AI Stack and Hexagon Tensor Processor: Empowering Edge Inferencing

Title of the Talk: Stack and Hexagon Tensor Processor- Empowering Edge Inferencing
Speakers: Dr. Saurabh Tewari
Host Faculty: Rajesh Kedia
Date: April 17, 2025
Time: 2:30 pm
Venue: CS-LH2

Abstract:
This presentation explores Qualcomm’s advanced technologies, focusing on the Hexagon Tensor Processor and AI Stack. It will provide an overview of Qualcomm’s role in advancing AI and machine learning through innovative hardware and software solutions. The presentation will also cover the high-level hardware architecture of the Hexagon Tensor Processor, its software architecture, and various optimization stages. Additionally, the presentation will introduce AI/ML sequencers and compilers, and their role in optimizing AI models for deployment, as well as profiling and debugging tools that help identify performance bottlenecks and optimize AI applications. The presentation will give an understanding of how Qualcomm’s technologies are driving advancements in AI, particularly in the context of deep neural network optimization and acceleration on edge devices.

Speaker Bio: Dr. Saurabh Tewari is a Senior Staff Engineer/Manager at Qualcomm, specializing in AI and machine learning. With a B.Tech from IIT Roorkee, an M.Tech, and a Ph.D. from IIT Delhi in Computer Science, Dr. Tewari has over 19 years of industrial experience. He has worked in software optimizations for embedded systems, 3G wireless protocol stack development, Global Positioning Systems (GPS), and MEMS algorithm development. His Ph.D. research focused on optimizing neural networks’ performance on parallel architectures. Since joining Qualcomm in 2022, Dr. Tewari has been contributing to the development of the Qualcomm Hexagon Tensor Processor and AI Stack, and is currently focused on optimizing deep learning models for edge devices.