Scheduler for mapping neural networks onto an array of neural cores in an inference processing unit

US12387082B2 · kind B2 · utility

Assignee

• International Business Machines Corporation · US

Inventors

• Pallab Datta · San Jose, US
• Andrew S. Cassidy · San Jose, US
• Myron D. Flickner · San Jose, US
• Hartmut Penner · San Jose, US
• Rathinakumar Appuswamy · San Jose, US
• Jun Sawada · Austin, US
• John V. Arthur · Mountain View, US
• Dharmendra S. Modha · San Jose, US
• Steven K. Esser · San Jose, US
• Brian Taba · Cupertino, US
• Jennifer Klamo · San Jose, US

Key dates

• Filing date: Jul 31, 2018
• Grant date: Aug 12, 2025
• Expiry date: Jul 15, 2041

Classification

• Technology area (CPC G): Physics
• CPC primary: G06N5/022
• WIPO field: Computer technology
• WIPO sector: Electrical engineering

Abstract

Mapping of neural network layers to physical neural cores is provided. In various embodiments, a neural network description describing a plurality of neural network layers is read. Each of the plurality of neural network layers has an associated weight tensor, input tensor, and output tensor. A plurality of precedence relationships among the plurality of neural network layers is determined. The weight tensor, input tensor, and output tensor of each of the plurality of neural network layers are mapped onto an array of neural cores.