Cloud computing platform provides on-demand IT services to users and advanced the technology. The purpose of virtualization is to improve the utilization of resources and reduce power consumption. Energy consumption is a major issue faced by data centers management. Virtual machine placement is an effective technique used for this purpose. Different algorithms have been proposed for virtual machine placement in cloud environments. These algorithms have considered different parameters. It is obvious that improving one parameter affects other parameters. There is still a need to reduce energy consumption in cloud data centers. Data centers need solutions that reduce energy consumption without affecting other parameters. There is a need to device solutions to effectively utilize cloud resources and reduce energy consumption. In this article, we present an algorithm for Virtual Machines (VMs) placement in cloud computing. The algorithm uses adaptive thresholding to identify over utilized and underutilized hosts to reduce energy consumption and Service Level Agreement (SLA) violations. The algorithm is validated with simulations and comparative results are presented.
Published projections suggest that information and communication technologies could account for up to 20% of global electricity use by 2030, yet these estimates are often based on short (<5 years) historical periods. Here, we present the first global long-run (1975-2022) analysis that jointly estimates electricity consumption, processed information, and efficiency. We find that these increased by 4, 11, and 7 orders of magnitude, respectively. However, after an initial exponential growth, the share of computing devices in world electricity consumption peaked at 2.5% in 2013, then decreased and stabilized at 1.8% since 2018. The stabilisation was due to the massive increases in information processing being offset by efficiency gains associated with the growing amount of computation in large datacenters and the shift from desktop computers to laptops and, more recently, to smartphones. These results indicate that concerns about the future electricity demand of computing may be overstated.
We present a graded-κ grating technique that enables stable single-mode lasing and robust modulation-bandwidth enhancement in directly modulated lasers. Internal reflection at the interface between the active and passive regions typically degrades detuned-loading and photon-photon resonance effects, which limit high-speed operation. The proposed graded-κ structure suppresses these reflections by smoothing the photon-density and refractive-index transition, resulting in fluctuation-free bandwidth scaling and reliable activation of single-cavity etalon effects. The device achieves a 3-dB bandwidth of 47.5 GHz and supports 112-Gbps PAM4 transmission with TDECQ values lower than 2.9 dB up to 45 ℃, demonstrating its suitability for next-generation optical interconnect solutions in datacenters.
The randomized, multicenter, prospective Phase 3 trial (NCT02777736) evaluated central nervous system (CNS) prophylaxis using either intravenous (i.v.) or intrathecal (i.t.) methotrexate (MTX) in diffuse large B-cell lymphoma (DLBCL). Treatment consisted of six cycles of R-CHOP + 2xR or DA-EPOCH-R + 2xR. Patients with intermediate or high-risk CNS International Prognostic Index (CNS-IPI) were randomized to receive CNS prophylaxis with either 2 doses of MTX 3 g/m2 i.v. (arm A) or 6 doses of MTX 12 mg i.t. (arm B). Patients with low-risk CNS-IPI did not receive MTX prophylaxis (arm C). The primary objective was to compare the cumulative incidence of CNS relapse between arms A and B. Secondary objectives included evaluation of overall response rate (ORR), complete remission rate (CRR), progression-free survival (PFS), overall survival (OS), and treatment-related safety across all arms. Between 7/2015 and 5/2024, a total of 100 patients were enrolled: 30 in arm A, 31 in arm B, and 39 in arm C. ORR did not differ among arms (p = 0.20). During a median follow-up of 54.9 months, CNS relapses were observed in three patients who had received MTX prophylaxis-one in arm A and two in arm B. The 5-year cumulative incidence of CNS relapse was 0% in arm A and 8.7% in arm B (p = 0.72). However, due to the small sample size, the primary endpoint results are inconclusive. Median PFS was comparable between arms A and B (HR 0.66, p = 0.20). MTX i.v. was associated with a significantly higher grade ≥ 3 neutropenia (p = 0.0003) and infection (p = 0.0063). The higher infection rate contributed to a worse 5-year OS in arm A versus B (47.2% vs. 72.4%, HR 0.46, p = 0.04). Conclusion: our trial faced limitations due to a low number of randomized participants, making the interpretation of results challenging. A larger, international randomized trial is necessary to determine the benefit of CNS prophylaxis.
We propose and experimentally validate a novel, to the best of our knowledge, LR-8 optical demultiplexer architecture based on a four-stage multiple delayed interferometer employing discrete path-length differences and multimode-interference couplers to achieve spectral flatness and suppress crosstalk. An intentional half-length offset path difference in the third stage breaks the inherent spectral periodicity in the wavelength domain, while interferometer-type band-rejection filters in the fourth-stage further suppress out-of-band noise without any penalty for the main signals, thereby improving the crosstalk extinction ratio (XER). Fabricated silicon-nanowire devices exhibit flat-topped, non-periodic LR-8 passbands under TE-polarized input, with a measured XER of >11dB in good agreement with analytical predictions. We further experimentally investigate the impact of statistical random phase errors, arising from fabrication imperfections and/or intentional waveguide design variations, on the spectral degradation of the LR-8 device. With its compact footprint, flat spectral response, and high integrability, the proposed device shows strong potential for next-generation 800-GbE-class data-center interconnects.
Desirable data-center features include cost efficiency, high bandwidth, scalability, and protection against component failures. In this study, we propose an architecture that employs a small number of optical sources to transport data over a large number of links inside a data center. To further increase the bandwidth, two bits per symbol are transmitted by employing a novel technique for the generation and demodulation of pulse amplitude modulated (PAM) signals having a data rate of 100 Gbps. The proposed architecture provides protection against two major sources of breakdown, the laser source and the channel used for the transmission of signals. Furthermore, diversity gain can also be achieved by simultaneously transmitting the same optical signals through different channels inside the data center. The performance of the proposed architecture is observed in terms of bit error rate under different amplified spontaneous emission (ASE) noise powers.
The rapid advancement of Artificial Intelligence (AI) is driving unprecedented computational demands, posing significant challenges to datacenter infrastructure and threatening the stability and resilience of modern power grids. This study presents an open-access dataset featuring a diverse set of AI training sessions recorded at sub-second resolution, designed to advance research on the energy consumption profiles of AI workloads and their interactions with power grid dynamics in datacenter environments. The dataset contains 32 training sessions on high-performance H100 and B200 8-GPU nodes and 40 sessions on consumer-grade NVIDIA GeForce RTX 3060 GPUs, encompassing over 1.8 million samples. Each session records power demand, CPU and GPU utilization, per-GPU power, memory usage, and temperature across diverse AI tasks (at the node scale, temperature refers to the GPUs temperature), including forecasting, classification, reinforcement learning, and text and image generation. Data quality was verified through detailed technical validation, including timing accuracy, hardware limit conformance, and cross-metric correlation analysis. Measurements remained within manufacturer-specified thermal and power envelopes, and observed correlations among power, utilization, temperature, and current were consistent with established processor and GPU behavior. The dataset provides a robust foundation for modeling AI datacenter energy behavior, system-level performance analysis, and power grid connection impact assessment studies.
We report on a 3-dB bandwidth extension technique achieved through a coplanar waveguide (CPW) electrode design that imposes inductive gains in a backside-lens-integrated InGaAs positive-intrinsic-negative (PIN) photodiode. The inductive CPW electrodes are analyzed in detail using an equivalent circuit model in which the transit time of the photo-generated carriers has been included. The transit times for the electrons and holes have been analyzed numerically. By including the transit time effect in the equivalent circuit model, the extracted RLC parameters show a good consistency with measured frequency response data. As a result, the 3-dB bandwidth has been extended from 48 GHz to over 67 GHz at a -2.5 V bias voltage simply by reconfiguring the CPW electrode design.
We demonstrate a polarization-independent silicon photonics optical switch based on mode conversion and interference of multimode interferometers (MMIs). The proposed architecture consists of a mode converter for TE and TM separation and TM to TE conversion, a cascaded MMIs for forming inner and outer Mach-Zehnder Interferometer, phase shifters using thermal heater and pn-junction, and a 4×2 asymmetric MMI for combining TE and TM (converted to TE). This architecture converts the input TM mode into a first-order TE mode and splits it into two fundamental TE modes. It employs asymmetric MMI structures to route both TE and TM inputs to the same output ports via inner and outer interferometers, respectively. This approach significantly reduces device complexity while maintaining polarization-independent operation. The device design is presented together with an analysis of the wavelength dependence of insertion loss and polarization-dependent loss (PDL). A practical calibration method is developed to compensate for the phase difference of TE and TM modes using a limited number of electrical control parameters. Experimental results show stable continuous wave switching with low crosstalk, PDL below 3.4 dB over the C band, and robust operation against wavelength and temperature variations. Bit-error-rate measurements at 25 Gb/s with OOK modulation format confirm a power penalty below 1 dB for all tested polarization states and output ports. The wavelength dependence observed for mixed TE/TM input states is analyzed based on group-delay imbalance between interfering optical paths, showing good agreement between calculated and measured free spectral ranges. These results indicate that the proposed polarization-independent optical switch is a promising building block for low-latency, high-throughput optical interconnects in future data-center and computing systems.
Shop-floor weld inspection may appear to be a solved problem until a camera is deployed near a galvanized-sheet MIG welding line. The seam reflects light, the texture changes from frame to frame, and the defects of interest are often small and visually subtle. Additionally, the hardware near the line is rarely a data-center GPU. With those constraints in mind, this paper presents YOLO-MIG, a compact detector built on YOLOv10n for weld-seam inspection in practical production conditions. We make three focused changes to the baseline: a C2f-EMSCP backbone block to better preserve weak defect cues with modest parameter growth, a BiFPN neck to keep small-target information alive during feature fusion, and a C2fCIB head to clean up predictions that otherwise get distracted by seam edges and illumination artifacts. On a workshop-collected dataset containing 326 original images, with the training subset expanded through augmentation to 2608 labeled samples in total, YOLO-MIG achieves 98.4% mAP@0.5 and 56.29% mAP@0.5:0.95 on the test set while remaining lightweight (1.83 M parameters, 3.87 MB FP16 weights). Compared with YOLOv10n, the proposed model improves mAP@0.5 by 9.36 points and mAP@0.5:0.95 by 4.89 points, while reducing parameters, GFLOPs, and model size by 43.4%, 19.9%, and 29.9%, respectively. The results suggest that YOLO-MIG is not only accurate but also realistic to deploy at the edge for intelligent weld quality control.
It has become increasingly important to accurately identify cancer patients at high risk for venous thromboembolism (VTE) who could have greater benefit with anticoagulants. Cancer-associated genomic variants could have a potential clinical utility for prediction of VTE. This study was a single-center observational study using the C-CAT database, the national datacenter for cancer genomic medicine in Japan, and evaluated 412 cancer patients who underwent comprehensive genome profiling with FoundationOne CDx at Kyoto University Hospital. We comprehensively investigated the association between cancer-associated genomic variants and VTE development. In the entire cohort, 77% had distant metastasis, and 90% were under chemotherapy. During a median follow-up period of 693 days, 59 patients (14.3%) developed VTE events. The cumulative incidence of VTE events after specimen collection was 26.0% at 5 years. In the multivariable Fine-Gray sub-distribution hazard models adjusted for age, sex, cancer type, metastasis, and chemotherapy at baseline, several genomic variants showed a trend toward an increased risk of VTE, including ERBB2 (HR 2.43, 95% CI 1.21-4.87), APC (HR 1.94, 95% CI 0.84-4.50), KRAS (HR 1.67, 95% CI 0.79-3.53), ATM (HR 1.64, 95% CI 0.74-3.65), and NOTCH1 (HR 1.51, 95% CI 0.73-3.11). However, no variants remained statistically significant after correction for multiple testing. The current exploratory study identified several genomic variants potentially associated with a high risk of cancer-associated VTE, which may indicate potential clinical utility of comprehensive genomic profiling for accurate prediction of VTE events in cancer patients.
Artificial intelligence (AI) promises to redefine production systems, decision-making processes, and even social relationships, but it raises critical (and ethical) questions about its environmental impact. The paradox is clear: AI can be both part of the solution and part of the problem. On one hand, it enables better forecasting of extreme weather events thanks to data collected from satellites and sensors, supports "smart" electrical grids that integrate renewable sources such as solar and wind power, strengthens "precision agriculture" by optimizing irrigation and fertilization, and makes smart cities more sustainable through intelligent mobility systems and energy optimization. On the other hand, it introduces unavoidable environmental costs linked to high energy and water consumption, the limited availability of raw materials needed to build data-center components, and the challenge of disposing of them sustainably. Experts propose shifting from energy-hungry Red AI to Green AI, based on lighter, less complex models powered by renewable energy. Techniques such as federated learning and pruning, combined with the use of sustainable data centers, recyclable hardware, and distributed architectures, make it possible to drastically reduce consumption without sacrificing performance. The challenge for the future will be to govern AI with a vision that balances innovation and sustainability, environmental justice and technological progress.
Enhancing the efficiency and beam directivity of GaN micron-scale light-emitting diodes (µLEDs) is critical for visible-light communication, which has emerged as a promising platform for high-bandwidth optical links in data-center environments. We demonstrate a µLED design where the emitting mesa is laterally enclosed by a distributed Bragg reflector (DBR). This design achieves ∼20% higher optical output through air-side emission and ∼130% higher optical output through substrate-side emission with ∼30% reduced divergence compared to reference devices enclosed by a TiO2 film. Our results present a manufacturable route to efficient, directional µLEDs with applications in optical interconnects and advanced display technologies.
Transmission of high-speed optical data over fiber is inherently limited by optical attenuation and chromatic dispersion, with dispersion impairments becoming increasingly critical as data rates scale and symbol durations shorten. One of the practical solutions is to utilize an integrated photonic device that produces dispersion with the same magnitude but opposite sign, so that net zero aggregate dispersion is achieved in the link. In this work, we demonstrate dispersion compensation of 30 Gb/s NRZ data using an integrated silicon nitride (SiN) grating-based device. The SiN Bragg grating operates in transmission and generates dispersion to counteract the anomalous dispersion intrinsic to the optical fiber using the slow light effect. The dual-period sidewall design is engineered to provide tailored differential group delay characteristics across channels 53, 55, 57, and 59 on the ITU coarse wavelength division multiplexing grid using both transverse electric (TE) and transverse magnetic (TM) modes, enabling effective compensation of accumulated fiber dispersion while maintaining low insertion loss. Experiments show clear improvements in the eye diagrams and reductions in the measured bit error rates after dispersion compensation, confirming the effectiveness of the proposed SiN grating for high-speed 1 by 4 wavelength division multiplexed links. Owing to its CMOS-compatible fabrication, low propagation loss, and sufficiently high dispersion magnitudes for compensation of long fiber propagation lengths, the demonstrated approach offers a compact and scalable solution for dispersion management in short-reach and data-center optical communication systems.
Computational theories of memory posit that the dentate gyrus and CA3 (CA3DG) hippocampal subfields reduce mnemonic interference via a process called pattern separation. While the CA3DG is viewed as a domain-general pattern separator, the parahippocampal and perirhinal cortices may play a role in content-specific (e.g., spatial or object-related) interference reduction. Recent work suggests that frontal and parietal control areas may allocate resources during mnemonic discrimination, but the evidence is still limited. Moreover, mnemonic discrimination tasks designed for humans almost exclusively use everyday items as stimuli, confounding retrieval processes with pattern separation. To address these challenges, we acquired high-resolution structural images of the medial temporal lobe, and full-brain high-resolution functional MRI data of 39 participants while they studied non-meaningful fractals with varying degrees of interference in either their spatial or object features. We found that the parahippocampal cortex contributes to interference reduction in the spatial domain, while the perirhinal cortex contributes to interference reduction in the object domain. The dorsolateral frontal and parietal regions were recruited during the encoding of interfering stimuli in both object and location domains and displayed strengthened within- and cross-network connectivity in response to interference. Contrary to our expectations, we did not find significantly increased activation in the CA3DG to similar trials relative to repeats, indicating a lack of sensitivity to small differences in interfering stimuli. Altogether, these results are in line with content-specific interference reduction in the medial temporal lobe, possibly orchestrated by frontoparietal regions, but challenge the view of the CA3DG as a universal pattern separator of the human brain.
Visible-light communication is increasingly being regarded as a pivotal complementary paradigm for emerging wireless infrastructures, while optical interconnects are widely acknowledged as a disruptive enabler of energy-efficient, ultra-dense data-center architectures. Based on this idea, we proposed an InGaN/GaN LED array for high-speed, short-distance optical links. By utilizing InGaN/GaN multi-quantum-well (MQW) devices and matrix electrode injection structures, the LED array (10 × 10) was fabricated for multi-channel communication using an on-off keying modulation scheme. The electroluminescence (EL), modulation bandwidth, and signal transmission properties of individual LEDs were thoroughly characterized. Experimental results demonstrated that the individual unit was about 200 × 200 μm in size, and it showed a green emission peak near 530 nm, achieving a -3 dB bandwidth in the region of 5.3 MHz to 13.7 MHz with currents in the region of 10 to 70 mA. The data rate remained stable at 50 Mbps with a bit error rate below 2 × 10-5 and can reach a limiting data rate exceeding 100 Mbps for drive currents over 80 mA. We also confirmed that these values are high enough to support video and audio transmissions. Furthermore, as the LED array was designed in a row-common-ground configuration with individually addressable columns, the total data rate in a multiple-input multiple-output (MIMO) mode theoretically exceeded 500 Mbps. Our study reveals the promising potential of compact LED arrays for integrated optical links, with significant room remaining for improvement, particularly in reducing power consumption.
As the exponential growth in advanced compute workloads drives intra-datacenter interconnects to ever increasing bitrates, optical networking equipment has risen to the challenge by shifting from NRZ signaling to bandwidth efficient modulation methods such as PAM4. As these modulation schemes introduce an inherent SNR penalty, maintaining low bit error rates (BER) forces optical links to operate at significantly higher optical powers. However, increasing the optical power leads to photodetectors reaching one of their fundamental bottlenecks caused by the space-charge effect, limiting their ability to provide a high-speed response under high-power illumination. This work presents the design, fabrication, and characterization of a waveguide-integrated photodiode with dual optical inputs (DIPD) designed to overcome this limitation. Specifically, we demonstrate that combining a dual-fed architecture with targeted cross-sectional geometric optimizations effectively distributes the photocurrent density to delay the onset of space-charge saturation. Experimental validation demonstrates a high responsivity of ≈0.91 [A/W] (for O-band wavelengths) and a large electro-optic bandwidth (EOBW) of ≈58 [GHz], all under high-power illumination and CMOS driving voltages.
Since the inception of integrated photonics, multimaterial integration has served as a primary avenue for new technology innovations. Now, with an ever-increasing demand for integrated photonics as a platform for both high-performance links from/within datacenters and AI acceleration, multimaterial integration has begun to play an even more critical role in pushing capabilities beyond their current limits. In this work, we review photonics for AI and datacenter applications, the current landscape of multimaterial integration in photonics, and the ways in which multimaterial integration techniques have been recently utilized to push the performance of modulators on silicon and chip-scale optical frequency combs.
Multimodal artificial intelligence (AI) has the potential to revolutionise healthcare by enabling the simultaneous processing and integration of various data types, including medical imaging, electronic health records, genomic information and real-time data. This review explores the current applications and future potential of multimodal AI across healthcare, with a particular focus on orthopaedic surgery. In presurgical planning, multimodal AI has demonstrated significant improvements in diagnostic accuracy and risk prediction, with studies reporting an Area under the receiving operator curve presenting good to excellent performance across various orthopaedic conditions. Intraoperative applications leverage advanced imaging and tracking technologies to enhance surgical precision, while postoperative care has been advanced through continuous patient monitoring and early detection of complications. Despite these advances, significant challenges remain in data integration, standardisation, and privacy protection. Technical solutions such as federated learning (allowing decentralisation of models) and edge computing (allowing data analysis to happen on site or closer to site instead of multipurpose datacenters) are being developed to address these concerns while maintaining compliance with regulatory frameworks. As this field continues to evolve, the integration of multimodal AI promises to advance personalised medicine, improve patient outcomes, and transform healthcare delivery through more comprehensive and nuanced analysis of patient data. Level of Evidence: Level V.
In-phase and quadrature (IQ)-interleaved 2-channel optical time-division multiplexing (OTDM) is a promising approach to overcoming transmitter bandwidth limitations in high-speed intensity-modulation and direct-detection systems. However, we show analytically that IQ-interleaved OTDM signals generated with conventional constant-phase pulses suffer from a severe chromatic dispersion (CD)-induced power imbalance between two multiplexed channels, which limits the 200-Gb/s pulse-amplitude modulation 4-level (PAM4) transmission reach to less than 1 km in the C-band. To address this fundamental limitation, we propose to use sinusoidally modulated alternating-phase pulses (with a 50-GHz repetition rate) generated by a null-biased Mach-Zehnder modulator driven by a 25-GHz electrical clock. We demonstrate analytically and through simulation that these pulses are inherently immune to CD-induced power imbalance regardless of the sign of the relative phase between the two return-to-zero (RZ) PAM4 tributaries during IQ combining (ϕRZ2 - ϕRZ1 = ±90°), and further demonstrate through simulation that the +90° configuration is preferable to the -90° counterpart, since it extends the transmission reach to 6 km. To fully exploit this advantage, we further propose a two-stage equalization scheme in which a half-symbol-spaced symbol-timing recovery block suppresses aliasing before an adaptive channel equalizer that compensates residual inter-symbol interference. The proposed system is experimentally implemented using a silicon photonics transmitter chip. Transmission of 200-Gb/s C-band PAM4 signals is experimentally demonstrated over 1.9 km, validating the feasibility of C-band 2-km data-center networks, while simulation results further show transmission reach of up to 6 km, suggesting that sufficient margin can be secured for future 400-Gb/s/λ operation.