Multi-Layer Digital Twin for Terahertz Data Centers

This research introduces a sophisticated multi-layer digital twin (DT) framework specifically designed for Terahertz (THz) wireless data centers, aiming to meet the escalating demands for flexible and high-capacity interconnections driven by AI computing. The proposed framework systematically constructs physical, channel, evaluation, and manipulation layers, building from foundational measurements upwards. The process begins with extensive tri-band channel measurements conducted at 140, 220, and 300 GHz to thoroughly characterize the frequency-dependent propagation behaviors within a data center environment. These measurements are then used to calibrate a physical twin, which involves jointly optimizing geometric, material, antenna, and hybrid propagation models. Building upon this calibrated physical twin, an AI channel twin is developed using a line-of-sight (LoS)-aware implicit neural field. This AI twin learns location-dependent channel statistics, enabling efficient and real-time prediction of received power and LoS probability. Finally, a system-level evaluation layer is derived from the reconstructed channel field to analyze coverage and interference for both access point-to-rack and rack-to-rack communications. Experimental results demonstrate that the AI twin achieves lower power reconstruction error compared to existing neural-field baselines while maintaining real-time inference capabilities. The study also shows that ceiling-mounted access point deployment can achieve over 90% coverage under a 10 dB SINR threshold, validating the effectiveness of the DT framework for THz wireless data center planning and optimization.