Quantum random number generators play a crucial role in securing high-demand information contexts by producing true random numbers. Nevertheless, the large volume and high cost limit their widespread use. Here, a system on-chip is proposed that fully leverages the advantages of different photonic integrated platforms, where the interference optical paths and photodiodes are integrated on a standard silicon process, while the laser source on-chip is realized on a III–V platform. Using micro-lens coupling package technology with coupling loss lower than 2 dB, the components on different platforms are combined and packaged with the amplifier circuits in a 42 mm × 24 mm footprint in a butterfly form. This miniaturized and cost-effective entropy source enables outputting a vacuum noise signal with a 3-dB bandwidth of over 500 MHz. After sampling and post-processing, a random number generation rate of up to 6.57 Gbps is achieved.

With the increasingly urgent demand for non-toxic, high-specific-impulse propellants in space propulsion systems, ammonium dinitramide (ADN)-based liquid propellants have attracted extensive attention as a new-generation green alternative to hydrazine-based fuels. However, existing ADN-based thrusters all adopt the catalytic ignition technology route, which suffers from inherent drawbacks: the catalyst cannot withstand temperatures exceeding 1500 K, long preheating times are required prior to startup (in the Prisma mission, in-flight preheating lasted 600–720 s with a single-event energy consumption of 25 kJ), and insufficient preheating may lead to "hard start" or even explosion—these issues severely constrain the rapid response capability and operational safety of the thrusters. Although active ignition methods such as resistive ignition and laser ignition have been validated at the single-droplet level, how to achieve catalyst-free electric ignition at the thruster system level and systematically evaluate its combustion characteristics remains a critical engineering problem urgently to be addressed in the field of green space propulsion.

Plasma bubbles in the ionosphere over equatorial and low-latitude regions are a common nighttime plasma density depletion structure that can cause severe scintillation interference on high-frequency communications, making them an important target of space weather monitoring. Traditional detection methods rely on statistical thresholds derived from temporal fluctuations in ion density; however, plasma bubbles exhibit significant variability in spatiotemporal scales (ranging from kilometers to hundreds of kilometers), and their occurrence frequency is strongly modulated by solar activity and seasonal factors. Consequently, fixed-threshold methods struggle to accommodate this diversity, with miss rates reaching up to 40% during periods of low solar activity. Although deep learning has demonstrated powerful capabilities in automatically extracting and recognizing complex waveform features in fields such as medical image segmentation and signal peak detection, its systematic application to ionospheric plasma bubble detection—and thus addressing the adaptability deficiencies of traditional threshold methods—remains underexplored.

With the long-term on-orbit operation of space stations, extravehicular status monitoring and surface inspection tasks have imposed higher demands on the motion planning capabilities of manipulators. Space redundant manipulators, when performing visual monitoring tasks, must simultaneously satisfy multiple complex constraints including kinematics, dynamics, obstacle avoidance, and end-effector line-of-sight, making this a typical nonlinear non-convex optimal control problem. Although sampling-based methods can rapidly generate collision-free paths, they struggle to guarantee dynamic feasibility in high-dimensional systems and often yield poor-quality paths. Optimization-based methods, while capable of directly generating high-quality trajectories, are highly sensitive to initial solutions and lack effective warm-start mechanisms. Therefore, how to merge the rapid search capability of sampling-based methods with the precise solving advantage of optimization-based methods to plan high-quality executable trajectories under complex constraints for space manipulators has become a key challenge in advancing the intelligent development of space robotics.

With the rapid development of 6G space–air–ground integrated communication networks, low Earth orbit (LEO) satellite constellations, leveraging their advantages of wide-area coverage and freedom from geographic constraints, have become key infrastructure for achieving global seamless connectivity. However, due to the relative motion among satellites in different orbits, inter-satellite links (ISLs) exhibit periodic dynamic on–off characteristics, among which a portion of the interruptions are predictable topological changes. Conventional Open Shortest Path First (OSPF) protocols compute shortest paths solely based on link distance, without considering link effective duration or bandwidth constraints, rendering them inadequate for adapting to the dynamic topological characteristics of satellite networks. Although fast reroute strategies can address sudden link failures, they consume substantial bandwidth resources and cannot provide protection for all nodes across the entire network. Therefore, how to proactively switch service paths before predictable link interruptions occur, so as to minimize service disruption and resource overhead, has become a core issue in ensuring service continuity in LEO satellite networks.