Prof. Yu’s Team Won the “Qilin Cup” from the Integrated Circuit EDA Elite Challenge 2023!

Fangzhou Liu, Wuqian Tang, and Ruijie Wang, guided by Prof. Bei Yu and Prof. Chun-Yao Wang, earned the “Qilin Cup” (top-1 from all 400+ teams) for excelling in the challenge “Intelligent Synthesis Recipe Search in Logic Synthesis” at the 2023 Integrated Circuit EDA Elite Challenge.

From December 22nd to 24th, the 5th Integrated Circuit EDA Elite Challenge were held in Nanjing, closely aligned with the development needs of EDA design industry. Challenges were jointly crafted by corporate technical experts and university professors with expertise in pertinent research areas, encompassing a comprehensive set of nine EDA topics such as logic synthesis, placement and routing, static timing analysis, layout generation, and constraint solving, to name a few.

The team, “TcT’’, tackled the challenge towards combinational logic optimization and technology mapping by utilizing the domestic EDA open-source tool, “imap.” The challenge entailed devising an algorithmic solution aimed at efficiently navigating the design space for synthesis recipes—specifically, optimization sequences—while adhering to constraints on runtime and targeting quality of results (QoR). Team “TcT” stood out in the technical review and group defense among 27 competing teams, and then went on to receive unanimous praise from the judges in the culminating showcase round.

Abstract: Combinational logic optimization is a critical step in logic synthesis, where open-source tools commonly apply a series of operators such as rewriting, refactoring, and balancing to Boolean networks (e.g. And-Inverter Graphs). This process aims to streamline the network structure, thereby effectively reducing circuit’s area and delay. Given that identifying the optimal optimization sequence involves exploring an exponentially large solution space, devising an efficient method is especially important. Team “TcT” implemented the BETuing framework, which integrates Contextual Multi-Armed Bandits (CMAB) with Evolutionary Algorithms (EA). This hybrid approach enables a hierarchical search from fine-grained to coarse-grained levels, maximizing the extension of the optimization sequence within a constrained time period. Experimental results show that BETuning’s framework secured top-three placements for both area and delay optimizations in 63% and 97% of all the testcases, respectively, among all competing teams.