MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202621081957 A) filed by Teqsil Labs Private Limited on July 02, 2026, for Artificial Intelligence Processing Apparatus With Shared Register Fabric For Concurrent Dense-Tensor And Spiking-Neural Processing.
Inventors include Ms. Trushna Nishant Kasture; and Mr. Nishant Dinkarrao Kasture.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: ABSTRACT ARTIFICIAL INTELLIGENCE PROCESSING APPARATUS WITH SHARED REGISTER FABRIC FOR CONCURRENT DENSE-TENSOR AND SPIKING-NEURAL PROCESSING The present invention relates to an artificial intelligence processing apparatus (100), a corresponding method (200), and a System-on-Chip configured for concurrently executing heterogeneous neural-network workloads within a unified processing architecture. The apparatus (100) comprises an artificial intelligence processing core (100) supporting a plurality of hardware execution threads (102, 104), the mode-select register (108), a shared instruction decoder (106), a shared register fabric including a tensor register file (110) and a neural-state register file (112), and a shared on-core scratchpad (118). The mode-select register (108) selectively assigns computational paradigms to the hardware execution threads (102, 104), while the shared instruction decoder (106) selectively dispatches instructions to the tensor register file (110) or the neural-state register file (112) according to the assigned computational paradigm. The disclosed architecture reduces communication latency, external memory traffic, hardware duplication, and energy consumption while improving execution throughput, silicon utilization, and computational efficiency for heterogeneous artificial intelligence applications. FIG. 1 will be the reference figure.
Disclaimer: Curated by HT Syndication.
