Near-data-processing for Data-Intensive Applications
Abstract:
The informationtechnology sector has experienced explosive growth in data-intensiveapplications such as bioinformatics, big data analytics, and deep neuralnetworks (DNNs). These computing tasks have a tremendous economic impact andsocietal benefits, but their execution on conventional Von Neumannarchitectures is inefficient due to excessive data movement, a problem thatrapidly growing input data sizes have exacerbated. To tackle this bottleneck,the computer architecture research community has put forward many data-centricsolutions that place logic inside memory or the disk drive, commonly referredto as Near-data-processing (NDP), to reduce the latency and energy cost of dataaccess significantly. Additionally, NDP architectures usually offer much largerparallelism, higher data bandwidth, and lower peak power consumption than CPUand GPU, allowing them to achieve orders of magnitude speedup and energy savingwhen executing data-intensive kernels.
This dissertationoutlines four new contributions to NDP, including (1) a digital bit-serialDRAM-based processing scheme that targets a wide range of computing tasks,including bioinformatics, data analytics, pattern matching, and general-purposearithmetic, (2) a 3D-stacked memory technology with an integrated compute layerthat accelerates de novo genome assembly, (3) aprocessing-with-storage-technology (PWST) HW/SW codesigned framework thattargets k-mer counting, a key bottleneck of many bioinformatics tasks, and (4)a case study of how privacy and data integrity can be breached in a recentNDP-based DNN accelerator leveraging the non-volatile memory technologies(NVM), highlighting the importance of fostering future NDP accelerator designwith a security focus.
Committee:
- Mircea Stan, CommitteeChair (ECE/SEAS/UVA)
- Kevin Skadron, Advisor (CS/SEAS/UVA)
- Ashish Venkat, Co-advisor (CS/SEAS/UVA)
- Felix Lin (CS/SEAS/UVA)
- Adwait Jog (CS/SEAS/UVA)
- Sandhya Dwarkadas (CS/SEAS/UVA)