PhD Candidate
Cornell University
Areas of Interest
- Artificial Intelligence
- Computer Architecture and Engineering
- Cyber-Physical Systems and Design Automation
- Integrated Circuits
- Reconfigurable Computing
- High-Level Synthesis
- Physical Design
Poster
Machine Learning-Assisted Early Timing Prediction for Rapid FPGA Design Closure
Abstract
With the recent trends in technology scaling, specialized hardware accelerators such as field-programmable gate arrays (FPGAs) are increasingly employed to achieve high performance and energy efficiency. However, weak guarantees of existing CAD tools on achieving FPGA design closure is a significant barrier to their adaptation. Current methodologies require extensive manual efforts to explore a wide spectrum of optimizations across multiple design stages. Design closure has become remarkably challenging due to the size and complexity of the search space spanned by these optimizations and the time-consuming design stages such as placement and routing. The ultimate goal of my research is to achieve rapid end-to-end design closure and contribute to the rise of specialized accelerators. To realize this goal, my research focuses on developing machine learning-assisted methodologies to automatically learn from the inherent structure of today's designs and technologies, and hereby introducing high-fidelity predictions in early and low-fidelity design stages.
Bio
Ecenur is a Ph.D. candidate in Electrical and Computer Engineering at Cornell University, advised by Prof. Zhiru Zhang. She is vastly interested in accelerating FPGA design closure by leveraging various machine learning techniques (e.g., graph representation learning, reinforcement learning) and domain specification. She was a research intern at Xilinx Research Labs in Summer 2020. Ecenur received her B.S. degree in Electrical and Electronics Engineering from Bogazici University, Istanbul, Turkey in 2016. In Summer 2015, she was an undergraduate researcher at MIT. For links to papers and videos, please visit http://people.ece.cornell.edu/eu49/.
