Recent #AI news in the semiconductor industry

3 months ago

① MIT's Neural Jacobian Fields (NJF) enables robot motion control through visual self-observation, eliminating reliance on physical sensors or digital models;

② The system maps motor signals to movements using multi-camera training data, achieving real-time 12Hz control with a single camera post-training;

③ Demonstrated on soft grippers and robotic hands, it shows adaptability but requires per-device training and currently lacks tactile feedback capabilities.

AICamerarobots
3 months ago

➀ The optical transport market is forecast to grow at 5% annually through 2029 after a 9% decline in 2024, reaching $19 billion;

➁ Growth is driven by AI datacenters and cloud providers investing in DCI technologies, including ZR/ZR+ optics and DWDM systems;

➂ Metro DCI shifts to IPoDWDM, while DWDM Long Haul for DCI projects 15% CAGR, with ZR/ZR+ optics revenue growing 20% annually.

AIHPC
3 months ago

➀ NASA成功测试了搭载在CogniSAT-6立方星上的AI系统Dynamic Targeting,利用Ubotica的SPACE:AI平台实现实时地形分析,自主调整观测角度,避开云层干扰;

➁ 该技术通过预判轨道前方300英里范围,仅拍摄无云区域数据,将有效数据比例提升三分之二,未来将扩展至野火、极端天气等动态目标监测;

➂ 后续计划包括联邦自主测量(FAME)多卫星协同观测,并探索将技术应用于雷达监测冬季强对流冰暴等罕见气象事件。

AINASAUbotica
3 months ago

➀ The latest patch for Wuchang: Fallen Feathers allegedly forces hidden upscaling even at 100% resolution, falsely boosting FPS while compromising image quality;

➁ Testing by hardware analyst Daniel Owen revealed a 28–37 FPS jump at "native" settings, suggesting the game now caps internal resolution at 80–85% despite claims;

➂ Players report reduced visual fidelity without disclosure, sparking backlash against developer 505 Games for deceptive optimization tactics.

AIGPUgaming
3 months ago

➀ Intel spins off its Network and Edge Group (NEX) to streamline operations and focus on x86 CPUs and AI-driven strategies;

➁ Quarterly losses of $2.9 billion and workforce reductions highlight financial challenges amid internal restructuring;

➂ Critical 14A node development faces cancellation without major external customers, signaling strategic reliance on partnerships.

AIIntelfoundry
3 months ago

➀ Thingiverse deploys AI-driven detection and human moderation to block 3D-printed ghost gun designs;

➁ Collaboration with authorities targets Chinese manufacturer Creality to enforce anti-firearm policies;

➂ AI tool '3D GUN’T' scans CAD files and uses printer cameras for real-time monitoring of gun-shaped objects.

3D printingAIcybersecurity
3 months ago

➀ Raja Koduri joins SanDisk's HBF Technical Advisory Board to advance high-bandwidth flash memory for AI accelerators;

➁ HBF technology enables up to 4TB VRAM on GPUs, targeting 8-16X capacity gains through 3D NAND and TSV integration;

➂ The innovation aims to revolutionize AI training/inference efficiency and edge AI deployment.

AIHBM
3 months ago

➀ Intel's Q2 2025 revenue remained flat at $12.9B with losses deepening to $2.9B;

➁ The company announced radical restructuring plans, including layoffs of 30,000 employees by 2025 and CEO Lip-Bu Tan's personal oversight of chip designs;

➂ Strategic shifts include pivoting toward AI inference/full-stack solutions, re-evaluating foundry investments (18A/14A nodes), and reviving Hyper-Threading for desktop CPUs to counter AMD's EPYC.

AIIntelfoundry
3 months ago

➀ IISc and QSRTC developed an AI-powered tool using machine learning to automatically assess corrosion in industrial machinery through microscope image analysis;

➁ The system measures deposit thickness and porosity to determine corrosion severity, correlating pH levels with damage stages (critical threshold at pH=3) for early warnings;

➂ Demonstrated 73% accuracy in steam generator tube tests, the unsupervised model streamlines industrial inspections and supports digital monitoring transformation.

AI
3 months ago

➀ Intel reported flat Q2 2025 revenue at $12.9 billion YoY, with losses widening to $2.9 billion and gross margin dropping to 27.5%;

➁ The company announced workforce reduction to 75,000 employees, halted expansions in Germany and Poland, and adjusted fab investments to align with confirmed customer demand;

➂ Intel prioritizes financial discipline, AI development, and optimizing its foundry business, with Data Center and AI group revenue growing 4% while Client Computing declined 3%.

AIIntelsemiconductor
3 months ago

➀ Elon Musk's xAI startup raised $10 billion in June 2025, seeks an additional $12 billion, and has an annual cash burn rate of $13 billion, with minimal reported revenue;

➁ xAI is constructing custom AI datacenters (e.g., the Memphis-based 'Colossus' with 200,000 Nvidia GPUs) through leased GPUs and complex financial deals to keep pace with rapid hardware advancements;

➂ Despite skepticism about its economics, Musk’s proven track record with Tesla and SpaceX fuels confidence in xAI’s ambitious goal to deploy 1 million GPUs for training its AI model Grok.

AINVIDIA
3 months ago

➀ TH Köln and brehmergroup are developing an AI-based gesture control system with haptic feedback integrated into motorcycle grips, aiming to minimize safety risks from manual device operation;

➁ The system detects gestures (e.g., tapping, swiping) via grip sensors to control functions like call acceptance and turn signals while keeping hands on the handlebars;

➂ The project builds on patented heated grip technology, with potential applications in e-bikes and scooters, supported by €220,000 in government funding until 2026.

AIautomotivesemiconductor
3 months ago

➀ LTO tape storage shipments surged to 176.5 exabytes in 2024 with 15.4% YoY growth, marking four consecutive years of record highs;

➁ Enterprises increasingly adopt LTO for AI/ML implementations, leveraging its cost-effectiveness, security, and compatibility with modern storage architectures;

➂ Despite the U.S. DOGE's dismissal of tape as 'obsolete,' industry leaders emphasize LTO's irreplaceable role in sustainable and high-capacity data storage solutions.

AIHPCstorage
3 months ago

➀ Hygon launched the C86-5G server processor, a proprietary x86 design to compete with Intel Xeon and AMD Epyc;

➁ The chip features up to 128 cores with SMT4, DDR5-5600 memory support, AVX-512 for AI workloads, and PCIe 5.0/CXL 2.0 connectivity;

➂ Acquired by Chinese supercomputer firm Sugon, it targets enterprise servers, AI clusters, and large-scale data centers.

AIHPCcpu
4 months ago

➀ HTWK Leipzig's Boosted Robots team won the RoboCup 2025 Humanoid League – Kid Size world championship with an 11:0 victory over China's TH-MOS;

➁ The team utilized new Booster K1 humanoid robots from Booster Robotics, enhancing computing power and AI capabilities for the competition;

➂ HTWK's innovative software adaptation and collaboration with Booster Robotics highlight advancements in AI and robotics, reinforcing the university's leadership in applied research.

AIrobotics
4 months ago

➀ The Boosted HTWK Robots team from HTWK Leipzig won the 2025 RoboCup in Salvador, Brazil, defeating China's TH-MOS 11-0 in the Humanoid League final;

➁ The team utilized new Booster Robotics humanoid robots, enhancing computational power, sensors, and AI programming for navigation and gameplay;

➂ The victory underscores HTWK Leipzig's leadership in AI and robotics, aligning with RoboCup's 2050 goal of robots outperforming human soccer champions.

AIHPCsemiconductor
4 months ago

➀ AMD and Stability AI co-launch Stable Diffusion 3.0 Medium AI model optimized for XDNA 2 NPUs, enabling offline 4K image generation on Ryzen AI laptops;

➁ The model requires Ryzen AI 300-series or MAX+ processors with 24GB RAM, uses 9GB VRAM during operation, and supports text-to-4MP image conversion;

➂ Currently free for small businesses under $1M revenue via Amuse 3.1 Beta, emphasizing privacy-focused local AI processing against cloud dependencies.

AIAMDNPU
4 months ago

➀ Researchers at Sejong University developed STAU, a hardware accelerator enabling edge devices to run large AI models like BERT and GPT with 5.18× speedup over CPUs and 97% accuracy;

➁ The design uses a Variable Systolic Array (VSA) and Radix-2 softmax optimization to reduce computational complexity and power consumption, cutting processing time by 68% for long inputs;

➂ Implemented on FPGA with a custom 16-bit floating-point format, STAU supports multiple transformer models via software updates, advancing on-device AI deployment without cloud dependency.

AIAI ChipHPC
4 months ago

➀ The datacentre processor market is projected to grow from $147B in 2024 to $372B by 2030, driven by AI expansion, with NVIDIA dominating the server GPU market (over 90% share) through its CUDA ecosystem and Blackwell/Rubin platforms;

➁ AI ASICs and custom chips (e.g., Google TPUs, AWS Trainium, Huawei Ascend) are rising as cost-efficient solutions for hyperscalers, while Arm-based CPUs like Amazon Graviton and Nvidia Grace gain traction for power efficiency;

➂ Geopolitical tensions and export controls spur local AI infrastructure investments, with China accelerating domestic semiconductor development and NVIDIA adapting export-compliant chips amid industry consolidation (e.g., SoftBank-Graphcore, AWS-Tenstorrent).

AIArmNVIDIA
4 months ago

➀ Kioxia launches the world's highest-capacity 245.76TB SSD using BiCS8 3D QLC NAND, targeting AI and hyperscale data centers;

➀ Features 32-layer NAND packages, PCIe 5.0 x4 interface, 12GB/s read speeds, but prioritizes density over peak performance;

➂ Includes advanced encryption and reliability features, with sampling to partners ahead of August 2025 conference.

AIKioxiaSSD