A Framework for AI-Native Semantic-Based Dynamic Slicing for 6G Networks

TL;DR

This work proposes a novel AI-native semantic slicing framework that integrates semantic encoding/decoding, semantic service class identification, and dynamic compute-network orchestration within a 5G service-based architecture. By decomposing data into semantic content elements and aligning network and computing resources to semantic tasks, it enables fine-grained, context-aware slice management and proactive resource provisioning. A first responder use case demonstrates how knowledge-graph based semantic representations can drive adaptive slicing across mission-critical tasks, while initial single-link results suggest reduced data transmission and improved semantic reliability. The framework outlines key open challenges in semantic representation, AI-driven orchestration, inter-slice interactions, and security, charting a path toward AI-native, cross-domain semantic networking for 6G.

Abstract

In the ensuing ultra-dense and diverse environment in future \ac{6G} communication networks, it will be critical to optimize network resources via mechanisms that recognize and cater to the diversity, density, and dynamicity of system changes. However, coping with such environments cannot be done through the current network approach of compartmentalizing data as distinct from network operations. Instead, we envision a computing continuum where the content of the transmitted data is considered as an essential element in the transmission of that data, with data sources and streams analyzed and distilled to their essential elements, based on their semantic context, and then processed and transmitted over dedicated slices of network resources. By exploiting the rich content and semantics within data for dynamic and autonomous optimization of the computing continuum, this article opens the door to integrating communication, computing, cyber-physical systems, data flow, and AI, presenting new and exciting opportunities for cross-layer design. We propose semantic slicing, a two-pronged approach that builds multiple virtual divisions within a single physical and data infrastructure, each with its own distinct characteristics and needs. We view semantic slicing as a novel shift from current static slicing techniques, extending existing slicing approaches such that it can be applied dynamically at different levels and categories of resources in the computing continuum. Further it propels the advancement of semantic communication via the proposed architectural framework.

Figures (7)

• Figure 1: Two functions of semantic slicing: data slicing and network slicing.
• Figure 2: Comparison between traditional/static resource provisioning, dynamic network slicing, context-aware communication, and proposed semantic slicing.
• Figure 3: Signaling involved in 5G Network Slicing
• Figure 4: Proposed Semantic Slicing Architecture. The additional components related to Semantics are highlighted in red. The VNF and resources dedicated to two different slices are highlighted in yellow and green. UE1 and UE2 belong to Slice1 and Slice2 respectively.
• Figure 5: Service Class Identification