Kartik Hegde is a second-year PhD student at the University of Illinois at Urbana-Champaign, advised by Chris Fletcher. He received a B.Tech from NITK, Surathkal in 2015. His research focuses on developing efficient hardware accelerators for massively heterogeneous systems.
As computer architects turn towards hardware accelerators to satisfy the increasing performance and power efficiency demands, significant research effort is required towards designing reusable hardware abstractions, endowing programmability and flexibility in execution, achieving high utilization of resources, supporting dynamic power and performance modes etc. In this light, Kartik’s research broadly aims to develop a hardware accelerator that reaches “the efficiency of ASICs and the programmability of CPUs”.
Previously, Kartik’s research involved developing hardware accelerators for deep learning which proposed ways to dramatically reduce computations in inference by exploiting repetition in DNN models. His previous works also explored the design of a flexible accelerator for DNN inference by developing novel hardware abstractions to endow flexibility and programmability to accelerators for efficient inference at the edge. His work proposes high-performance programmable address generators for dense, sparse and quantized DNNs, representation-independent traversals of sparse data structures, flexible memory hierarchies, and other mechanisms for improved flexibility.
Visit his website for more information.