News
后摩智能亮出2D CIM+3D CIM路线!Agent竞速赛打响,双路径抢占端边算力先机
3 min read
Source: zhidx.com
Core East-West (Public account: aichip001) Author | Cheng Qian Editor | Mo Ying Core East-West reported on October 16 that on October 14, at the first Ecological Partner Conference of Houmo Intelligence, Houmo Intelligence officially announced that it will use SRAM-CIM as the root technology and lay out two technical paths of 2D CIM and 3D CIM, corresponding to the demand for greater capacity and greater bandwidth. This action hits a key turning point in the evolution of edge AI. Nowadays, the capabilities of large models combined with AI Agents are fully exploded, and personal agents that are highly compatible with end-side scenarios are ushering in an explosive period. Overseas, Meta, OpenAI, and Google have launched personal AI assistants one after another. Domestic major manufacturers such as Tencent, Ant, and Byte have successively reported news of internal testing of personal Agent products. The intensive layout of major manufacturers confirms the huge potential contained in this market. This type of personal agent requires long-term memory and autonomous execution of tasks, and is in line with the advantages of localization, personalization, real-time, and security of end-side AI. Under the surface of a prosperous industry, the competition for computing power between devices and edges has intensified, and pain points such as computing power, memory access bandwidth, power consumption costs, and scenario adaptation have become increasingly prominent. These hardware bottlenecks cannot be eliminated by algorithm optimization, forcing the industry to find new chip architecture ideas. According to Xin Xiaoxu, co-founder and vice president of product of Houmo Intelligence, the new computing paradigm is reshaping the entire industry chain, and its impact may be greater and more far-reaching than everyone thinks. Whoever chooses the right technical path and can lay out the technology in advance is likely to become the winner in this new era. Nowadays, the differentiated technology route represented by the integration of storage and computing is reshaping the pattern of end-side AI computing power, becoming the core key to unlocking the large-scale implementation of large-scale end-side and end-side models and supporting the explosion of the AI Agent industry. The dual technology route of Houmo Intelligence this time is an important appearance in this critical period of underlying change. 1. End-to-end AI has reached a commercialization singularity. How can chip manufacturers break through? Device-side AI has become an irreversible core direction for industrial development. Local devices carry model inference, and there is no need to continuously upload raw data and conversation context to the cloud. This not only reduces token overhead and compresses business costs, but also