摘要
人工智能与芯片技术正形成深刻的双向驱动关系,其影响主要体现在“用AI设计芯片(AI for Chip)”和“为AI设计芯片(Chip for AI)”这2个核心维度。前者利用AI技术赋能芯片设计流程,通过机器学习算法,AI能高效完成处理器微架构参数的智能探索、物理设计的自动化布局布线(PPA优化)以及软硬件协同调优,极大提升了复杂芯片的设计效率与质量。后者指为加速AI计算而专门设计芯片架构,如今英伟达凭借GPU+CUDA生态成为市场主导,而基于开放指令集RISC-V扩展AI指令(如向量、张量扩展)有望成为打破垄断的新趋势。展望未来,AI技术与芯片产业将构成了一个强大的创新循环:先进的AI工具设计出更强大的专用芯片,而这些芯片又为下一代AI技术的发展提供了至关重要的算力基石。
Artificial intelligence and chip technology are engaged in a deep,mutually reinforcing evolution,crystallized in two core directions:AI for Chip and Chip for AI.AI for Chip leverages AI to transform semiconductor design—enabling intelligent microarchitectural exploration,automated PPA-optimized physical implementation,and hardware-software co-design,significantly enhancing design efficiency and quality.Chip for AI focuses on architecting specialized hardware to accelerate AI workloads.While NVIDIA dominates through its GPU-CUDA ecosystem,extending open instruction sets like RISC-V with vector and tensor capabilities is emerging as a key path toward architectural innovation and ecosystem diversification.Looking ahead,a powerful virtuous cycle is taking shape:advanced AI tools enable more capable domain-specific chips,which in turn provide the essential computational foundation for next-generation AI advancements.
作者
包云岗
BAO Yungang(Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China)
出处
《微电子学与计算机》
2025年第10期1-8,共8页
Microelectronics & Computer
关键词
人工智能
芯片
微控制处理器
大语言模型
artifical intelligence
chip
microcon-troller unit
large language model
