About

Juejun Hu is the John F. Elliott Professor of Materials Science and Engineering at MIT. His research group develops novel materials and devices that harness light-matter interactions for a broad range of applications, including on-chip sensing and spectroscopy. His work leverages digital Fourier transform technology to build miniaturized, rugged sensors compatible with mass production for industrial process control, medical imaging, and space systems. A major focus of his research is on optical phase-change materials and meta-optics. These materials exhibit reversible changes in their optical properties during solid-state phase transitions, enabling the creation of reconfigurable optical devices that can be programmably adapted to specific functionalities. Additional research directions include flexible and polymer photonics for biomedical monitoring and high-speed data communications, advanced imaging and sensing optics for consumer and automotive electronics, and magneto-optical isolation. His group is also developing chip-scale nonreciprocal photonic devices functioning as one-way valves for light, which are crucial for next-generation optical communication and navigation systems.

Research topics

• Computer Science
• Materials science
• Optoelectronics
• Nanotechnology
• Physics
• Optics
• Engineering physics
• Engineering
• Systems engineering
• Software engineering
• Chemistry
• Mechanical engineering
• Telecommunications
• Embedded system

Selected publications

• Development of a machine learning model to predict low vision aid fitting for visually impaired patients

  Frontiers in Medicine · 2026-01-12

  articleOpen accessSenior authorCorresponding

  Background: At present, the fitting of low-vision aids (LVA) for patients globally necessitates the intervention of highly skilled ophthalmologists and certified rehabilitation specialists. To mitigate this limitation, we employed machine learning algorithms to develop an artificial intelligence (AI)-based model for automated LVA fitting assistance. Patients and Methods: Clinical characteristics and diagnostic data from patients with low vision in southeastern China were collected between October 26, 2015, and October 6, 2021, to establish the training and test datasets. We developed and compared three machine learning models-Random Forest (RF), Deep Neural Network (DNN), and Logistic Regression (LR)-to predict prescriptions for three LVA categories selected based on compliance with the World Health Organization's basic specifications for assistive products: Distant Optical Visual aids (DOV), Near Electronic Visual aids (NEV), and Near Optical Visual aids (NOV). Hyperparameter optimization was conducted through four rounds of internal cross-validation. Following model training, the best-performing model was identified and subsequently validated on external data to assess its predictive accuracy and sensitivity. Results: The dataset comprised a total of 1,241 patients diagnosed with low vision. Our model displayed satisfactory performance in LVA fitting when evaluated on the test set. Comparative analysis revealed the RF model as the optimal choice, achieving area under the curve (AUC) values of 0.93 for DOV, 0.83 for NEV, and 0.91 for NOV. Furthermore, feature importance analysis derived from the RF model weights indicated that patient age, best-corrected visual acuity (BCVA of the left eye), and consultation year were the predominant factors influencing LVA fitting decisions across all three aid categories, while visual disability grade specifically impacted DOV prescriptions. In external validation involving 112 prospective cases, the model demonstrated performance comparable to that of a mid-career ophthalmologist (5 years' experience). Conclusion: This study identified significant associations between clinical characteristics and LVA prescription patterns. Leveraging historical LVA fitting data, we developed a machine learning-based decision support system capable of predicting optimal fittings for the three fundamental LVA categories. The proposed tool demonstrates potential for clinical application by generating data-driven prescription recommendations.

  PublisherOA PDFDOI

• BanaServe: Unified KV Cache and Dynamic Module Migration for Balancing Disaggregated LLM Serving in AI Infrastructure

  Software Practice and Experience · 2026-01-21

  articleOpen access

  ABSTRACT Objective Large Language Models (LLMs) are increasingly deployed in modern AI infrastructure, creating a strong demand for high‐throughput and resource‐efficient serving systems. Disaggregated LLM serving, which decouples prompt prefill from auto‐regressive decode to accommodate their heterogeneous compute and memory characteristics, has emerged as a promising architecture. However, existing disaggregated serving systems suffer from three fundamental limitations: static resource allocation that fails to adapt to highly dynamic workloads, severe load imbalance between compute‐bound prefill and memory‐bound decode stages, and prefix‐cache‐aware routing that skews load distribution and creates performance hotspots. These issues collectively limit resource utilization, scalability, and the ability to meet service level objectives (SLOs) under real‐world workloads. Methods To address these challenges, we propose BanaServe, a dynamic orchestration framework for disaggregated LLM serving that continuously rebalances both computational and memory resources across prefill and decode instances. BanaServe introduces three key mechanisms: (i) layer‐level weight migration to enable coarse‐grained redistribution of computation, (ii) attention‐level Key–Value (KV) cache migration for fine‐grained memory load balancing, and (iii) a Global KV Cache Store with layer‐wise overlapped transmission to decouple routing decisions from cache placement. Together, these mechanisms eliminate cache‐induced hotspots and allow routers to perform purely load‐aware scheduling with minimal latency overhead. BanaServe is implemented on top of state‐of‐the‐art LLM serving frameworks, including vLLM and DistServe. Results We evaluate BanaServe under diverse and challenging workloads, including long‐context inference, bursty request arrivals, and mixed prompt–generation patterns. Experimental results show that, compared to vLLM, BanaServe improves throughput by 1.2–3.9× and reduces total processing time by 3.9%–78.4%. In comparison with DistServe, BanaServe achieves 1.1–2.8× higher throughput while reducing latency by 1.4%–70.1%. These gains are consistent across workload variations, demonstrating BanaServe's robustness under highly dynamic serving conditions. Conclusion BanaServe demonstrates that dynamic, multi‐granularity resource rebalancing and cache‐decoupled routing are essential for efficient disaggregated LLM serving. By jointly addressing resource elasticity, stage imbalance, and cache‐induced load skew, BanaServe substantially improves throughput, latency, and resource utilization in real‐world deployments. This work provides a practical and scalable foundation for next‐generation LLM serving systems operating under dynamic and heterogeneous workloads.

  PublisherOA PDFDOI

• Bidirectional causal associations between immune cell phenotypes and age-related macular degeneration subtypes: A mendelian randomization study

  Experimental Gerontology · 2026-02-19

  articleOpen accessSenior authorCorresponding

  Age-related macular degeneration (AMD) is a leading cause of irreversible vision loss in the elderly population, characterized by two primary subtypes: dry AMD (dAMD) and wet AMD (wAMD). While immune mechanisms are implicated in AMD pathogenesis, the causal relationships between specific immune cell phenotypes and AMD subtypes remain incompletely characterized. Bidirectional two-sample Mendelian randomization (MR) analysis was conducted to evaluate causal associations between 731 immune cell phenotypes and AMD subtypes. Immune cell phenotype data were sourced from publicly available GWAS datasets, encompassing 3757 individuals of European descent. AMD data were obtained from the Finnish Finngen R10 database, including 5239 wAMD cases and 6651 dAMD cases. The inverse-variance weighted (IVW) method was used as the primary analytical approach, complemented by reverse MR analysis to assess bidirectional relationships. The MR analysis identified significant causal associations between specific immune cell phenotypes and AMD subtypes. For wAMD, the expression of HLA-DR on CD14 − CD16 + monocytes exhibited a positive association (OR = 1.11, P = 1.65 × 10 −5 ), whereas CD25 on CD28 + CD4 + T cells showed a negative association (OR = 0.87, P = 0.70 × 10 −4 ). In the case of dAMD, the expression of HLA-DR on CD14 + CD16 + monocytes was positively associated with disease risk (OR = 1.13, P = 0.84 × 10 −6 ). Reverse MR analysis revealed that dAMD negatively influenced the expression levels of Effector Memory CD4 − CD8 − T cells (OR = 0.91, P = 0.36 × 10 −4 ) and CD28 + CD4 − CD8 − T cells (OR = 0.92, P = 0.90 × 10 −4 ). Notably, no evidence supported a causal effect of wAMD on immune cell phenotypes. Using updated GWAS data, this study confirms subtype-specific immune associations in AMD and reports the first evidence of reverse causality: dAMD may directly modulates adaptive immune cell phenotypes. These findings refine our understanding of AMD immunopathogenesis and highlight potential targets for subtype-specific therapies. • Systematic bidirectional MR screens 731 immune phenotypes across wet and dry AMD subtypes. • HLA-DR+ monocytes are identified as risk factors, while CD25+ T-cells exhibit protective roles in wAMD. • Novel reverse MR findings reveal dAMD-driven immune remodeling, offering new etiological insights. • Results provide a genetic rationale for targeted immunomodulation therapies in AMD.

  PublisherDOI

• Novel Optical Materials and Applications (NOMA): feature issue introduction

  Optical Materials Express · 2025-09-23

  articleOpen access

  We introduce the Optical Materials Express feature issue on Novel Optical Materials and Applications (NOMA). This issue comprises a collection of fifteen papers that highlight recent progress in the design, fabrication, characterization, and theoretical analysis of advanced optical materials and photonic structures. This feature issue is built upon the 2024 Novel Optical Materials and Applications (NOMA) Conference, part of the Optica Advanced Photonics Congress that was held in Québec City, Québec, Canada in 2024.

  PublisherDOI

• Electrically-driven nonreciprocal silicon photonic isolator based on monolithically integrated bismuth terbium iron garnet

  2025-05-26

  preprintSenior author

  Photonic isolation is an essential on-chip function needed to stabilize laser operation and prevent cross-talk between different components within a communication system. Monolithically integrated isolator designs incorporate magnetooptical cladding consisting of doped yttrium iron garnet (YIG), such as Ce:YIG and Bi:YIG, where a YIG seedlayer is required prior to the deposition and subsequent crystallization of doped YIG. This adversely impacts isolator performance through reduced mode overlap with the doped YIG layer. Furthermore, the Faraday rotations of YIG and Ce-doped YIG have opposing signs such that they partially cancel out. This work demonstrates a seedlayer-free, monolithic nonreciprocal photonic platform based on an alternative garnet formulation, bismuth doped terbium iron garnet (Bi:TbIG). We further realize on-chip integration of electromagnets to obviate the requirements of external magnets. An isolator based on the Mach-Zehnder interferometer (MZI) architecture is implemented, which records π nonreciprocal phase shift with a driving current of 114 mA. Optical isolation as high as 24 dB is achieved.

  PublisherDOI

• Large-scale self-assembled nanophotonic scintillators for X-ray imaging

  Nature Communications · 2025-07-01 · 8 citations

  articleOpen access

  Scintillators convert X-ray energy into visible light and are critical for imaging technologies. Their widespread use relies on scalable, high-quality manufacturing methods. Nanophotonic scintillators, featuring wavelength-scale nanostructures, can offer improved emission properties such as higher light yield, shorter decay times, and enhanced directionality. However, achieving scalable fabrication of these structures remains challenging. Here, we present a scalable fabrication method for large-area nanophotonic scintillators based on the self-assembly of chalcogenide glass photonic crystals. This technique enables the production of nanophotonic scintillators over wafer-scale areas, achieving a six-fold enhancement in light yield compared to unpatterned scintillators. By studying surface nanofabrication disorder, we show its impact on imaging performance and provide a route towards scintillation enhancements without compromising resolution. We demonstrate the practical applicability of our nanophotonic scintillators through X-ray imaging of biological and inorganic specimens. Our results could enable the industrial implementation of a new generation of nanophotonic-enhanced scintillators. Scintillators are used for converting X-ray energy into visible light in imaging technologies. Here, the authors present a scalable fabrication approach for large-area nanostructured scintillators, and achieve six-fold enhancement in light yield compared to unpatterned scintillators.

  PublisherOA PDFDOI

• Lossless resistive micro-heater design for reconfigurable phase-change photonics

  2025-05-07

  preprintOpen accessSenior author

  Electrically reconfigurable photonics based on optical phase change materials (PCMs) have attracted surging interest recently with broad potential applications ranging from analog computing to optical camouflage. PCM switching in these devices is customarily implemented using resistive micro-heaters. These electrically conductive micro-heaters, however, cause significant optical losses due to free carrier absorption. In this paper, we present a design concept that overcomes this limitation without compromising the electrical performance of the micro-heaters. Using doped Si heaters as an example, we show that such parasitic losses can be suppressed by engineering the optical mode to minimize the optical field overlap with the lossy doped region. An active optical meta-grating design was proposed based on the concept, which achieves 2π phase tuning range near 1.53 μm wavelength with only 10 nm thick Sb₂S₃ while maintaining over 85% optical efficiency

  PublisherDOI

• Microheater hotspot engineering for spatially resolved and repeatable multi-level switching in foundry-processed phase change silicon photonics

  Nature Communications · 2025-05-09 · 10 citations

  articleOpen access

  Nonvolatile photonic integrated circuits employing phase change materials have relied either on optical switching with precise multi-level control but poor scalability or electrical switching with seamless integration and scalability but mostly limited to a binary response. The main limitation of the latter is relying on stochastic nucleation, since its random nature hinders the repeatability of multi-level states. Here, we show engineered waveguide-integrated microheaters to achieve precise spatial control of the temperature profile (i.e., hotspot) and, thus, switch deterministic areas of an embedded phase change material. We experimentally demonstrate this concept using a variety of foundry-processed doped-silicon microheaters on a silicon-on-insulator platform featuring Sb2Se3 or Ge2Sb2Se4Te and achieve 27 cycles with 7 repeatable levels each. We further characterize the microheaters’ response using Transient Thermoreflectance Imaging. Our microstructure engineering concept demonstrates the evasive repeatable multi-levels employing a single microheater device, which is necessary for robust and energy-efficient reprogrammable phase change photonics in analog processing and computing. Stochastic nucleation prevents the repeatable multi-level response of phase change materials in integrated photonics. Here the authors circumvent this issue with a method using deterministic amorphization via spatially controlled microheater hotspots.

  PublisherOA PDFDOI

• PIC-based spectroscopic chemical sensing

  2025-03-21

  articleSenior author

  Photothermal sensing is an indirect absorption sensing method that measures the change in temperature caused by the analyte molecules absorbing at a particular wavelength. While this technique has been demonstrated on various photonic platforms, silicon nitride circuits are specifically known for their ultralow propagation losses and CMOS compatibility. Here, we demonstrate suspended silicon nitride microdisk resonators that can be employed for near-infrared photothermal sensing. The photonic circuits were made in a photonic foundry multi-project wafer run with simple post-processing to remove the oxide underneath the microdisk. We observed optical bistability in our suspended microdisks, making them suitable for the required application.

  PublisherDOI

• Modeling a metalens-based system for GHz fiber mode-locked lasers

  Optics and Lasers in Engineering · 2025-08-22 · 1 citations

  articleOpen accessSenior author

  • We propose a lens system design to enhance polarization control, optimize saturable absorber efficiency, and achieve a high free spectral range (FSR) in a mode-locked laser (MLL). • We present the design, simulation, and experimental characterization of a polarization-dependent metalens for spatial filtering applications. • We develop and characterize a two-photon polymerized (TPP) fiber-tip printed lens collimator to improve alignment and offset tolerance. • We model the complete lensing system in ZEMAX to assess the impact of misalignment and positional offsets on optical performance. Fiber-based passive mode-locked lasers (MLLs) are a well-established technology for high-speed optical communications, capable of generating ultrashort pulses with high energy. While most commercial MLLs operate at repetition rates around 100 MHz, increasing this frequency to the GHz range introduces significant challenges, including polarization control, efficient saturation of the saturable absorber, heat dissipation and the achievement of a high free spectral range (FSR). To address these limitations, we propose a system consisting of a metalens and a 3D-printed fiber-tip collimator. The metalens is designed to selectively focus one polarization while diverging the orthogonal component, thereby addressing the polarization control. To enhance its performance and increase tolerance to positional offsets and angular tilts, we fabricated a fiber-tip collimator using two-photon polymerization (TPP). Our model suggests that this integrated system could enable the miniaturization of fiber-based MLLs while controlling polarization, enhancing the efficiency of the saturable absorber through better heat dissipation, and increasing the FSR with a shorter fiber length.

  PublisherDOI

Recent grants

• Collaborative Research: Combinatorial solution processing of optical phase change materials

  NSF · $320k · 2022–2026

• Collaborative Research: Conformal and robust integrated infrared spectroscopic sensors

  NSF · $250k · 2017–2020

• Collaborative Research: FuSe: High-throughput Discovery of Phase Change Materials for Co-designed Electronic and Optical Computational Devices (PHACEO)

  NSF · $315k · 2023–2026

• ASCENT: PROWESS: Phase-change Reconfigurable Optical WavEfront Synthesis System

  NSF · $1.5M · 2021–2026

• CAREER: Glass-Based Fexible Integrated Photonic Devices

  NSF · $500k · 2015–2020

Frequent coauthors

• Anu Agarwal

  203 shared

• Kathleen Richardson

  190 shared

• Tian Gu

  Massachusetts Institute of Technology

  186 shared

• Lionel C. Kimerling

  Massachusetts Institute of Technology

  180 shared

• Hongtao Lin

  State Key Laboratory of Modern Optical Instruments

  127 shared

• Qingyang Du

  Zhejiang Lab

  92 shared

• Vivek Singh

  L V Prasad Eye Institute

  85 shared

• Mikhail Y. Shalaginov

  Massachusetts Institute of Technology

  81 shared

Labs

• Photonic Materials GroupPI

Education

• Ph.D., Materials Science and Engineering

  Massachusetts Institute of Technology

  1999

• M.S., Materials Science and Engineering

  University of California, Berkeley

  1994

• B.S., Materials Science and Engineering

  University of Science and Technology of China

  1991

Awards & honors

• Vittorio Gottardi Prize, International Commission on Glass (…
• SPIE Early Career Achievement Award (2019)
• Robert L. Coble Award, American Ceramic Society (2017)
• Faculty Early Career Development Award, National Science Fou…