MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202631072117 A) filed by C. V. Raman Global University on June 10, 2026, for Ephemeris-Driven Predictive Sharding For Thermalaware Distributed Orbital Computing.
Inventors include Debadutta Sahoo; Ramakant Senapati; Debankur Pal; Asmita Agarwal; and Dr. Soumya Mishra.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: The present invention discloses an ephemeris-driven predictive sharding system and method for thermal-aware distributed orbital computing, comprising a constellation management architecture in which a plurality of orbital compute nodes (OCNs), each equipped with a radiation-hardened AI accelerator, a passive thermal radiator system, and an inter-satellite optical link terminal, cooperatively execute a continuous thermal prediction and workload orchestration framework. An Ephemeris Predictor Module (EPM) on each OCN computes, from stored Two-Line Element (TLE) orbital data, real-time solar incidence angles and time-to-eclipse (TTE) values for all nodes in the constellation neighbourhood, classifying each node into a defined thermal state and identifying those nodes imminently entering or currently within the planetary shadow. A Model Sharding Engine (MSE) decomposes monolithic neural network inference models into a directed acyclic graph of independently executable computational shards, each annotated with a thermal dissipation cost value. A Follow-the-Shade Routing Protocol (FSRP) assigns the highestdissipation shards exclusively to eclipse-phase nodes, synchronising peak computational heat generation with maximum radiative cooling capacity and eliminating the performance degradation imposed by conventional reactive thermal throttling mechanisms. A Migration Cost Analysis (MCA) module enforces energetic self-consistency by restricting shard transfers to scenarios where thermal benefit exceeds inter-satellite link transmission cost. A Thermal Inertia Extension (TIE) mechanism exploits post-eclipse processor thermal mass to extend productive computation windows beyond strict eclipse boundaries. The system operates autonomously via a gossip-based distributed consensus protocol without ground station intervention, achieving approximately twofold improvement in constellation-wide AI inference throughput and significantly reducing processor thermal cycling amplitude compared to throttling-based prior art, thereby extending accelerator silicon operational lifetime and enabling deterministic real-time latency guarantees for mission-critical orbital edge AI applications.
Disclaimer: Curated by HT Syndication.
