The New Era of AI Hardware

Chapter 3 by Jeff Burns illustrates IBM's current efforts and future plans for deep learning acceleration. AI systems are necessary to handle the expected 40 trillion gigabytes of data generated by over 30 billion IoT devices in the near term. AI research consists of data requirements, algorithms for data manipulation, and then computing for prediction from the data. Accelerating AI computing involves software, architecture, and hardware development. Accelerating training and inference are two goals of AI hardware research.