Tensor Processing Unit: Accelerating AI Workloads Arshiq A Under the guidance of
A S Arshiq, R Anoop, Sheena K M · 2025
This paper explores the architecture and impact of the Tensor Processing Unit (TPU), a specialized hardware accelerator developed by Google to enhance machine learning workloads. TPUs are tailored for deep learning applications, offering high throughput and energy efficiency for training and inference tasks. By leveraging matrix multipliers, systolic array design, and dedicated memory bandwidth, TPUs outperform general-purpose CPUs and GPUs in AI tasks. The paper analyzes TPU's role in model deployment, its integration into cloud services, and its contributions to advancing artificial intelligence. Future directions include improvements in hardware scalability, energy optimization, and hybrid computing frameworks.