随着数字化的发展，嵌入式系统已成为支撑智能设备运转的核心。从可穿戴设备的实时健康监测到自动驾驶汽车的环境感知，从工业机器人的精密控制到 5G 基站的高速数据处理，这些场景对内存的性能、功耗和成本提出了日益严苛的要求。传统存储技术逐渐显露出短板：静态随机存取存储器（SRAM） 虽能提供高速访问，但高昂的成本和有限的容量使其难以满足大规模数据处理需求；动态随机存取存储器（DRAM）虽具备较高的存储密度，却需要复杂 的刷新电路支持，功耗控制难题始终制约着其在嵌入式领域的深度应用。 在此情况下，伪静态随机存取存储器（PSRAM）凭借独特的技术架构脱颖而出，成为嵌入式内存领域的新选择。这种融合了 SRAM 接口便利性与 DRAM 存储密度优势的创新产品，正在重塑嵌入式存储的市场格局。 01 PSRAM 是什么？ PSRAM（Pseudo Static Random Access Memory，伪静态随机存储器）是面向嵌入系统、消费电子、IoT物联网、可穿戴设备、端侧AI产品打造的新型存储 方案。顾名思义，其是一种通过技术手段模拟SRAM操作特性的DRAM。 要理解PSRAM的优势，我们需要先了解传统存储器的特点。SRA ...

Core Viewpoint - The Chinese semiconductor industry has undergone significant development from 2000 to 2021, driven by the collaboration of national, corporate, and financial forces, culminating in a complex landscape shaped by globalization and subsequent challenges from geopolitical tensions [2][57]. Group 1: Historical Development - From 2000 to 2012, the Chinese semiconductor industry entered its "root and sprout" phase, establishing an initial industrial chain [1]. - The year 2014 marked a pivotal moment with the establishment of the National Integrated Circuit Industry Investment Fund, which provided essential financial support for the industry [4][6]. - Between 2014 and 2017, this financial influx accelerated the growth of Chinese semiconductor companies through mergers, acquisitions, and investments, integrating them into the global supply chain [1][5]. Group 2: Financial Forces - The emergence of various financial institutions, such as Zhilu Capital and Wuyuefeng Capital, provided market-oriented investment capabilities, enhancing the ability to acquire overseas semiconductor firms [8][10]. - Notable acquisitions included the purchase of Rui Neng Semiconductor for 800 million RMB, which significantly advanced China's position in the power semiconductor sector [9]. - The establishment of the Zhongguancun Rongxin Industrial Alliance in 2015 facilitated collaboration among various investment institutions, enhancing the success rate of overseas acquisitions [22]. Group 3: Corporate Forces - Major corporations like Unisoc and Changjiang Electronics have expanded their influence through strategic acquisitions, such as Unisoc's purchase of RDA Microelectronics for 910 million USD, enhancing its capabilities in mobile communication chips [13][31]. - The collaboration between large enterprises and national funds has enabled significant mergers, such as Changjiang Electronics' acquisition of STATS ChipPAC for 780 million USD, positioning it among the top semiconductor packaging and testing companies globally [11][13]. - The rapid growth of these companies has contributed to the establishment of a competitive landscape in the semiconductor industry, with firms like Unisoc becoming a leading player in the global market [13][31]. Group 4: Challenges and Adjustments - The period from 2018 to 2020 saw increasing challenges due to U.S. sanctions and trade tensions, which hindered the ability of Chinese companies to pursue overseas acquisitions [39][46]. - The establishment of the second phase of the National Integrated Circuit Industry Investment Fund in 2019 aimed to strengthen domestic capabilities and support key sectors like IC design and AI [47][48]. - By 2021, the focus shifted towards restructuring and preparing for a new phase of development, as the global landscape for semiconductor investments became increasingly restrictive [56][54]. Group 5: Future Outlook - The Chinese semiconductor industry is entering a phase of self-reliance and independence, with a focus on enhancing domestic capabilities and reducing reliance on foreign technology [57]. - The collaboration among national, corporate, and financial forces is expected to continue driving innovation and growth in the sector, despite external pressures [57].

Doug Schirle, CFO Didier Lasserre, VP of Sales and Investor Relations Spring 2025 Safe Harbor The statements contained in this presentation that are not purely historical are forward-looking statements within the meaning of Section 21E of the Securities Exchange Act of 1934, as amended, including statements regarding GSI Technology's expectations, beliefs, intentions, or strategies regarding the future. All forward- looking statements included in this press release are based upon information available to GS ...

报告导读： GPT-5 将于 2025 年夏季发布，将整合现存模型功能。 Marvell 发布 2nm SRAM ，并上调市场预期。 MiniMax 相继推出推理模型、视频生成模型、通用智能体。 投资建议： 维持行业增持评级，推荐 AI 算力方向、云厂商方向、 AI 应用方向、 AI 社交方向以及 AI 注 入周期下的中概巨头。 GPT-5 将于 2025 年夏季发布，将整合现存模型功能。 当地时间 6 月 18 日， OpenAI 首席执行官 Sam Altman 在公司首期博客节目中探讨了通用人工智能（ AGI ）、 " 星际之门 " 计划及用户隐私等议 题。他在节目中宣布 GPT-5 预计于 2025 年夏季发布，具体日期尚未披露。该模型计划整合 OpenAI 全部旗舰功能，包括 GPT-4o 的自然语言处理能力与 o3 在代码及科学推理领域的优势，形成综合性能 更强大的新模型。 Altman 同时表示不排斥在 ChatGPT 中投放广告，此举或将成为公司拓展盈利的新渠 道。 Marvell 发布 2nm SRAM ，并上调市场预期。 在 6 月 18 日结束的 Custom AI 投资者日活动上， M ...

智通财经APP获悉，迈威尔科技(MRVL.US)在上周的"定制人工智能(AI)投资者日"活动上给人留下了深刻印象。从强大的定制芯片到业界首创的2nm SRAM，再到与亚马逊和微软的大规模合作，迈威尔科技为未来价值数十亿美元的定制芯片业务奠定了基础。 过去一个月，迈威尔科技的股价上涨了20%，但年初至今仍累计下跌33%。Seeking Alpha撰稿人Muslim Farooque表示，考虑到迈威尔科技non-GAAP远 期市盈率为26倍，该股可能是最物有所值的人工智能股票。 迈威尔科技的定制人工智能投资者日取得了巨大成功。 该公司的定制ASIC业务发展迅猛。该公司管理层称，为云巨头开发的定制人工智能芯片在2025财年带来了6.5亿美元的销售额，占总销售额的11%，占 数据中心芯片销售额的25%以上。预计这一比例将跃升至50%以上，使迈威尔科技的定制人工智能业务成为价值数十亿美元的增长引擎。 在内存方面，该公司推出了业界首款2nm定制SRAM，封装了6 GB的超高速片上内存，使待机功耗降低了三分之二，芯片面积减少了15%。 换句话说，该公司可以在每片芯片中塞入更多内存，同时降低能耗并提高数据吞吐量。正如迈威尔科 ...

如果您希望可以时常见面，欢迎标星收藏哦~ 来源：内容 编译自 semiengineering 。 随着人工智能 (AI) 工作负载日益庞大、复杂，用于处理所有数据的各种处理元件对功率的需求也 空前高涨。然而，高效可靠地提供这种功率，同时又不损害信号完整性或引入热瓶颈，却带来了半 导体历史上最严峻的设计和制造挑战。 与通用处理器不同，专为 AI 工作负载设计的芯片将密度推向极致。它们将更多晶体管封装到更小 的空间内，同时增加晶体管的总数（通常以芯片的形式）。其结果是更大、更密集的系统级封装， 其中电力传输不仅仅是一个电气问题，而是一个从单个芯片到服务器机架的封装、材料和系统集成 挑战。 由于内存和计算单元之间来回传输的数据量巨大，动态功耗占据主导地位。这些传输跨越庞大的内 存层级结构，使用各种高速互连。但移动所有这些数据是有代价的，这会产生层层叠加的设计约 束，从内存层级结构决策一直延伸到电源传输网络 (PDN)。 Imec研发副总裁Julien Ryckaert表示："随着我们转向背面和3D堆叠，热量变得更加局部化，也更 难消散。这种物理压缩加剧了电迁移和局部热点等挑战。" 为了使这些级别的电源传输易于处理， ...

行情速递 ① 主 板 领 涨 ， 大 为 股 份 (+10.02%)/ 深 华 发 A (+110.01%)/ 中 京 电 子 (+10.01%)/ 沪 电 股 份 (+10.00%)/楚天龙 (+10.00%)/ ② 创业板领涨， 逸豪新材(+20.01%)/科翔股份(+19.98%)/联建光电(+19.90%)/ ③科创板领涨， 清越科技(+20.05%)/生益电子 (+16.31%)/源杰科技(+9.85%)/ ④活跃子行业， SW印制电路板(+5.82%)/SW品牌消费电子(+3.94%)/ 国内新闻 ① 艾邦半导体网，中电化合物与甬江实验室正式签署战略合作框架协议/双 方将围绕AR眼镜用光学碳化硅（SiC）晶片展开深度研发合作/共同推动SiC 材料在增强现实（AR）领域的创新应用/为下一代智能穿戴设备提供更优的 光学解决方案/ ② 电子工程专辑，阿里巴巴Qwen团队宣布Qwen3全系模型完成苹果MLX框 架的官方适配/涵盖从0.6B（6亿参数）到235B（2350亿参数）的8个尺寸/ 并支持4bit/6bit/8bit/BF16四种量化精度/总计32个版本模型/ ③ 集微网，据盛美上海最新机构调研信 ...

Summary of GSI (GSIT) Conference Call - May 21, 2025 Company Overview - GSI was founded 30 years ago by Leland Shu, the current president and CEO, and went public in February 2007 [2][3] - The company has a market cap of just under $100 million and maintains a significant insider ownership of 27% [6] Financials - GSI finished fiscal year 2025 with revenues of $20.5 million [4] - The company has $13.4 million in cash and cash equivalents and has never carried debt [5] - Operating costs have decreased to $5.6 million in the past quarter, with a cash burn of just over $1.5 million last quarter [40] - Revenues have been growing, attributed mainly to the build-out of AI, with a previous revenue of approximately $4.5 million a year ago [37] Product Lines and Market Focus - GSI specializes in high-density, high-performance memory products, particularly in the SRAM area, which has been profitable and is experiencing growth from the Sigma Quad family [6][7] - The company is expanding into aerospace and AI markets, specifically targeting edge and inference applications with their AI chip, the APU [7][8] - The APU is described as a unique compute-in-memory architecture, allowing for extreme parallel processing with 2 million bit processors [12][14] Technology and Innovation - GSI's APU technology is positioned as a true compute-in-memory solution, contrasting with other companies that offer near-memory processing [12][61] - The APU architecture allows for significant power savings by eliminating the need to transfer data back and forth between memory and processing units [15][16] - The company has two product families: Gemini One and Gemini Two, with Gemini Two expected to be production-worthy soon [24][44] - The upcoming PLATO chip aims to address multimodal generative AI and large language models at the edge, targeting a power consumption of under 10 watts [26][34] Market Opportunities - The AI industry is projected to grow at over 20% CAGR, while the space market is expected to grow at just under 10% CAGR [8] - GSI aims to capture 10-20% of a $100 million market opportunity in radiation-hardened devices for the space industry [10][11] - The company is actively pursuing government funding through SBIRs, having won three grants totaling approximately $3.35 million [41][42] Strategic Initiatives - GSI is focusing on short-term sales with government and military sectors while maintaining discussions with hyperscalers for long-term growth [49] - The company is looking to raise funds for the development of PLATO and the launch of Gemini Two, with a timeline for funding within the next two to three quarters [57][58] - GSI is open to various funding avenues, including equity raises, partnerships, and potential mergers or acquisitions [47] Challenges and Future Outlook - The company acknowledges the need to demonstrate the value of its technology to the market, as current stock prices do not reflect its potential [59] - GSI is optimistic about the future, expecting milestones to kick in during the second half of the year, particularly with the launch of Gemini Two and the development of PLATO [64]

Core Insights - GSI Technology, Inc. (GSIT) shares have increased by 8.1% since the earnings report for the quarter ended March 31, 2025, outperforming the S&P 500 index, which grew by 0.7% during the same period [1] - The stock has seen a significant rise of 47.5% over the past month compared to the S&P 500's 2.8% increase [1] Financial Performance - For Q4 fiscal 2025, GSI Technology reported a net loss of $0.09 per share, an improvement from a net loss of $0.17 per share in the same quarter last year [2] - Net revenues reached $5.9 million, reflecting a 14% year-over-year increase, while gross margin improved to 56.1% from 51.6% a year earlier [2] - The net loss narrowed to $2.2 million from $4.3 million in the previous year [2] Operating Metrics - Operating expenses decreased to $5.6 million from $7.2 million year-over-year, primarily due to a reduction in R&D costs, which fell to $3 million from $4.8 million [3] - Government funding under the SBIR program contributed $0.9 million to offset R&D expenses [3] - SG&A expenses increased slightly to $2.6 million from $2.4 million [3] Revenue Breakdown - Sales to KYEC surged to $1.7 million, representing 29.5% of quarterly revenues, up from $0.5 million (10.6%) in the prior year [4] - Sales to Nokia declined to $0.4 million, or 7.5% of revenues, down from 13.5% a year earlier [4] - Military and defense sales accounted for 30.7% of quarterly shipments, slightly down from 35.5% last year [4] - SigmaQuad sales remained stable at 39.3% of shipments [4] Management Insights - The CEO highlighted strong demand for legacy SRAM products as a key driver of revenue growth and noted a strategic breakthrough with a North American prime contractor for higher-margin radiation-hardened SRAM chips [5] - Additional orders are expected in fiscal 2026, which is seen as pivotal for market access [5] Product Development - Progress continues in the AI product segment with the Gemini-II APU, with production-ready chips expected by the end of Q1 fiscal 2026 [6] - Successful milestones have been achieved under SBIR contracts with the U.S. Space Development Agency and Air Force Research Laboratory [6] - A new enhancement to the Plato processor includes a built-in camera interface, positioning it as a compact solution for edge AI applications [7] Earnings Drivers - Improvements in both top-line and bottom-line performance were attributed to a favorable product mix and cost discipline [8] - Gross margin benefited from increased sales of higher-margin products and better absorption of fixed manufacturing costs [8] - R&D expenses for the full year dropped to $16 million from $21.7 million in fiscal 2024 [9] Fiscal Year Overview - For the full fiscal year, revenues totaled $20.5 million, a decrease of 5.7% from fiscal 2024, while the net loss narrowed to $10.6 million, or $0.42 per share, from $20.1 million, or $0.80 per share a year ago [11] Future Guidance - For Q1 fiscal 2026, GSI expects revenues between $5.5 million and $6.3 million, with gross margin projected at 56% to 58% [12] - Continued momentum in SRAM demand and potential Gemini-II commercialization milestones are expected to support near-term growth [12] Other Developments - The company completed the sale and leaseback of its Sunnyvale headquarters, resulting in a $5.8 million gain and a reduction in working capital [13] - Working capital declined to $16.4 million from $24.7 million a year earlier, with cash and cash equivalents reported at $13.4 million and stockholders' equity at $28.2 million as of March 31, 2025 [13]

如果您希望可以时常见面，欢迎标星收藏哦~ 来源：本文编译自semiengineering，谢谢。 混合方法有所帮助，堆叠更多层HBM也同样有效。三星、SK 海力士和美光是仅有的几家生产 HBM 的公司。三星以此为跳板，开始针对特定工作负载定制 HBM。但最佳解决方案是 HBM 和 SRAM 的结合，而代工厂最新的路线图展示了不同存储器的复杂组合，其互连间距非常紧密，以 促进数据移动。 英特尔最新的架构显示，14A 逻辑层直接堆叠在 SRAM 层上方。 英特尔代工厂、台积电和三星代工厂正在争相提供完整 3D-IC 的所有基础组件，这些组件将在未 来几年内以最小的功耗实现性能的大幅提升。 人们将大量注意力集中在工艺节点的进步上，但成功的3D-IC实现远比仅仅扩展数字逻辑复杂得 多，也更全面。它需要新材料，以及处理更薄基板并将其组装的不同方法。它涉及不同的背面供电 方案、各种类型的桥接器、多芯片通信的接口标准以及新的互连技术和方法。此外，它还需要对 EDA 工具和方法、数字孪生、多物理场仿真进行重大变革，以及重组工程团队和流程，并在从设 计到制造的整个流程的多个阶段融入人工智能。 3D-IC 已出现在代工厂的内部路 ...