证券代码：688387 证券简称：信科移动 公告编号：2025-025 中信科移动通信技术股份有限公司 本公司董事会及全体董事保证本公告内容不存在任何虚假记载、误导性陈述 或者重大遗漏，并对其内容的真实性、准确性和完整性依法承担法律责任。 中信科移动通信技术股份有限公司（以下简称"信科移动"或"公司"）2025 年 6 月 12 日召开了第二届董事会第六次会议及第二届监事会第五次会议，审议 通过了《关于募投项目延期的议案》，同意公司将首次公开发行募集资金投资项 目"5G 无线系统产品升级与技术演进研发项目"、"5G 行业专网与智能应用研发 项目"和"5G 融合天线与新型室分设备研发项目"的实施期限延长至 2026 年 12 月 31 日。上述事项在董事会审批权限范围内，无需提交股东大会审议。保荐机 构申万宏源证券承销保荐有限责任公司（以下简称"申万宏源承销保荐"）对上 述事项出具了明确的核查意见。 本次变更仅涉及募投项目实施期限，未改变募集资金投资用途、投资总额和 实施主体，不会对募投项目的实施造成实质性影响，不存在变相改变募集资金用 途或损害公司和股东利益的情形。 一、募集资金基本情况 根据中国证券监督管理 ...

Core Viewpoint - Ren Zhengfei, the founder of Huawei, recently gave an interview to People's Daily, marking his return to the public eye amid significant geopolitical tensions, particularly the ongoing US-China trade negotiations and the impact on China's chip industry [4][5]. Group 1: Huawei's Response to External Pressures - Huawei is facing a full-chain blockade from the US on its chip industry, which has forced China to rebuild an independent chip supply chain [4]. - In the interview, Ren emphasized a proactive approach to external pressures, stating, "Just do it, step by step" [4]. - Ren acknowledged the exaggeration of Huawei's achievements by the US and expressed that external criticism helps Huawei remain vigilant [4]. Group 2: Research and Development Focus - Huawei's R&D expenditure for 2024 is projected to be 179.7 billion yuan, accounting for approximately 20.8% of its total revenue, with cumulative R&D investment exceeding 1.249 trillion yuan over the past decade [5][6]. - Ren highlighted that around 60 billion yuan of the R&D budget is allocated to fundamental theoretical research, which is not subject to performance evaluation, while about 120 billion yuan is focused on product development [6]. - The company employs a significant number of R&D personnel, with approximately 54.1% of its total workforce dedicated to research and innovation [10]. Group 3: Intellectual Property and Innovation - Huawei leads in global patent applications, with 6,600 applications in 2024, significantly outpacing competitors like Samsung and Qualcomm [10]. - The company has transformed its patent portfolio into a revenue-generating asset, earning 560 million USD from intellectual property in 2023 [10]. - Ren's commitment to innovation is rooted in Huawei's history, where he emphasized the importance of having proprietary technology over merely acting as a reseller [8]. Group 4: Broader Implications for Chinese Technology - The interview reflects a broader trend among Chinese companies prioritizing independent R&D and innovation, which is reshaping global industry dynamics [11][12]. - China's stance in international negotiations, particularly regarding technology and resources, indicates a firm commitment to maintaining its position without concessions [12]. - Ren's assertion that "the era of begging for survival is over" signifies a shift in mindset towards self-reliance and resilience in the face of external challenges [10][12].

Group 1 - Shenzhen has attracted top global talent, with 920 scholars selected as top 2% scientists, leading the nation in R&D personnel [1][6] - The recent policy document emphasizes the integration of education, technology, and talent reform as a priority for Shenzhen's development [1][2] - Shenzhen's R&D expenditure intensity reached 6.46% in 2023, with corporate R&D investment consistently above 90%, peaking at 93.3% [6][7] Group 2 - The new policy aims to create a collaborative network among government, industry, academia, and research, enhancing the role of enterprises in innovation [2][9] - Shenzhen's educational institutions are rapidly evolving, with 17 universities established and a focus on aligning higher education with economic needs [6][11] - The establishment of the Shenzhen Medical Academy, led by renowned scientist Yan Ning, reflects the city's commitment to innovative research and talent cultivation [9][10] Group 3 - The policy encourages a market-oriented approach to research management, allowing for greater autonomy in project selection and funding allocation [9][10] - Young researchers and teams are highlighted as key drivers of innovation, with a significant proportion of research personnel being under 40 years old [8][11] - Shenzhen's experience in integrating education and industry serves as a model for national talent development strategies [11][12]

Core Viewpoint - Nokia is significantly expanding its IP routing portfolio to support utilities transitioning to smart grid technologies, enhancing its 7705 Service Aggregation Router and 7250 Interconnect Router platforms to meet the growing demand for secure and high-performance networking infrastructure [1][3]. Group 1: Industry Context - Utilities globally are implementing smart grid technologies to address challenges such as climate disruptions and cyber threats, while also integrating distributed energy resources to achieve net-zero emissions targets [2]. - Smart grids enhance operational efficiency and real-time monitoring, ensuring compliance with evolving regulations and preparing the grid for future demands [2]. Group 2: Nokia's Technological Advancements - Nokia's new routing solutions provide end-to-end, secure, and adaptable IP routing that scales from the enterprise edge to the data center core, facilitating the evolution of utilities' communications infrastructure for smart grid technologies [3]. - The 7705 SAR and 7250 IXR platforms offer advanced capabilities for application-aware communications, supporting legacy systems and ensuring precise frequency and time synchronization across the grid [3]. - The platforms are designed to counter cybersecurity threats, including those posed by quantum computing, through advanced quantum-safe MACsec encryption [3][4]. - With increasing bandwidth demands from substation CCTV cameras and sensors, the new platforms can scale to 100 GE and 400 GE to support critical applications and future high-capacity services [3]. Group 3: Customer Needs and Commitment - Energy customers require networks that provide not only bandwidth but also resilience against harsh conditions and preparedness for quantum-era threats, which Nokia aims to address with its latest platform enhancements [4].

中兴通讯AIM团 队 投稿 量子位 | 公众号 QbitAI 如何让AI代码补全更懂开发者？ 中兴通讯团队提出了 两个新的评测指标 ，以及 一套仓库级代码语料处理框架 。 按照团队的说法，这套方法论不仅为评测代码大模型提供了新视角，也为提升模型在真实工业场景中的代码补全性能开辟了新路径。 目前在编写代码时，智能补全工具如GitHub Copilot和Cursor等，极大地提升了程序员的开发效率。 然而这些AI工具给出的建议经常"差了点火候"，不完全符合用户预期。 对此团队认为，这实际上暴露了当前代码大模型使用中的两个关键痛点： 这些问题，无疑限制了代码大模型在复杂工业环境，尤其是在ZTE-Code-Copilot （中兴通讯自研的通信领域代码开发助手） 这类专业场景 中的应用潜力。 那么，如何使AI的补全建议更符合开发者的需求呢？ 两个新指标+一套新框架 事实上，开发者们苦恼于AI的"自说自话"久矣。团队认为要解决这个问题，必须回答两个灵魂拷问： ① 何谓高质量的代码补全，能够让开发者愉快地按"Tab"键采纳？ ② 如何教会AI"高瞻远瞩"，理解整个代码仓库的复杂结构和内在逻辑，而不是只盯着眼前的一亩三分地？ ...

训练阶段：类似猛踩油门，但算力、内存、通信如果搭配不当，效率会暴跌； 而数字化风洞在这个过程中起到的作用，就像是一位让AI算力"少踩坑、跑得更快更稳"的智能调度专家。 金磊 发自 凹非寺 量子位 | 公众号 QbitAI 就在刚刚， 华为 首次亮相了一套 "虚" 的技术—— 数字化风洞 ，一个在正式训推复杂AI模型之前，可以在电脑中"彩排"的 虚拟环境平台 。 这套有种《黑客帝国》意味般的技术 （都是通过虚拟世界预演现实） ，是由 华为马尔科夫建模仿真团队 构建，可以 小时级预演万卡集群 方案 。 而之所以要在真枪实弹训推复杂AI模型之前来这么一个步骤，是因为华为研究团队发现，超过60%的算力浪费在硬件资源错配与系统耦合 上。 于是，就像汽车设计师用风洞测试新车性能一样，华为通过这个平台在电脑里模拟AI大模型训练和推理的过程，便提前发现问题并优化配 置。 一言蔽之，为的就是 避免浪费时间和算力 。 更具体来看，如果把运行大模型类比成开一辆高性能赛车，那么当下的痛点就主要集中在三点： 并且是对上述三大痛点逐一击破的那种。 Sim2Train：小时级自动寻优 训练大模型这件事可以说是越来越复杂，例如由于参数量越发 ...

想象一下，你正在用手机导航规划长途路线，背后可能有几十个 AI 模型同时在分析路况、预测拥 堵；医院用 AI 辅助诊断癌症时，系统需要瞬间处理成百上千张 CT 影像。这些看似简单的智能应 用，背后都依赖着像 " 超级大脑 " 一样的 AI 算力集群在 24 小时不停运转。 超节点需要做大超节点规模，才能充分发挥超平面网络的优势，目前业界没有使用光链路来构建超节 点的成功案例，因此华为团队提出了相应的超节点光链路软件容错方案。通过多层防护体系，借助超 时代答，绿色通道等关键技术实现无超节点级故障，通过链路级重传，光模块动态升降 Lane ， HCCL 算子重执行，借轨通信，双层路由收敛， Step 级重调度等特性，实现光模块闪断的故障率容 忍度 >99% 。在新增 10 倍 + 光模块后，通过软件可靠性措施，以及光链路压测技术等，实现光模 块闪断率低至电链路水平，保障了超平面的可靠性。通过构建 Step 级重调度能力，高频的 HBM 多 比特 ECC 故障恢复时间 缩短至 1min ，对于因为故障造成的用户的算力 损失下降 5% 。 高可用助力业务——万卡集群上千亿模型的线性度和训推快恢 线性度指标用于衡量训练 ...

秒级快恢、超95%线性度，华为如何让算力集群高效稳定工作？ 编辑丨李希 01 引言 想象一下，你正在用手机导航规划长途路线，背后可能有几十个 AI 模型同时在分析路况、预测拥堵；医院用 AI 辅助诊断癌症时，系统需要瞬间处理成百上千张 CT 影像。这些看似简单的智能应用，背后都依赖着像 "超 级大脑" 一样的 AI 算力集群在 24 小时不停运转。 高可用核心基础 ——面向超节点的故障感知、管理及容错 AI大集群问题定位复杂，系统规模大、软硬技术栈复杂、调用链长，先要跨域故障定界，然后各域内部故障定 界定位，故障诊断面临巨大挑战；当前定位时间从数小时到数天，技能要求高 ，难以找到故障设备和根因。 华为团队为了让集群运维工具能够快速找到问题原因，有效提升现网问题的闭环效率，提出了 全栈可观测能 力，构建了大规模集群的故障感知能力，主要由集群运行视图、告警视图、网络链路监控、告警接入和配置、 网络流可观测能力组成；同时还提出了包括全栈故障模式库、跨域故障诊断、计算节点故障诊断、网络故障诊 断等四大能力的故障诊断技术。 当前行业水平下，万卡级别的 AI集群平均每天会出现一次甚至多次故障，这不仅严重影响了训练效率，还 ...

Core Insights - VIAVI Solutions Inc. has entered into a Memorandum of Understanding with Hanyang University to collaborate on AI-RAN, 5G, and 6G research at the university's Beyond-G Global Innovation Center [1][2] - The partnership aims to advance next-generation communication technologies and foster talent in this field [5][6] Collaboration Details - VIAVI will provide wireless lab test solutions and expertise, while Hanyang University will become a VIAVI 6G Forward academic partner [1][4] - The Beyond-G Global Innovation Center was selected for the Global Innovation Research Center Support Project, receiving over KRW 5 billion annually for 10 years to support top research initiatives [3] Technological Contributions - VIAVI will contribute its NITRO® Wireless test and optimization suite, which includes a 6G testbed used by leading network equipment manufacturers and research institutions [4] - The technology will enable validation of AI-RAN, 5G, 6G, and quantum technologies, facilitating the planning and implementation of network changes [4][6] Educational Impact - The partnership will provide students with hands-on experience in cutting-edge test environments, laying the foundation for impactful research in 6G and AI-RAN [5]

金磊 发自 凹非寺 量子位 | 公众号 QbitAI 大模型的落地能力，核心在于性能的稳定输出，而性能稳定的底层支撑，是强大的算力集群。其中，构建万卡级算力集群，已成为全球公认 的顶尖技术挑战。 但是，在 华为 ，昇腾万卡算力集群，已经可以做到近乎 "永不罢工" 了： 或许有小伙伴要问了：AI算力需要如此24小时不间断的运作吗？ 关于这一切背后的秘密，华为在今天首次把技术给公开了出来。 高可用的核心：三套"智能保险系统" 答案是肯定的，需要，且有必要。 因为小到我们用手机导个航，背后都会有几十个AI模型在发力来分析路况、预测拥堵；再如医院用AI辅助诊断癌症，系统得在瞬间处理掉成 百上千的CT照片。 这些看似简单的智能应用，其实都离不开如 "超级大脑" 般的AI算力集群，需要它们全天候不停歇地运转着。 而要保证有这样的能力，高训练可用度、高线性度、快速消除故障，就相当于给AI发动机上了一份强有力的保险。 更严格来说，AI推理的可用度甚至还需要 达到99.95%的程度。 那么华为又是如何做到这点的？ AI大集群出问题时，定位故障特别麻烦；毕竟系统规模庞大，软件和硬件组成的技术栈错综复杂，而且调用链条还很长。 要解 ...