2025/11/20 Prof. Zhou awarded 10-Year Most Influential Paper at ICCAD 2025!

📣 📣 📣 Prof. Zhou was honored with the 10-Year Retrospective Most Influential Paper Award at the 2025 Institute of Electrical and Electronics Engineers (IEEE)/Association for Computing Machinery (ACM) International Conference on Computer-Aided Design (ICCAD), held in Munich, Germany in October. The award recognizes the lasting impact of the 2016 paper, “Caffeine: Towards uniformed representation and acceleration for deep convolutional neural networks,” co-authored by Zhou with her Ph.D. adviser Jason Cong (UCLA), Chen Zhang (Shanghai Jiao Tong University), Guangyu Sun (Peking University), Zhenman Fang (Simon Fraser University), and Peichen Pan (Falcon-Computing Solutions).

The paper presented Caffeine, a full-stack design automation tool that synthesizes deep neural networks specifically onto field-programmable gate-arrays (FPGA). At the hardware level, it automatically extracts operator sets from AI models, optimally maps them to a reconfigurable intermediate representation, and generates efficient FPGA hardware implementations. At the software level, it automatically generates instruction streams based on the hardware configuration (matrix engines, caches, pipelines, etc.), optimizing operator scheduling and memory management.

Deep neural networks are built from distinct computational stages called layers. Before Caffeine, most FPGA accelerators focused only on the convolution layers, which limited performance at the fully-connected layers. Caffeine introduced a novel unified convolutional representation to efficiently accelerate the entire network on a single FPGA. This approach included key techniques like automated memory access transformation to solve the critical challenge of optimizing for the limited memory bandwidth of FPGAs. The paper’s lasting influence is highlighted by its nearly 800 citations, including from nearly every major company in the AI hardware acceleration industry such as AMD, Google, Intel, and Nvidia.

Brown Engineering News

LinkedIn