Architecture of Computing Systems
24-27 March 2015
Tutorial: Myriad2 architecture, programming and computer vision applications
Date and Time: 9AM-12:30AM, March 25, 2015
Mircea Ionica, Movidius Ltd, Ireland
Michael Bromberger, KIT, Germany
David Moloney, PhD, CTO & Founder, Movidius Ltd., Ireland
For the past 40 years computer scientists and engineers have been building technology that has allowed niche machine vision from factories to Mars rovers. Until now the availability of computational power has limited the application of these technologies to niches with a strong enough need to overcome the cost and power hurdles
This is changing rapidly and the computational means have now become available to bring computer vision to mass market applications in mobile phones, tablets, drones and robots enabling brand new user-experiences within the cost, power and volumetric constraints of mobile platforms.
Movidius is focused on bring human vision and scene understanding to mobile devices rather than exclusively in the cloud allowing low power always-on vision capabilities in devices with the very low latencies required for interactive services, self-driving cars and robots etc.
Movidius is enabling a swathe of new computer vision applications to be brought to the mass market for the very first time in embedded devices such as mobile phones, tablets and cameras. And the first generation of our Myriad VPU powers the computer vision subsystem in Google’s Project Tango where we handle all of the high performance ISP, feature tracking and tracking tasks in 10x less power than any other solutions available on the market today
This video gives more detail on what project tango does
A very detailed technical presentation on our new 28nm Myriad2 product that received an enthusiastic welcomed at HotChips 2014 in Cupertino (CA) in August is available here (starts at 28:00):
The tutorial will cover:
1. Challenges and rationale for Vision Processing unit
2. Myriad2 hardware architecture including SHAVE processor, SIPP image/video processing HW accelerators, multicore memory subsystem and SoC and communications interfaces
3. Software tools and design methodology including ISP/video pipeline GUI and runtime environment
4. Video pipeline examples including:
o ISP pipeline
o FAST9 feature detector
5. Demonstrations of pipelines and hands-on examples of how to program Myriad2
Short Speakers’ Bios:
Mircea-Horea Ionică is an embedded software engineer with Movidius, currently pursuing a PhD from the school of Computer Sciences, Trinity College Dublin. During his career in Movidius, he led the firmware development for the company's first chips, and made contributions to drivers, RTOS support, and tools development; his research in TCD focuses on performing scientific computing on embedded systems, with a particular focus on the Myriad architecture. He graduated with a degree in Informatics from the Politehnica University of Timișoara, Romania.
Michael Bromberger is currently a Ph.D. student in computer science at the Karlsruhe Institute of Technology. He received his diploma in computer science at the Karlsruher Institute of Technology in 2013. In 2014 Michael did a four months HiPEAC internship on approximate computing at Movidius. His research interests include special hardware accelerators, FPGAs, hybrid systems, and approximate computing. Therefore, he considers applications from computer vision, computational biology, machine learning and uncertainty quantification. Furthermore, he is a member of the data mining and uncertainty quantification group at the Heidelberger Institute of Theoretical Studies.
David Moloney, Chief Technology Officer, has worked for over 25 years in the semiconductor industry since qualifying with a BEng from DCU in 1985. He has a wealth of international experience having worked for Infineon (Siemens Semiconductor Division) in Munich for 5 years and SGS- Thomson Microelectronics (STM) in Milan for 4 years respectively. In 1994 he returned from STM to lead the engineering team for the first product development at Parthus Technologies where he was a key member of the management team and where he spearheaded the development of the Parthus Bluetooth technology. David left Parthus in 2003 to work towards his PhD in Trinity College Dublin and as an independent consultant for Frontier Silicon and Dublin City University. He received a PhD from Trinity College Dublin in 2010 for his research into high performance computer architectures. He has participated in FP7 projects like PEPPHER and EXCESS as well as the forthcoming H2020 Eyes of Things (EoT) project on low-power wearable cameras. David is inventor/co-inventor of 18 issued US patents, with an additional 13 patents pending and is the author of over 20 conference and journal papers.