MUMBAI, India, March 13 -- Intellectual Property India has published a patent application (202541020842 A) filed by Amrita Vishwa Vidyapeetham, Guntur, Andhra Pradesh, on March 7, 2025, for 'energy-efficient neural processing system for embedded artificial intelligence inference.'
Inventor(s) include Bikku, Thulasi.
The application for the patent was published on March 13, under issue no. 11/2026.
According to the abstract released by the Intellectual Property India: "An energy-efficient neural processing system (100) for embedded AI inference includes an input preprocessing module (102) generating preprocessed input data and a neural processing unit (NPU) (104) executing neural network inference. A dynamic workload analysis module (106) determines workload demand from at least one of the input data, the preprocessed input data, or runtime inference metrics from the NPU (104). A dynamic voltage and frequency scaling (DVFS) module (108) adjusts voltage or frequency of the NPU (104) based on the workload demand. A hierarchical memory (110) reduces data-movement energy, based on memory-access patterns from the NPU (104), via energy-aware caching, memory compression, or locality-driven data retrieval. An approximate computing module (112) applies approximate arithmetic for non-critical computations while meeting an accuracy threshold. An energy feedback loop module (114) monitors energy usage or performance via counters and controls the DVFS module (108), the hierarchical memory (110), and the approximate computing module(112)."
Disclaimer: Curated by HT Syndication.
