英特尔在半导体工厂（晶圆厂）核心设备管理中引入智能体AI，实测结果显示：设备异常征兆的检出速度较此前提升30%，仅单一场景每年就可节约约750万美元成本。据报道，英特尔近8个月以智能体AI平台Seeq作为协调器，部署多组AI智能体，取得了上述成效。
Seeq平台采用多AI分工协作模式，由一个“协调器”统一调度各个AI智能体。一旦监测到异常信号，系统可自动完成原因分析、生成作业指令初稿，最终由工程师负责审批确认。核心价值在于大幅省去工程师以往在多个系统间来回切换、人工排查的流程。
双方的合作始于约3年前，当时英特尔正在寻找可大规模开发、部署分析能力的自助式分析平台。随着英特尔全面推进AI落地，合作现已拓展至智能体AI方向。
落地对象为超纯水设备、冷水机组、特种气体、化学品供给装置等支撑晶圆厂运转的核心公用工程设备。这类设备虽不直接参与晶圆制造，但只要其中一台出现故障，极易造成晶圆污染或是产线停摆。尤其是超纯水系统，哪怕极微量杂质都不被允许，快速捕捉异常直接关系良率水平。
过去的模式是警报触发后，工程师再打开多个界面调取数据、排查根因；现在系统捕捉到异常信号后，会自动汇总关联数据、历史同类案例、推荐处置方案并推送给工程师。工程师只需核验内容并审批，即可生成作业工单。
带来的效益不止异常识别提速。自动生成工单还将技术人员的工作量降低15%~20%，每年节省超12万工时；冷水厂的能源效率提升约15%。
Seeq已接入英特尔全球约70处厂区的系统，整合6500万至7000万条标签（时序信号数据），可直接调用。目前每周约有1000名英特尔工程师在使用该软件。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
The introduction of AI agent AI in the management of core equipment at the semiconductor plant (the crystal circle plant) by Intel has shown that the speed at which equipment signs of anomalies are detected is 30 per cent higher than before, and that a single scene would save approximately $7.5 million per year. This was reported to have been achieved in Intel for nearly eight months with the AI agent AI platform Seeq as a coordinator for the deployment of multiple groups of AIAI agent.
The Seeq platform uses a multi-AI division of labour collaboration model, with a “coordinater” to move the AZAI agents. Once the anomaly signals are detected, the system automatically completes the cause analysis, generates the first draft of the operational instructions and is ultimately certified by the engineer. The core value lies in the significant savings in the process by which engineers used to switch back and forth between multiple systems and manually screen them.
Cooperation between the two sides began some three years ago when Intel was looking for a self-help analysis platform that could be developed and deployed for analysis on a large scale. With Intel's full push to land AI, cooperation has now expanded to AI agent AI.
Landing targets are core utility engineering equipment that supports the operation of the Clip plant, such as super-purified water equipment, chillers, special gases, and chemical supply units. Such equipment, although not directly involved in the manufacture of the crystal circle, is highly susceptible to the contamination of the crystal circle or the shutdown of the production line if one of them fails. In particular, super-purified water systems, even minimal impurities, are not allowed to quickly capture levels of abnormal direct correlation.
In the past, the pattern was for engineers to open multiple interfaces to retrieve data and sequence root causes after the alarm had been triggered; now that the system captures anomalous signals, it automatically aggregates associated data, historical cases, recommends disposal options and sends them to the engineer. Engineers can generate worksheets only by checking and approving the contents.
The benefits go beyond anomalous recognition of speed. Auto-generated worksheets also reduce the workload of technicians by 15-20 per cent and save more than 120,000 working hours per year; energy efficiency improvements in cold water plants are about 15 per cent.
Seeq has access to a system of about 70 plant areas worldwide in Intel that integrates between 65 and 70 million labels (time series signal data) and can be called directly. Currently, about 1,000 Intel engineers use the software every week.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.