Presentation
XPert: Peripheral Circuit & Neural Architecture Co-search for Area and Energy-efficient Xbar-based Computing
DescriptionDeep Neural Networks (DNNs) have achieved state-of-the-art performance in large scale tasks such as image classification, segmentation and object detection among others. In the context of energy-efficient deep learning, In-Memory Computing (IMC) architectures are a promising alternative for overcoming the memory wall bottleneck of traditional von- Neumann computing. However, it is found that peripheral circuits are the significant contributors to energy and the area overhead. To this end, we propose XPert, a fast network architecture and crossbar peripheral circuit co-search framework for area
and energy-efficient IMC-implementations. XPert achieves 3.5× lower EDAP than VGG16 networks at
iso-accuracy at 20mm2 area.
and energy-efficient IMC-implementations. XPert achieves 3.5× lower EDAP than VGG16 networks at
iso-accuracy at 20mm2 area.
Event Type
Research Manuscript
TimeWednesday, July 12th3:40pm - 3:55pm PDT
Location3004, 3rd Floor
Design
In-memory and Near-memory Computing Circuits