MUMBAI, India, July 24 -- Intellectual Property India has published a patent application (202641085356 A) filed by Sri Eshwar College Of Engineering on July 11, 2026, for Hierarchical Predictive Leakage Orchestration Controller For Leakage Power Optimization In Neural Processing Units For Industrial Edge Ai Applications.
Inventors include Ms. K. Gowthami; Ms. J. Yashwandra; Mr. S. Aravind; and Dr. R. Babitha Lincy.
The application for the patent was published on July 17, 2026, under issue no. 29/2026.
Abstract: The present disclosure relates to a Hierarchical Predictive Leakage Orchestration Controller for adaptive leakage power optimization in Neural Processing Units (NPUs) for industrial edge Al applications. The present invention is directed to leakage management in advanced semiconductor technology, wherein the commonly used leakage reduction techniques may operate independently and are often not able to adapt to changing workloads, process variation, temperature, device aging and Quality-of-Service (QoS) parameters.The proposed controller collects runtime information such as information on the current workload, temperature, voltage, leakage characteristics, timing information and historical information on how the chip was used in the past. This information is then processed by a predictive leakage model, which is based on machine learning that predicts future leakage before it results in excessive energy dissipation. The hierarchical orchestration strategy then makes decisions on leakage control at block, core and chip level to ensure that sufficient performance, thermal and reliability figures are met while applying the optimal leakage control. Leakage optimizer controller generates unified hardware control commands by intelligently combining multiple leakage power optimization techniques, i.e., power gating, clock gating, Dynamic Voltage and Frequency Scaling (DVFS), adaptive body biasing, sleep transistor control, memory retention, and program operating mode transition. It supports adaptive feedback to automatically and iteratively refine its prediction models and future program execution optimization policies, to learn and optimize it to effectively handle various system execution environments.For example, in the scenario of an industrial robotic assembly line where the AI-powered visual inspection of assembled parts is carried out by an industrial robotic assembly system’s controller, the controller forecasts the computation idle times between the inspections and reduces the leakage in inactive processing resources while maintaining the real- time processing for the AI inference. This result in reduced leakage power and improved thermal management, extended lifespan of semiconductors, preserved performance of the inference, and increased overall energy efficiency and reliability of the industrial edge AI systems.
Disclaimer: Curated by HT Syndication.
