芯和代文亮博士：AI时代，要把EDA 这条“脖子”练粗

Core Viewpoint - The EDA industry is undergoing a methodological reconstruction due to the exponential increase in AI model scale and computing power requirements, transitioning from a "design tool" to the underlying operating system of AI computing systems [1][3]. Group 1: Challenges in EDA - The main challenge for domestic EDA in ensuring the autonomy of AI computing systems lies in the need to consider the entire system rather than just the chip, especially as Moore's Law becomes less applicable [3]. - The evolution of computing architecture from single chips to multi-chiplets and supernodes presents new challenges in maintaining flexibility, scalability, and high-bandwidth interconnects [4]. Group 2: Chiplet Architecture - The Chiplet design requires new verification and collaboration processes, with a two-phase approach: the first phase focuses on integrating computing and storage through a 2.5D structure, while the second phase involves hybrid bonding for 3D stacking [4]. - The integration of various components such as sensors, storage, and RF devices into a compact design necessitates multi-physical field collaborative analysis, which increases the complexity of simulation across different scales [4][6]. Group 3: System-Level Simulation - System-level simulation differs significantly from traditional EDA, as it must account for multi-physical field interactions and the potential for issues arising from high current and impedance fluctuations [5]. - Chiplet architecture offers advantages in system-level considerations, allowing for a more comprehensive approach to design and integration [5]. Group 4: AI Integration in EDA - The strategy of "EDA for AI" focuses on providing comprehensive solutions from chip design to system integration, addressing the challenges posed by AI's increasing computational demands [10]. - The "AI + EDA" strategy aims to integrate AI models into the design and simulation processes, significantly enhancing efficiency and enabling a shift from rule-driven to data-driven design approaches [12]. Group 5: Future Outlook - The future of EDA in the AI era is characterized by cross-scale, cross-physical, and cross-system engineering, with expectations for more domestic design tools to become practical and for system-level issues to be resolved during the simulation phase [14]. - The company is positioned as a key player in this transformation, leveraging advancements in Chiplet technology, supernodes, and AI factories to enhance the stability and power of AI computing infrastructure [14].