*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Design Mehodology for 3D-stacked Imaging Systems with Integrated Deep Learning
Committee:
Dr. Saibal Mukhopdhyay, ECE, Chair , Advisor
Dr. Sudhakar Yalamanchili, ECE
Dr. Asif Khan, ECE
Dr. Tushar Krishna, ECE
Dr. Paul Kohl, ChBE
Abstract:
The Internet of Things (IoT) revolution has brought along with it billions of always on, always connected devices and sensors. Associated with these billions of sensors are huge amounts of data that must be transmitted to an off-chip host for classification. However, sending these large volumes of unprocessed data incurs large latency and energy penalties, often hindering proper performance and functionality of IoT systems which are typically resource constrained. Instead, moving computations to the sensor side can reduce data volume and hence improve performance and energy efficiency of the end application.
The objective of the presented research is to explore sensor integrated computing with Neurosensor, a 3D-stacked image sensor with integrated deep learning. Such an architecture enables the deployment of smart sensors that perform advanced neural acceleration in-field, with 3D integration helping to avoid the various pitfalls (such as routing challenges and associated latency) of designing this type of high bandwidth systems with a large degree of parallelism. The architecture of the system is explored and the various design trade-offs are investigated. Next, we examine technology based solutions to further increase system performance through the use of 3D stacked digital sensors and emerging device based processing-in-memory neural accelerators. Furthermore, the various circuit issues involved with the design of these sensor based systems are investigated through the discussion of post-silicon results from an image sensor SOC. Finally, the dissertation concludes with a brief discussion on how energy harvesting can be used to power these systems.
