人工智能，已然成为推动全球产业变革与技术创新的核心引擎。从智能生活的细微渗透到传 统产业的深度赋能，AI的触角无处不在、势不可挡。在这个AI飞速演进的时代，每一颗芯片 都值得被重新定义；而构建AI的算力基础设施，则为技术落地与产业升级注入了无限潜力。 2025年4月15日，慕尼黑电子展期间，由 半导体行业观察与慕尼黑电子展 联合主办的 "2025 AI技术创新论坛" 隆重举行。论坛汇聚了 概伦电子、兆易创新、速显微、英飞凌、得一微、 达摩院、万国半导体、光羽芯辰、Imagination Technologies、上海百图、德州仪器等 行业 前沿企业与重量级嘉宾，共们围绕AI趋势展开深入解读，现场干货不断，观点碰撞激烈。 嘉宾演讲精彩纷呈，AI产业全面加速 上海概伦电子副总裁马玉涛发表了主题为 "AI时代模拟和定制电路设计的挑战和机遇" 的开幕演 讲，系统阐释AI技术在电子设计自动化（EDA）领域的创新实践与产业突破。他指出，当前 EDA 工具在精度与速度上，面临双重挑战。 概伦电子在AI技术上探索已久，覆盖芯片制造到设计优化 全流程，贯穿成熟工艺与先进工艺，已部署和规划多维度AI/ML融合解决方案。概伦电子从 ...

Core Viewpoint - Artificial intelligence (AI) is becoming the core engine driving global industrial transformation and technological innovation, with its influence permeating various sectors and redefining chip technology and AI infrastructure [1]. Group 1: AI Industry Acceleration - The AI industry is experiencing comprehensive acceleration, with significant advancements in various sectors [2]. - The "2025 AI Technology Innovation Forum" gathered leading companies and experts to discuss AI trends and innovations [1]. Group 2: EDA Innovations - Shanghai Gaoneng Electronics' Vice President, Ma Yutao, highlighted the challenges and opportunities in analog and custom circuit design in the AI era, emphasizing the dual challenges of precision and speed in EDA tools [3]. - Gaoneng Electronics is focusing on AI/ML integration solutions across the entire chip manufacturing and design optimization process, aiming to transform the semiconductor industry into a data-driven paradigm [3]. Group 3: Flash Memory Demand - Tsinghua Unigroup's market manager, Tian Yue, discussed the rapid growth of the AI server market, predicting it will exceed $233 billion by 2028, with each AI server requiring approximately $100 worth of Flash memory [5]. - Tsinghua Unigroup holds the second-largest market share in SPI NOR Flash, with cumulative shipments exceeding 27 billion units [5]. Group 4: GPGPU Opportunities - Dr. Xiang Tian from Suxian Microelectronics presented on the DeepSeek technology, which significantly reduces computational load during inference, facilitating the deployment of large models on edge devices [8]. - Suxian Microelectronics' "Tianyuan" GPU architecture supports high concurrency and integrates open-source ecosystems, providing comprehensive solutions for AI product deployment [8]. Group 5: AI Power Solutions - Infineon's market manager, Zhou Chengjun, emphasized the need for efficient power applications in AI systems, introducing integrated power modules that reduce power loss to 2% compared to traditional methods [11]. - Infineon is recognized as a global leader in AI power management due to its advanced packaging technology and manufacturing capabilities [11]. Group 6: AI Solutions for SMEs - Zhang Haonan from DeYi Microelectronics discussed the challenges faced by SMEs in deploying AI models and introduced an integrated AI training and inference solution that significantly reduces costs and technical barriers [14]. - The solution supports local processing of large models, offering features like breakpoint training and flexible parameter configuration [14]. Group 7: RISC-V Development - Alibaba's DAMO Academy's Li Jue highlighted the rapid growth of RISC-V, which has surpassed 40% annual growth in mainstream markets, and its potential in high-performance computing and AI acceleration [17]. - The DAMO Academy is iterating on the Xuantie series processors to enhance capabilities for AI applications [17]. Group 8: AI Power Management - AOS Semiconductor's Liu Song discussed the increasing power demands of AI servers and introduced innovative power MOSFET solutions to meet these challenges [19]. - AOS's products are designed to optimize power efficiency and reliability in high-frequency applications [19]. Group 9: Edge AI Trends - Yang Lei from Guangyu Xincheng emphasized the rise of edge AI model chips, which are crucial for the intelligent upgrade of various industries, highlighting the commercial opportunities for hardware companies [21]. - Edge AI offers advantages in real-time processing, reliability, and privacy protection, although challenges remain in performance and cost [21]. Group 10: AI Hardware Evolution - Imagination's Huang Yin discussed the evolution of AI models and the importance of balancing performance with storage and communication needs in edge AI applications [23]. - The future demand for AI hardware will focus on efficiency, integration, and flexibility, necessitating collaboration across the industry [23]. Group 11: High-Performance Thermal Solutions - Lu Cheng from Baidu Technology discussed the increasing demand for high-performance thermal materials, emphasizing the need for customized solutions in various applications [25]. - Current leading thermal materials include aluminum oxide and boron nitride, which meet diverse thermal conductivity requirements [25]. Group 12: Real-Time Fault Monitoring - Wu Jixuan from Texas Instruments highlighted the advantages of edge AI in real-time fault detection systems, showcasing the effectiveness of integrated NPU architectures [28]. - The edge AI solutions provide faster response times and enhanced privacy compared to cloud-based systems [28]. Group 13: Future of AI Chips - A panel discussion featuring Yang Lei, Xiang Tian, and Huang Yin focused on the future design logic and ecosystem collaboration for AI chips, emphasizing the shift from general computing to diverse, modular, and low-power designs [30]. Conclusion - The successful hosting of the "2025 AI Technology Innovation Forum" showcased significant advancements in AI technology across various dimensions, highlighting the importance of industry collaboration and ecosystem synergy [32].