2026-09-19T04:40:04Z https://api.figshare.com/v2/oai

oai:figshare.com:article/32995724 2026-05-01T00:00:00Z portal_693 item_type_8 month_year_05_2026

Gaussian Splatting Device-Architecture Co-Design for Accelerated Physical AI Inference and Reasoning Sureshkumar Senthilkumar (24400619) Computer Science Engineering, Electronics and Electrical Modern autonomous systems, including robotics, autonomous driving, and XR/AR, demand 3D scene representations that are simultaneously photorealistic, memory-efficient, and computationally lean. Gaussian Splatting has established itself as the state-of-the-art technique for 3D scene representation, delivering photorealistic rendering quality that surpasses traditional methods. However, this fidelity comes at a steep computational cost, requiring extensive memory bandwidth and desktop-grade GPU resources to achieve real-time performance, which effectively precludes deployment on edge devices. This thesis addresses these hardware limitations through a novel device-architecture co-design leveraging Gaussian transistors. Unlike conventional silicon devices, the Gaussian transistor exhibits a current-voltage (I-V) characteristic that naturally mimics the Gaussian function, enabling intrinsic computation of splatting operations. We present a specialized accelerator utilizing this technology to minimize computational overhead. This research also demonstrates the practical utility of this accelerator in enabling real-time, power-efficient physical reasoning and inference within the 3D world. 2026-05-01T00:00:00Z Text Thesis 10.25417/uic.32995724.v1 https://figshare.com/articles/thesis/Gaussian_Splatting_Device-Architecture_Co-Design_for_Accelerated_Physical_AI_Inference_and_Reasoning/32995724 In Copyright Open Access after 2031-05-01