Tensor processing unit
Vinod Kumar Khanna · 2025
The terms ‘tensor’, ‘TensorFlow’ and ‘tensor processing unit (TPU)’ are introduced. Chronological history and features of TPU are described and its specialties for AI work are enumerated. The driving factors behind the motivation for TPU become evident by understanding the components of a perceptron and the calculation steps performed in its working. The structural units of TPU and its instruction set are described, including both off-chip and on-chip components and the complex instructions set computer (CISC) instruction set. TPU operation is explained by drawing the block diagram of the workflow of an inference task and discussion of inference implementation by TPU.