Author page description
May 30
- Partners Ask, SK hynix Answers at 2025 Shared Growth Committee General MeetingSK hynix hosted the 2025 Shared Growth Committee General Meeting on April 25 at Grand Walkerhill Seoul, sharing with partner companies the committee’s operational plans for the year and the company’s latest business updates. Established by SK hynix in 2001 to strengthen mutual growth and cooperation with the company’s partners, the Shared Growth Committee holds […] The post Partners Ask, SK hynix Answers at 2025 Shared Growth Committee General Meeting first appeared on SK hynix Newsroom .
May 29
- SK hynix’s Choungki Song Awarded Bronze Tower Order of Industrial Service Merit for Registering Over 300 Patents on Next-Gen MemoryTechnical Leader Choungki Song of the HBM Development team received the Bronze Tower Order of Industrial Service Merit at the 60th Invention Day ceremony held on May 19 at Coex Magok in Seoul. To commemorate the 60th anniversary of Invention Day, the Korean Intellectual Property Office (KIPO) hosted a ceremony to present government awards to […] The post SK hynix’s Choungki Song Awarded Bronze Tower Order of Industrial Service Merit for Registering Over 300 Patents on Next-Gen Memory first appeared on SK hynix Newsroom .
May 27
- Decoding AI: How AI Is Powering the Next Scientific RevolutionThe following article is an edited and abridged version of the original Korean contribution from Dr. Sungju Kang, also known as the YouTuber “Hang-sung.” A regular contributor to Korea’s top science YouTube channel Unreal Science, Dr. Kang reveals some of the most exciting scientific breakthroughs made possible by AI. The pace of progress in AI […] The post Decoding AI: How AI Is Powering the Next Scientific Revolution first appeared on SK hynix Newsroom .
April 1
- GlobalFoundries and UMC Merger Speculation: UMC Responds
➀ GlobalFoundries has been in contact with UMC regarding a potential merger, exploring a partnership two years ago but without progress.
➁ The proposed merger aims to create a larger wafer fabrication company, ensuring the supply of mature process chips in the US and investing in R&D.
➂ UMC and GlobalFoundries rank fourth and fifth in the wafer fabrication market with 5.5% and 4.7% shares respectively in Q4 2024.
➃ The combined revenue would reach $3.7 billion, potentially surpassing Samsung to become the second-largest wafer foundry after TSMC.
➄ Both companies have complementary process technologies and a production footprint across three continents.
➅ The merger faces challenges such as antitrust reviews by regulatory departments and handling agreements with Intel.
➆ UMC responded to the rumors by stating that there are no ongoing merger cases.
June 28
- Unveiling RowPress in Sub-20 nm DRAM Through Comparative Analysis With Row Hammer: From Leakage Mechanisms to Key Features1、探索了亚20纳米DRAM中RowPress引起的泄漏的物理机制；2、对比了单面和双面RowPress与Row Hammer，重点研究电子迁移和电容串扰；3、确定增加的电场是RowPress效应的主要贡献者，与Row Hammer的e-trap辅助机制不同；4、研究了RowPress的位翻转方向性、温度依赖性、访问模式依赖性和数据模式依赖性；5、评估了RowPress效应的可扩展性，为开发缓解技术和增强DRAM的可靠性和安全性提供了理论基础。
June 27
- Impact of Geometrical Factors on the Effectiveness of a Flexible Organic Thermoelectric Generator1、本研究利用分配印刷技术，探讨了几何因素对柔性有机热电发电机（f-OTEG）效率的影响。2、由噻吩并噻吩聚合物和单壁碳纳米管组成的f-OTEG，在7腿设计下，在69.1 K的温差下产生了95 nW的输出功率，显示出显著的性能提升。3、该设备在超过200天的时间内未出现退化，并在500次弯曲测试中保持了稳定的性能。
- Study on the Load Model of Microwave Power Detection Chip Based on MEMS Fixed Beams1、提出了一种基于MEMS固定梁的微波功率检测芯片，以提高性能和准确性。2、通过考虑梁的变形，利用挠曲方程建立了负载模型。3、与集总参数模型和关键力学模型相比，该模型分别提高了91.56%和43.67%的准确性。
- Adjoint Equations and Their Application to Helix TWT Design1、推导并应用伴随方程分析设计参数微小变化对螺旋行波管输出的影响。2、采用一维盘模型进行束流分析。3、伴随方法有助于行波管设计的敏感性分析和基于导数的优化。
April 16
April 10
- Wafer-Scale GaN-Si(100) Monolithic Heterogeneous Integration Inverters With Improved Output Voltage Swing and Fast Switching Capability by Transfer Printing and Self-Aligned Etching Technology
- Characterization and Modeling of Silicon-on-Insulator Lateral Bipolar Junction Transistors at Liquid Helium Temperature
June 11
- 苹果Vision Pro头显AI助手来袭：会调酒、能打麻将，甚至能开飞机本文介绍了苹果Vision Pro头显AI助手Otter，它可以实时识别现实场景并回答人类提问，支持八种语言，包括中文。Otter的训练过程中，团队专门使用了适用于AR头显的第一视角视频，并使用了MIMIC-IT数据集。Otter使用视觉识别模块分析场景信息，并结合ChatGPT生成答案。在主线流程之外，团队还引入了冷启动机制，用于发现数据库中可用的情景实例。Otter在各测试项目上的平均成绩比传统的MiniGPT-4、OpenFlamingo等传统模型高出十余个百分点。
- ChatGPT只讲这25个笑话！实验上千次有90%重复，网友：幽默是人类最后的尊严德国学者对GPT3.5做了大型测试，发现它只会讲25个笑话，1008次结果中有90%都是25个笑话的变体。所有25个高频出现的笑话都符合同一模式：先提一个让人摸不着头脑的问题，答案里出现双关语、谐音梗或其他技巧，大致都属于“冷笑话”范畴。这些笑话其实都是经典笑话，在网上有迹可循。研究人员得出的结论是，幽默对大语言模型来说仍然是挑战。ChatGPT没有创造笑话，只是在训练中记住了一些。用到一些“威逼利诱”的拷打技巧可以逼AI好好动脑子，引导ChatGPT讲出一个原创的笑话。
- 图像生成过程中遭「截胡」：稳定扩散的失败案例受四大因素影响近期发布的一篇论文《Stable Diffusion is Unstable》提出了一种基于梯度的文本到图像模型的对抗攻击算法，能够高效且有效地生成大量的攻击性文本提示，能够有效的发掘 Stable diffusion 模型的不稳定之处。该算法为研究文本到图像生成模型的失败模式提供了丰富的案例，为图片生成可控性研究奠定了基础。文章总结出四点生成失败的原因：生成速度的差异、粗粒度特征的相似性、单词的多义性和词语在 prompt 中的位置。
- 汽车传感器芯片，进入新阶段根据相关机构的研究，汽车传感器芯片正在进入快速迭代演进的新阶段。各大厂商纷纷转向“重感知，轻地图”的技术路线，加快布局城市NOA，打破对高精地图的依赖。雷达芯片市场中国厂商取得突破，打破海外垄断。4D雷达快速渗透到中高端车型和自主车型中。在常规雷达芯片领域，英飞凌和恩智浦几乎处于垄断地位。视觉传感器芯片方面，巨头竞相布局8MP产品。ISP集成SOC可以大大降低感知硬件的成本。
- 金刚石半导体，全球首创日本佐贺大学理工学部教授加须诚率领的研究小组成功开发出使用金刚石的半导体电源电路，证明了高速开关和长时间运行是可能的。金刚石半导体可以承受更高的电压，可以以更高的速度和频率运行，并且可以用于外层空间等高辐射环境。金刚石半导体作为下一代功率半导体的发展势头强劲。该电源电路有望应用于下一代通信标准“6G”、量子计算机等最新技术。该研究已申请专利。
- Apple Vision Pro的OLED屏幕，前所未见Apple于WWDC上宣布推出其首款增强现实（AR）头戴设备Vision Pro，该设备包含两个领先的微型OLED显示器，拥有令人难以置信的2300万像素，每个像素的大小与人类红细胞大小相当。Vision Pro有极高的像素密度，为有用的AR应用程序打开了大门，这些应用程序不会增加上一代AR显示模型的眼睛疲劳。该设备将配备Apple的M2芯片，但即使是中档Apple笔记本电脑中最常见的芯片M2，如果被迫以全分辨率驱动Vision Pro的显示屏，也会受到严重的负担，因此Apple使用全新的芯片Apple R1解决了这个问题，它承担了处理耳机摄像头输入的负担。像素密度的新前沿使得Vision Pro在所有AR/VR头戴式耳机中具有竞争力。
- 深挖BLE Phy：低功耗蓝牙基础这篇文章主要介绍了人工智能在医疗领域的应用。其中，人工智能可以帮助医生准确诊断疾病、预测疾病的发展趋势、设计个性化治疗方案、优化医疗流程等。在肺癌、心脏病、糖尿病等常见疾病的诊断方面，人工智能已经取得了良好的效果。此外，人工智能还可以通过对大量医疗数据的分析，提高医疗资源的利用效率，降低医疗成本。然而，人工智能在医疗领域的应用也面临着数据安全、隐私保护、伦理道德等问题。
- 晶圆代工厂，营收大跌40%晶圆代工厂世界先进与力积电5月营收同步滑落，世界先进降至新台币31.4亿元，月减12.04%。力积电为37.12亿元，月减4.03%，为今年次低。晶圆代工产业在2023年初面临了重大的营收危机，因为通货膨胀以及中国经济情势转坏，包含PC、手机等各种消费性电子的需求大幅衰退，晶圆代工业者也都面临了程度不等的营收衰退。目前在晶圆代工领域，仍持续经营先进制程代工与技术研发的只剩台积电与三星，英特尔代工服务要到2025年才会上线，因此暂不把英特尔列入代工业者内。
- iPhone两秒出图，目前已知的最快移动端Stable Diffusion模型来了Fusion 是一种移动端高性能 Stable Diffusion 模型，能够在移动设备上实现快速的图像生成。它有两个主要贡献：一是通过对现有 UNet 的逐层分析，提出一种新的高效 UNet 结构，可以等效替换原 Stable Diffusion 中的 UNet，实现 7.4x 加速；二是对推理阶段的迭代步数进行优化，提出一种全新的步数蒸馏方案（CFG-aware Step Distillation），减少步数的同时可显著提升 CLIP score，实现 6.25x 加速。在实验中，SnapFusion 的 CLIP score 表现优于 SD-v1.5，同时 FID 也没有变差。SnapFusion 在 iPhone 14 Pro 上实现 2 秒内出图，这是目前已知最快的移动端 Stable Diffusion 模型。
June 10
- 彻底反转：号称「碾压」LLaMA的Falcon实测得分仅49.08，HuggingFace决定重写排行榜代码本文介绍了近期开源模型领域的一次评测事件。LLaMA-65B是一款备受关注的开源语言模型，但近期发布的Falcon-40B模型在HuggingFace的OpenLLM排行榜上表现更好，引起了争议。一些团队重新编写了评测代码，并在相同的设定下进行了比较，结果发现LLaMA-65B的表现比排行榜上的低得多的得分更高，而Falcon-40B的得分则更低。这次评测事件彻底反转了之前的结论。
- 博通评论英伟达竞品：没有独到之处博通高管表示，英伟达在推出其Spectrum-X平台时，并没有提供独特的功能，而是构建了一个垂直整合的以太网平台来管理拥塞。这与Broadcom对其Tomahawk5和Jericho3-AI交换机ASIC的做法没有什么不同。但是Nvidia表示，在某些环境中，尤其是多租户云，以太网是首选。Nvidia的Spectrum-X平台声称可以应对这一挑战。需要明确的是，Nvidia的Spectrum-X不是产品。它是硬件和软件的集合。核心组件包括Nvidia的51.2Tbit/sec Spectrum-4以太网交换机和BlueField-3数据处理单元(DPU)。
March 31
- Jiang Shangyi: Intel is Now a 'Minor Player', Should Merge with Established Chip Technology Companies
➀ Jiang Shangyi, former Co-CEO of TSMC, openly discussed Intel's chip manufacturing challenges in a speech in Taiwan.
➁ He suggested that Intel should adopt mature chip manufacturing processes to compete, as it is far behind TSMC and unable to catch up.
➂ Jiang believes that Intel, once the 'king' of the chip industry, is now a 'minor player' and should merge with a company specializing in mass production of mature chip technology.
- Texas Instruments Announces Layoffs!
➀ Texas Instruments announced layoffs after receiving a $1.61 billion chip subsidy;
➁ The layoffs are part of a reform measure to support long-term operational plans, including job cuts at a factory near Tampa Bay Highway;
➂ Texas Instruments did not disclose the number of employees laid off, but said the figure was below the WARN Act notification threshold;
➃ The company received $9 billion for its Texas business and $7 billion for its Utah business as part of the $16.1 billion funding from the U.S. Department of Commerce;
➄ Despite the layoffs, Texas Instruments maintains its plan to build a second factory in Utah, but market concerns persist over the weakness of the analog chip market;
➅ Texas Instruments' fourth-quarter revenue was $4.01 billion, down about 2% year-on-year, and net profit was $1.205 billion, down 12% year-on-year;
➆ The company's embedded processing business revenue declined 18% year-on-year to $613 million, while other departments' revenue increased 7.3% year-on-year to $220 million.
March 26
- Investing in Shanghai Chip Unicorn: 200 Million Yuan!
➀ Shanghai Yachuang Electronics Group Co., Ltd. plans to up to 200 million yuan in Shanghai Analog Semiconductor Technology Co., Ltd.;
➁ Yachuang Electronics has been focusing on the automotive, industrial, and power sectors, and is a well-known domestic agent and provider of automotive entertainment navigation solutions;
➂ Analog Semiconductor, founded in 2018, specializes in chip design for automotive intelligent driving linear products, and data converters, mainly targeting industrial and automotive markets.
- Major Breakthroughs in Compound Semiconductor Field: Jinfengshan Lab Achieves Three Innovations
➀ Jinfengshan Lab achieves global first 8-inch silicon-based polar GaN substrate;
➁ China's first 100nm high-performance GaN flow PDK platform;
➂ 20-meter long-distance wireless energy transmission technology;
➃ Breakthrough in GaN material, device, and industrial application;
➄ Enabling 'instant charging' for mobile phones and 'while-driving charging' for electric vehicles;
➅ Accelerating the localization of high-end devices such as 6G base stations and autonomous driving radars.
- New Kai Lai, a Sensation!
➀ SEMICON China 2025 opened in Shanghai, with New Kai Lai participating for the first time to connect with downstream manufacturers.
➁ New Kai Lai's booth was a hit at the show, featuring a unique and stylish exhibition area that combined technology and fashion, attracting a lot of attention.
➂ The company showcased several upcoming products, including PVD, CVD, ALD for thin film products, RTP, EPI for diffusion equipment, ETCH for etching products, and various optical detection and measurement products.
- T-Bird: Born for UFS Testing
➀ The era of embedded storage requires;
➁ The challenges of UFS testing in product development and verification;
➂ The differences between UFS and SSD testing, including platforms, system environments, and drivers;
➃ Introduction of Thunder Bird (T-Bird) as a professional UFS testing platform;
➄ Features of T-Bird: high performance, comprehensive coverage, one-stop testing solution, high reliability, and application scenarios;
➅ Contribution of T-Bird to UFS testing technology development and its benefits to embedded storage development and application enterprises.