▌University of California Research Suggests Wafer-Level Accelerators Could Reshape AI Computing Efficiency and Sustainability
The engineering team at the University of California, Riverside, published a review in the journal Device, indicating that wafer-level accelerator chips developed by companies like Cerebras, with their integrated design and high-density integration, can provide AI models with computational power and energy efficiency far exceeding traditional GPUs. By integrating millions of AI cores within a single wafer, they significantly reduce data transmission energy consumption, helping to alleviate environmental pressures from computational power and cooling. The research suggests that wafer-level chips are suitable for ultra-large-scale AI tasks and will become an indispensable new type of hardware for high-end AI training, promoting sustainable computing and AI development.
▌OpenAI Secures $200 Million AI Contract with the U.S. Department of Defense
OpenAI has signed a $200 million contract with the U.S. Department of Defense to develop AI tools for the Pentagon, supporting critical areas such as management, data collection, and proactive network defense. This is OpenAI’s first government-customized AI project, with the contract running until 2026, involving the development of prototype AI capabilities and optimization of administrative processes. OpenAI stated that all use cases must comply with its policies, prohibiting weapon development and harmful actions. This collaboration marks an acceleration of AI applications in the defense sector, as many tech companies in the U.S. have recently relaxed AI military restrictions to meet national security needs.
▌U.S. Develops New Nanofiber Technology to Enhance Carbon Fiber Composite Strength
The Oak Ridge National Laboratory (ORNL) in the U.S. has developed an innovative method that combines carbon nanofibers to significantly enhance the strength and toughness of carbon fiber composites. This technology utilizes electrospinning to achieve both chemical and mechanical bonding of nanofibers with carbon fibers and polymer matrices, resulting in a 50% increase in tensile strength and nearly double the toughness. This achievement is expected to lower material costs and broaden the application prospects of carbon fiber in automotive, aerospace, and energy sectors, and a patent has been applied for, seeking industrial collaboration for promotion.
▌DARPA Achieves Breakthrough in Wireless Laser Power Transmission at 10 kW Level
DARPA has achieved a milestone in its “Persistent Optical Wireless Energy Relay” project by successfully transmitting 10 kW of energy wirelessly over nearly 1 kilometer using lasers, setting records for both distance and power in ground-based wireless energy delivery. The core receiving device, PRAD, can flexibly convert light energy into electricity. This achievement lays the foundation for applications in long-distance wireless power supply, emergency rescue, and ground delivery of solar energy from space, marking a new phase in the practical application of wireless energy technology.
▌DARPA Launches PROSE Project to Promote High-Throughput Protein Sequencing Innovations
DARPA has launched the “PROSE” protein sequencing project, aiming to develop a high-accuracy, scalable molecular reader within three years to achieve real-time high-throughput sequencing of at least 100 billion amino acids per day, directly identifying unknown and modified proteins. This project aims to enhance biological threat monitoring and defense capabilities, promote synthetic biological safety, disease warning, and technological commercialization, strengthening U.S. public health and innovation competitiveness.
▌UK Company Develops “ChatGPT for Biology” to Challenge Gene Data Application Bottlenecks
Basecamp Research, a UK company, has discovered over a million new species and nearly ten billion new genes by collecting extremophilic microorganisms, intending to use this database to train a “ChatGPT for biology.” Despite the wealth of data, experts point out that the lack of functional and physiological information limits AI models’ understanding and application of new genes. The industry believes that merely “brute-forcing” data expansion will not significantly enhance AI’s predictions of unknown biological functions, and traditional experimental validation is still necessary. Nevertheless, this project provides valuable resources and new ideas for biodiversity and generative biological AI exploration.
▌U.S. Department of Energy Opens MARVEL Micro Nuclear Reactor Testbed
The U.S. Department of Energy (DOE) has announced that the MARVEL micro nuclear reactor testbed at the Idaho National Laboratory (INL) has entered the final design phase and is now soliciting proposals from private enterprises for advanced nuclear energy technology experiments and demonstrations. MARVEL is a sodium-potassium cooled micro-reactor with 85 kW thermal/20 kW electrical output, which will provide real-world testing for nuclear energy applications such as microgrids, remote operations, and process heat utilization, with a target operational date by the end of 2027. This opening will promote innovation in the micro-reactor industry and accelerate the development and commercialization of advanced nuclear energy technologies.
▌UK Team Launches AI Tool to Support Green Chemistry Development
A team from the University of Nottingham in the UK has developed a new AI tool to help chemists and engineers achieve more sustainable chemical processes. Researchers optimized commonly used AI algorithms to improve the predictive capability of molecular retrosynthesis pathways and integrated it into the open-source electronic lab notebook platform AI4Green for global user access. The team also introduced a machine learning-based method for selecting green solvents to reduce environmental impact and foster students’ awareness of sustainable chemistry through the AI4Green electronic lab notebook. These innovations are expected to promote green chemistry practices and assist the industry in transitioning to environmentally friendly practices.
▌U.S. Biomimetic Swarm Robots Achieve Blueprint-Free Autonomous Construction
Engineers at the University of Pennsylvania in the U.S., inspired by insect swarms, have developed the world’s most flexible robotic swarm system that can autonomously construct structures without blueprints or coordination. The team designed mathematical rules that allow micro-robots to react autonomously to their surrounding environment and assemble collectively, demonstrating high adaptability and resilience. Simulation experiments show that adjusting the local behavior of robots can automatically generate diverse structures, enhancing overall crack resistance. This decentralized strategy has the potential to revolutionize traditional manufacturing and construction methods, promoting the application of self-organizing robots in complex environments.
▌Cambridge Team Develops Low-Cost Multisensory Robot Skin
A joint research team from the University of Cambridge and University College London has developed a new type of stretchable “robot skin” that is soft and highly conductive, capable of covering a robot’s hand and sensing pressure, temperature, pain, and multi-point touch in real-time. This skin can recognize various contact information through 860,000 signal pathways and AI algorithms, and it is easy to manufacture and low-cost. Compared to traditional multi-sensor solutions, this single-material multimodal sensing technology is more durable and suitable for large-scale applications, with potential uses in humanoid robots, prosthetics, automotive, and disaster rescue fields.