Test and Reliability Aspects in Neuromorphic Computing

 

Krishnendu Chakrabarty

Krishnendu Chakrabarty , Duke University, USA

Syllabus

The Lecture will cover the following topics:

Part 1 (30 minutes)
  • Motivation for neuromorphic computing
  • Memory wall, data transfer bottleneck, limitation of current architectures
  • Basics of RRAM
  • Introduction to RRAM crossbars, training and inferencing
  • RRAM crossbar-based architectures: PRIME, ISAAC, PipeLayer, TIME
Part 2 (1 hour)
  • Defects and variations, fault models
  • Impact of defects and variations on training and inferencing
  • Defect understanding and impact analysis
  • Testing methods
Part 3 (1 hour)
  • Fault tolerance: Training
  • Fault tolerance: Inferencing