Researchers built more energy-efficient and high-performance phase change random access memory by learning how to sculpt heat within a nanometer-thin sandwich of materials. They engineered a "double-confined" structure that stacks alternating slices of antimony telluride (Sb₂Te₃) with two distinct helper compounds: highly conductive nickel ditelluride (NiTe₂) and heat-blocking molybdenum ditelluride (MoTe₂).

As artificial intelligence moves beyond controlled settings and into dynamic, uncertain and socially complex environments, new questions arise about how these systems understand their surroundings, act autonomously, and interact with people. Rather than concentrating intelligence in a single complex system, the four-year EIC Pathfinder project EMERGE took a different approach. It considered the simplest possible artificial agents, such as small robots in a swarm, components of a robotic body, or connected devices. Individually, each unit may possess limited intelligence and information. Together, however, from their interactions, sharing information and coordinating their behaviour, they can establish a representation of shared existence, environment and goals, which the project calls collaborative awareness.

Graph neural networks have been used in various graph-related tasks, where they have achieved excellent performance. However, incomplete graph structures, commonly found in real-life scenarios, can severely degrade performance. In a new study, researchers develop a novel graph completion network that reconstructs missing features and structure simultaneously while also maintaining performance with extremely weak supervision, which is difficult to achieve with conventional models.

Researchers have demonstrated a scalable 3.2 Tb/s O-band silicon photonic interconnect architecture designed to address the growing bandwidth demands of artificial intelligence (AI) computing. Published in PhotoniX Synergy, the work combines hybrid wavelength-port multiplexing with adaptive wavelength stabilization to enable high-bandwidth and robust optical I/O, providing a scalable pathway toward terabit-scale silicon photonic interconnects for next-generation AI computing systems.