Dipankar Das
Papers
1
Total Citations
462
H-Index
1
About
Dipankar Das is a computer architecture and systems researcher whose work sits at the intersection of deep learning and hardware acceleration. His most notable contribution is SIGMA (Sparse and Irregular GEMM Accelerator), a landmark 2020 paper that addresses one of the fundamental challenges in training deep neural networks: efficiently handling sparse and irregular matrix multiplication workloads. Published to significant acclaim, SIGMA has accumulated over 460 citations, reflecting its substantial influence on the hardware design community. The work introduced a flexible interconnect architecture that enables efficient general matrix multiplication (GEMM) operations — the computational backbone of virtually all deep learning training pipelines — even under sparse and irregular data patterns that typically bottleneck conventional accelerators. By tackling inefficiencies across vision, speech, language, and recommendation workloads, Das's research speaks to the breadth of modern AI applications. His contributions are particularly relevant as the research community grapples with scaling neural network training efficiently amid exploding model sizes. For students and practitioners in computer architecture or ML systems, Das's work represents a critical bridge between algorithmic demands and practical hardware innovation.
Research Focus
Key Achievements
Top Papers
- 1