Reconfigurable Terahertz Neuromorphic Chip for Penetrative Dynamic Vision and Compressed Information Transmission

Terahertz (THz) waves have enormous potential for high‐penetration imaging and broadband information transmission with high biosafety. However, due to challenges such as non‐volatile regulation and the construction of a high‐density mode, there is no THz‐based dynamic visual sensing‐storage‐computing system. Here, we demonstrate a THz neuromorphic chip integrating high‐penetration dynamic vision, high‐density information transmission, and in‐sensor computing. By integrating a specially designed metasurface with a ferroelectric liquid crystal layer to establish a dual‐physical‐layer mechanism, this chip utilizes 6‐level states across 16 input ports and triple‐frequency multiplexing to dynamically encode 3 × 6 16 vortex interference fields (scalable to m × S n ) in the spatiotemporal‐frequency domain, achieving a 27‐fold performance enhancement over prior works. Using an artificial neural network and Caesar transformation, the device implements a brute‐force‐attack‐resistant high‐dimensional encryption transmission scheme with a bit error rate of 0.09%, representing a leading performance in THz wireless communications. Furthermore, via multi‐level encoding and two‐dimensional complex‐amplitude output, the system achieves motion direction recognition under low‐visibility environments, with an experimental accuracy of 92.3% and a horizontal visibility range 12‐fold that of near‐infrared devices. This strategy integrates the superiorities of THz techniques and neuromorphic computing, potentially applicable to dynamic artificial vision with high‐penetration, dynamic information processing, compressed and encrypted information transmission abilities.