Computer Architecture in Post-Moore Era

All contemporary computers are designed following the von Neumann architecture that separates computing from memory. Even before the eventual end of Moore's law, von Neumann architecture has been struggling with fundamental problems such as memory wall, prohibitively high energy cost of data transfer, and power (end of energy scaling and dark silicon) wall. These issues are especially challenging in data intensive applications such as machine learning or bioinformatics, where the effectiveness of classic von Neumann architecture is limited. In future years, when device sizes will no longer be scalable, computer and electronic industry will need innovations "beyond" Moore and von Neumann. A new architecture paradigm known as data centric processing (processing-in-memory) promises resolving many of von Neumann architecture inefficiencies by bringing computing to data, instead of bringing data to computing units.

I will discuss the headwinds faced by contemporary high-performance computing and present GIRAF, a non von Neumann computer architecture that intertwines processing with storage, and functions simultaneously as memory and a massively parallel processor. GIRAF combines associative processing with emerging (resistive) memory technologies. It outperforms state of the art solutions in a variety of data intensive applications, such as machine learning, deep learning and bioinformatics.

I will also briefly introduce bioinformatics, a new computing field which is quickly becoming a target of intensive hardware acceleration research. I will present my recent work in the field, including RASSA, a resistive accelerator for long read DNA mapping, and BioSEAL, a hardware accelerator for DNA sequence alignment.