SS10: ANN HW-Accelerators
Mario Porrmann
Osnabrück University, Germany
Ulrich Rückert
Bielefeld University, Germany
Abstract
Artificial neural networks (ANNs) are simple models of abstract brain-like computing principles performing massively parallel computations for artificial intelligence tasks. As the performance increase of microprocessors slowed down, the demand for application-specific solutions for ANN architectures is increasing in applications like autonomous driving, cognitive robotics, cognitive edge computing, and Internet-of-Things. Many analogue, digital and mixed analogue/digital chip implementations have been proposed in the last 40 years. In nanoelectronics, the growing complexity of ultra-large-scale integration turns difficult problems (e.g., power, reliability, testing, connectivity, design complexity, design support, …) into great challenges. Because of these challenges, digital implementations are dominating today.
In this special session, recent hardware implementations for ANNs will be discussed. Nowadays, the number of chip proposals from academia and companies (from start-ups to enterprises) is heavily increasing. The focus is on ANN models such as Deep Neural Networks, Long Short-Term Memory, and Auto-Encoder Networks, which are mainly used in applications today. Their performance will be compared to ANN implementations on commercial off-the-shelf chips like multi-core chips, graphics processing units, and field-programmable gate-arrays. Based on current applications, future requirements for ANN HW accelerators will be estimated.
Organizers
Dr. Mario Porrmann is professor and head of the “Computer Engineering” group at Osnabrück University, Germany, since April 2019. He graduated as “Diplom-Ingenieur” in Electrical Engineering at the University of Dortmund, Germany, in 1994. In 2001, he received a PhD in Electrical Engineering from the University of Paderborn, Germany, for his work on “Performance Evaluation of Embedded Neurocomputer Systems”. From 2010 to March 2012, he was Acting Professor of the “System and Circuit Technology” research group at the Heinz Nixdorf Institute, University of Paderborn. He then joined the research group “Cognitronics and Sensor Systems” in the Center of Excellence “Cognitive Interaction Technology” at Bielefeld University as Academic Director. Mario Porrmann’s scientific work focuses on resource-efficient computing with a special emphasis on adaptive heterogeneous architectures for embedded systems and cognitive edge computing.
Dr. Ulrich Rückert studied Computer Science (Dipl. Inf.) and received a Dr.-Ing. degree in Electrical Engineering from the University of Dortmund, Germany (1989). From 1992 to 1995 he was Associate Professor at the Technical University of Hamburg-Harburg, Germany. In 1995 he joined as a Full Professor the Heinz Nixdorf Institute at the University of Paderborn, Germany, heading the research group ‘‘System and Circuit Technology’’. Since 2009 he is Professor at Bielefeld University, Germany heading the “Cognitronics and Sensor Systems” group of the “Cluster of Excellence – Cognitive Interaction Technology”. He is working on innovative circuit design and development of nanoelectronic systems for massively parallel and resource-efficient information processing. His main research interests are bio-inspired architectures for nanotechnologies and cognitive robotics.