging and often prohibitively mismatched in scalability to early quantum-based sensing and information systems. Here, we demonstrate a microresonator optical-parametric oscillator (OPO) that converts a pump laser to an output wave within a frequency span exceeding an octave. We achieve phase matching for oscillation by nanopatterning the microresonator to open a photonic-crystal bandgap on the mode of the pump laser. By adjusting the nanophotonic pattern and hence the bandgap, the ratio of output OPO wave frequency span to required pump laser tuning is more than 10,000. We also demonstrate tuning the oscillator in free-spectral-range steps, more finely with temperature, and minimal additive frequency noise of the laser-conversion process. Our work shows that nanophotonics offers control of laser conversion in microresonators, bridging phase-matching of nonlinear optics and application requirements for laser designs.

physics.optics↗

arXiv2024

Foundry manufacturing of octave-spanning microcombs

Soliton microcombs provide a chip-based, octave-spanning source for self-referencing and optical metrology. We explore use of a silicon-nitride integrated photonics foundry to manufacture octave-spanning microcombs. By group-velocity dispersion engineering with the waveguide cross-section, we shape the soliton spectrum for dispersive-wave spectral enhancements at the frequencies for f-2f self-referencing. With the optimized waveguide geometry, we control the carrier-envelope offset frequency by adjusting the resonator radius. Moreover, we demonstrate the other considerations for octave microcombs, including models for soliton spectrum design, ultra-broadband resonator external coupling, low-loss edge couplers, and the nonlinear self-interactions of few-cycle solitons. This design process permits highly repeatable creation of soliton microcombs optimized for pump operation less than 100 mW, an electronically detectable offset frequency, and high comb mode power for f-2f detection. However, these design aspects must also be made compatible with the foundry fabrication tolerance of octave microcomb devices. Our experiments highlight the potential to manufacture a single-chip solution for an octave-spanning microcomb, which is the central component of a compact microsystem for optical metrology.

physics.optics↗

arXiv2024

Laser-power consumption of soliton formation in a bidirectional Kerr resonator

Laser sources power extreme data transmission as well as computing acceleration, access to ultrahigh-speed signaling, and sensing for chemicals, distance, and pattern recognition. The ever-growing scale of these applications drives innovation in multi-wavelength lasers for massively parallel processing. We report a nanophotonic Kerr-resonator circuit that consumes the power of an input laser and generates a soliton frequency comb at approaching unit efficiency. By coupling forward and backward propagation, we realize a bidirectional Kerr resonator that supports universal phase matching but also opens excess loss by double-sided emission. Therefore, we induce reflection of the resonator's forward, external-coupling port to favor backward propagation, resulting in efficient, one-sided soliton formation. Coherent backscattering with nanophotonics provides the control to put arbitrary phase-matching and efficient laser-power consumption on equal footing in Kerr resonators. In the overcoupled-resonator regime, we measure 65% conversion efficiency of a 40 mW input pump laser, and the nonlinear circuit consumes 97% of the pump, generating the maximum possible comb power. Our work opens up high-efficiency soliton formation in integrated photonics, exploring how energy flows in nonlinear circuits and enabling laser sources for advanced transmission, computing, quantum sensing, and artificial-intelligence applications.

physics.optics↗

arXiv2023

Threshold and laser-conversion in nanostructured-resonator parametric oscillators

We explore optical parametric oscillation (OPO) in nanophotonic resonators, enabling arbitrary, nonlinear phase-matching and nearly lossless control of energy conversion. Such pristine OPO laser converters are determined by nonlinear light-matter interactions, making them both technologically flexible and broadly reconfigurable. We utilize a nanostructured inner-wall modulation in the resonator to achieve universal phase-matching for OPO-laser conversion, but coherent backscattering also induces a counterpropagating pump laser. This depletes the intra-resonator optical power in either direction, increasing the OPO threshold power and limiting laser-conversion efficiency, the ratio of optical power in target signal and idler frequencies to the pump. We develop an analytical model of this system that emphasizes an understanding of optimal laser conversion and threshold behaviors, and we use the model to guide experiments with nanostructured-resonator OPO laser-conversion circuits, fully integrated on chip and unlimited by group-velocity dispersion. Our work demonstrates the fundamental connection between OPO laser-conversion efficiency and the resonator coupling rate, subject to the relative phase and power of counterpropagating pump fields. We achieve $(40\pm4)$ mW of on-chip power, corresponding to $(41\pm4)$% conversion efficiency, and discover a path toward near-unity OPO laser conversion efficiency.

physics.optics↗

arXiv2022

Tunable lasers with optical-parametric oscillation in photonic-crystal resonators

By design access to laser wavelength, especially with integrated photonics, is critical to advance quantum sensors like optical clocks and quantum-information systems, and open opportunities in optical communication. Semiconductor-laser gain provides exemplary efficiency and integration but merely in developed wavelength bands. Alternatively, nonlinear optics requires control of phase matching, but the principle of nonlinear conversion of a pump laser to a designed wavelength is extensible. We report on laser-wavelength access by versatile customization of optical-parametric oscillation (OPO) with a photonic-crystal resonator (PhCR). By controlling the bandgap of a PhCR, we enable OPO generation across a wavelength range of 1234-2093 nm with a 1550 nm pump and 1016-1110 nm with a 1064 nm pump. Moreover, our tunable laser platform offers pump-to-sideband conversion efficiency of >10% and negligible additive optical-frequency noise across the output range. From laser design to simulation of nonlinear dynamics, we use a Lugiato-Lefever framework that predicts the system characteristics, including bi-directional OPO generation in the PhCR and conversion efficiency in agreement with our observations. Our experiments introduce tunable lasers by design with PhCR OPOs, providing critical functionalities in integrated photonics.

physics.optics↗

arXiv2021

First-Generation Inference Accelerator Deployment at Facebook

In this paper, we provide a deep dive into the deployment of inference accelerators at Facebook. Many of our ML workloads have unique characteristics, such as sparse memory accesses, large model sizes, as well as high compute, memory and network bandwidth requirements. We co-designed a high-performance, energy-efficient inference accelerator platform based on these requirements. We describe the inference accelerator platform ecosystem we developed and deployed at Facebook: both hardware, through Open Compute Platform (OCP), and software framework and tooling, through Pytorch/Caffe2/Glow. A characteristic of this ecosystem from the start is its openness to enable a variety of AI accelerators from different vendors. This platform, with six low-power accelerator cards alongside a single-socket host CPU, allows us to serve models of high complexity that cannot be easily or efficiently run on CPUs. We describe various performance optimizations, at both platform and accelerator level, which enables this platform to serve production traffic at Facebook. We also share deployment challenges, lessons learned during performance optimization, as well as provide guidance for future inference hardware co-design.

cs.AR↗

arXiv2021

Low-Precision Hardware Architectures Meet Recommendation Model Inference at Scale

Tremendous success of machine learning (ML) and the unabated growth in ML model complexity motivated many ML-specific designs in both CPU and accelerator architectures to speed up the model inference. While these architectures are diverse, highly optimized low-precision arithmetic is a component shared by most. Impressive compute throughputs are indeed often exhibited by these architectures on benchmark ML models. Nevertheless, production models such as recommendation systems important to Facebook's personalization services are demanding and complex: These systems must serve billions of users per month responsively with low latency while maintaining high prediction accuracy, notwithstanding computations with many tens of billions parameters per inference. Do these low-precision architectures work well with our production recommendation systems? They do. But not without significant effort. We share in this paper our search strategies to adapt reference recommendation models to low-precision hardware, our optimization of low-precision compute kernels, and the design and development of tool chain so as to maintain our models' accuracy throughout their lifespan during which topic trends and users' interests inevitably evolve. Practicing these low-precision technologies helped us save datacenter capacities while deploying models with up to 5X complexity that would otherwise not be deployed on traditional general-purpose CPUs. We believe these lessons from the trenches promote better co-design between hardware architecture and software engineering and advance the state of the art of ML in industry.

cs.LG↗

arXiv2021

FBGEMM: Enabling High-Performance Low-Precision Deep Learning Inference

Deep learning models typically use single-precision (FP32) floating point data types for representing activations and weights, but a slew of recent research work has shown that computations with reduced-precision data types (FP16, 16-bit integers, 8-bit integers or even 4- or 2-bit integers) are enough to achieve same accuracy as FP32 and are much more efficient. Therefore, we designed fbgemm, a high-performance kernel library, from ground up to perform high-performance quantized inference on current generation CPUs. fbgemm achieves efficiency by fusing common quantization operations with a high-performance gemm implementation and by shape- and size-specific kernel code generation at runtime. The library has been deployed at Facebook, where it delivers greater than 2x performance gains with respect to our current production baseline.

cs.LG↗

arXiv2019

Probabilistic quantum cloning of N quantum states

Probabilistically creating n perfect clones from m copies for one of N priori known quantum states with minimum failure probability is a long-standing problem. We provide a rigorous proof for the geometric approach to this probabilistic quantum cloning problem when N = 2. Then, we give the general geometric form of the sufficient and necessary condition of probabilistic cloning for N known quantum states. By this general geometric approach, we realize the optimal probabilistic quantum cloning of N known quantum states with priori probabilities. The results are also applicable to the identification of those N quantum states.

quant-ph↗