AI增强型SoC

Search documents
AI系列专题报告(三):AIot端侧:智能硬件百花齐放,国产SoC大有可为
Ping An Securities· 2025-06-19 11:09
Investment Rating - Semiconductor industry maintains a strong market rating [1] Core Insights - The report highlights the rapid penetration of AI applications at the edge, driven by technological innovations that promote hardware upgrades [5][9] - The integration of AI-enhanced SoCs with NPU and reconfigurable computing units is breaking traditional processor efficiency bottlenecks, enabling real-time inference and decision-making capabilities at the edge [2][9] - Audio is emerging as a primary information dimension for AI applications, with AI-enhanced audio models driving product innovation [10][15] Summary by Sections Processing: Edge Intelligence Drives NPU Adoption - The widespread application of NPUs is being propelled by edge intelligence, with low power consumption making them ideal for edge devices [2][32] - NPU architecture is evolving alongside AI algorithms and application scenarios, enhancing performance in tasks like facial recognition and voice recognition [32][41] Connectivity: Wireless Communication as the Main IoT Implementation Method - The demand for edge connectivity in IoT is growing, leading to the expansion of local wireless connection technologies such as WiFi, Bluetooth, and ZigBee [2][73] - The integration of various wireless technologies enhances device functionality and application scenarios, allowing for flexible communication methods based on specific needs [73][74] Edge Applications: AI's Core is Upgrading User Experience - The shift of AI development focus towards end devices is creating a strong demand for intelligent perception and natural dialogue across multiple modalities [2][9] - AI technologies are being integrated into devices like smart glasses, headphones, and smart speakers, significantly enhancing user experience [9][10] Investment Recommendations - The report suggests focusing on companies like Rockchip, Allwinner Technology, Hengxuan Technology, and others, as the demand for edge AI computing in low-power devices is expected to rise significantly [2][9]