Today's Semiconductors Pulse
SMIC clears final hurdle for $5.97B takeover of SMNC
China’s Semiconductor Manufacturing International Corp (SMIC) secured regulatory approval to acquire the remaining 49 % of its Beijing‑based unit SMNC, valuing the stake at 40.6 billion yuan (≈$5.97 bn). The transaction will be funded by issuing 547.2 million A‑shares at ¥74.20 ($10.91) each, with a 12‑month lock‑up for sellers, making SMNC a wholly‑owned subsidiary.
Siemens U2U 3D IC Design and Verification Panel
Siemens convened a panel of industry experts to discuss the shift from 2D chips to 3D ICs and chiplet architectures. The discussion highlighted how stacking multiple dies improves performance, power efficiency, and bandwidth for AI, HPC, and data‑center workloads. Panelists warned that the transition brings system‑level complexity, thermal and power challenges, and supply‑chain bottlenecks in advanced packaging. They emphasized early architectural planning, standardization, and advanced simulation tools such as multiphysics and digital twins as essential to accelerate adoption.
AI100 Finalist Interview: ChipAgents
ChipAgents, an AI 100 finalist, is building a dedicated AI infrastructure for semiconductor chip design. Its platform deploys coordinated multi‑agent systems that can read specifications, plan tasks, generate RTL code, create verification environments, and autonomously iterate designs. By automating the entire...
JSR to Build First Taiwan Photoresist Plant to Co-Develop Advanced Resists with TSMC — Multi-Million Dollar Plant Could Come Online...
JSR, which holds about 20% of the global photoresist market, announced a joint‑venture to build its first production plant in Taiwan, targeting an operational date as early as 2028. The multi‑million‑dollar facility will co‑develop advanced photoresists directly with TSMC, closing...
China Silicon Wafers Push Boosts Eswin Capacity
China has set an informal mandate for domestic silicon wafer suppliers to meet 70% of the 12‑inch wafer demand by 2026, intensifying its push to localise the semiconductor stack amid AI‑driven demand and U.S. export controls. Xi’an‑based Eswin Material Technology...
Availability of Arasan UFS 5.0 Host Controller IP
Arasan Chip Systems announced the immediate availability of its UFS 5.0 host controller IP for ASIC and FPGA designs. The IP supports up to 46.694 Gb/s throughput using M‑PHY HS‑Gear 6, targeting high‑end smartphones and edge‑AI devices. Arasan, a long‑time member of the...
Your GPUs Aren’t Slow, They Just Have a Short Memory
Graid Technology is tackling the growing KV‑cache bottleneck that plagues long‑context, agentic AI models. When the cache exceeds GPU HBM, it overflows to slower storage, causing latency spikes of up to 18× and throughput drops of tenfold. Graid’s KV Cache Server,...
The Great Chip Divide: AI Chip War Pivots From Hype to Execution
Intel cleared a U.S. antitrust review for its investment in AI‑chip developer SambaNova, bolstering its generative‑AI roadmap. The chipmaker also hired former Qualcomm executive Alex Katouzian to run its client computing and physical AI group, signaling a push into edge‑AI...
OpenAI Accelerates AI Phone with Dual NPUs, Custom SoC
OpenAI's version of Jarvis -> Kuo: OpenAI appears to be fast-tracking its AI agent phone with two NPUs and a custom MediaTek Dimensity 9600 SoC, targeting mass production as early as H1 2027 Real-world visual sensing: "The ISP is the headline spec,...
THL Partners to Sell AMI to Lattice Semiconductor for $1.65bn
THL Partners announced the sale of its semiconductor design firm AMI to Lattice Semiconductor for a cash consideration of $1.65 billion. The transaction, subject to customary closing conditions, is slated to finalize in the third quarter of 2026. AMI brings a...
AMD Preps Linux For CPPC HighestFreq Feature Coming With Future ACPI Spec
AMD’s Linux P‑State driver is being updated to support a new “HighestFreq” ACPI register that is expected to land in the upcoming ACPI 6.7 specification. The register addresses cases where boost ratios cannot be derived through linear interpolation because core performance‑to‑frequency...
LLM System Design Interview #42 - The Global Memory Trap
In a mock DeepMind interview, candidates are asked why a 5× increase in raw teraFLOPs yields only a 1.2× boost in end‑to‑end throughput. The correct answer points to the memory wall: GPU compute has outpaced global memory bandwidth, leaving the...
GlobalFoundries Reports First Quarter 2026 Financial Results
GlobalFoundries posted first‑quarter 2026 revenue of $1.634 billion, a 3 % increase year‑over‑year but down from the prior quarter’s $1.830 billion. Gross profit rose to $451 million, lifting the IFRS gross margin to 27.6 % and the non‑IFRS margin to 29.0 %. Operating profit reached $180 million,...
Europe Is Building a System without a Physical Foundation
Europe is pouring billions into AI, semiconductors and data centres, yet its hardware backbone—advanced printed‑circuit‑board (PCB) manufacturing—remains largely offshore. Vytautas Ilgūnas, chief commercial officer of TLT PCB, warns that without a domestic PCB supply chain, the region’s technological sovereignty is incomplete....
Intel Slides 3.8% to $95.78 as AI‑Fueled Rally Tests New Resistance
Intel (INTC) dropped 3.84% to $95.78 on May 4, snapping a 166% year‑to‑date rally sparked by strong AI‑related earnings and new client contracts. The pullback, accompanied by 118 million shares traded, puts the stock at a key technical resistance level and...
Lattice to Acquire AMI for $1.65 B, Boosting AI‑Driven Operations Platform
Lattice Semiconductor signed a definitive agreement to buy platform‑firm AMI for $1.65 billion, split between $1 billion in cash and $650 million in stock. The deal is expected to be immediately accretive to gross margin, free cash flow and non‑GAAP EPS, and to...
MSI Launches 2026 Stealth 16 AI+ Laptop with Upgraded Cooling and AI‑focused GPU
MSI announced the 2026 Stealth 16 AI+ notebook, priced from $2,699, with a new all‑aluminum chassis, a beefed‑up cooling architecture and an AI‑optimized Nvidia RTX 5060 GPU. The launch signals MSI’s push to blend high‑performance gaming hardware with creator‑grade AI workloads.
Cerebras Systems Launches $3.5 B IPO at $115‑$125 per Share, Targeting $26.6 B Valuation
Cerebras Systems announced an IPO of 28 million shares priced between $115 and $125, aiming to raise up to $3.5 billion and achieve a peak valuation of $26.6 billion. The offering is led by Morgan Stanley, Citigroup, Barclays and UBS, and includes a...
OpenCL 3.1 Released To Bolster AI & HPC Workloads
The Khronos Group announced OpenCL 3.1, the first major update since the provisional 3.0 release six years ago. The new version pulls several previously optional extensions—such as SPIR‑V kernel ingestion, sub‑group execution, integer dot‑product operations, and a device UUID query—into the...
AXT’s Revenue Grows 17% in Q1 After Greater-than-Expected Export Permits
AXT reported Q1 2026 revenue of $26.9 million, up 17% year‑over‑year and slightly above the $26 million forecast, driven primarily by a surge in indium‑phosphide (InP) sales. Export permits from China arrived better than expected, enabling higher InP shipments to U.S. AI...
ESD Alliance Outlook Spotlights Agentic AI in Chip Design
The Electronic System Design Alliance’s 2026 Executive Outlook, slated for June 10 in San Jose, will put agentic AI at the center of the semiconductor conversation. Hosted at Cadence Design Systems headquarters, the event gathers leading EDA firms and AI‑focused startups...
Colour Lidar Launch From Ouster
Ouster has introduced its Rev8 portfolio of OS digital LiDAR sensors, featuring the world’s first patented native colour LiDAR powered by the next‑generation L4 silicon architecture. The L4 chip delivers up to double the range and resolution of previous models,...
ST Unveils 100W VIPerGaN Converters for Energy-Efficient Appliances
STMicroelectronics announced two new 100 W VIPerGaN high‑voltage converters—VIPerGaN100W and VIPerGaN100WB—targeting energy‑efficient appliances and consumer electronics. The devices handle 3.5 A and 4.2 A drain currents respectively, support a universal 85‑265 V AC input, and can deliver up to 125 W peak power. Built on...
The Next Phase of Europe’s Semiconductor Strategy
The EU is moving its semiconductor policy from a high‑profile fab push to a realistic, market‑driven approach. New analysis highlights the shift toward de‑risking without decoupling, focusing on application‑critical chips such as automotive, power and secure devices. At the same...
Memory Stocks May Finally Merit 20× PE Valuation
What if this time is really different for „cyclical“ memory stocks? What if SK Hynix should trade at a PE of 20x I mean, how arrogant would anyone have to be to seriously claim that this time is no different at all?...
Research Bits: May 5
MIT and the MIT‑IBM Watson AI Lab unveiled a lightweight model that predicts the power draw of AI workloads on GPUs and accelerators with roughly 8% error, cutting estimation time from days to seconds. The tool incorporates software‑level optimizations and...
Team Group Expands DDR5 Lineup with New 8000 MT/S ELITE Kits
Team Group announced the expansion of its DDR5 desktop memory portfolio with ELITE PLUS and ELITE DDR5 kits rated at 8000 MT/s. The modules run at 1.1 V with CL56‑56‑56‑128 timings and meet JEDEC specifications, reducing reliance on extreme overclocking. They feature Same‑Bank...
Microchip Expands dsPIC33A Controllers for AI Data Center Power
Microchip has expanded its dsPIC33A family with the dsPIC33AK256MPS306 digital signal controller, targeting high‑density AI‑driven data‑center power, motor control and intelligent sensing. The part features a 200 MHz 32‑bit core, double‑precision FPU, 78 ps PWM, 40 MSPS 12‑bit ADCs and hardware‑accelerated post‑quantum cryptography....
MIT Researchers Use AI to Uncover Atomic Defects in Materials
MIT researchers have unveiled an artificial‑intelligence model that can identify and quantify up to six distinct atomic‑scale point defects in semiconductor materials without destroying the sample. The model was trained on a database of 2,000 semiconductor specimens, representing 56 elements,...
Congatec Conga-TC300 COM Express Module Features up to Intel Core 7 350 Wildcat Lake Processor
congatec unveiled the conga‑TC300, a Type‑6 COM Express module powered by Intel’s 15 W Core Series 3 “Wildcat Lake” SoCs, including up to a six‑core Core 7 350. The board supports up to 64 GB DDR5‑6400 memory, optional UFS 3.1 storage, and a rich I/O set featuring...
Intel Is Said to Refresh Raptor Lake for LGA1700 Once Again: Old Socket, Renewed Life Extension, and a Rather Awkward...
Intel is reportedly planning a second Raptor Lake refresh for the LGA1700 socket, targeting an early 2027 launch. The leak suggests the update will reuse the Raptor Cove P‑cores and Gracemont E‑cores on the Intel 7 process, focusing on low‑to‑mid‑range models rather than...
AMD RDNA4: Driver Leaks Indicate a Change of Strategy for Upcoming GPUs
AMD’s driver updates reveal that its upcoming RDNA4 GPUs will likely skip a high‑end halo chip, focusing instead on smaller, power‑efficient designs for the mid‑ and upper‑mid‑range market. The leaks show new device IDs but no flagship SKU, suggesting a...
Intel Delays 18A Schedule: Manufacturing Problems Slow Down the Hopeful Centerpiece of the Foundry Offensive
Intel has delayed parts of its 18A production schedule as yield problems surface on the new RibbonFET and PowerVia node. The 18A process, touted as a breakthrough with Gate‑Around transistors and backside power delivery, is central to Intel's push to...
Shuttle XPC Cube SB860R8 Targets Workstation Workloads with Core Ultra 200 Support
Shuttle introduced the XPC cube SB860R8, a 14‑liter barebone workstation built around the LGA1851 socket for Intel Core Ultra 200 series processors. The system supports up to 24 cores, 192 GB DDR5‑5600 memory, and offers PCIe Gen5 x16 expansion, dual 2.5 GbE networking,...
Capture Data at Every Chip Manufacturing Stage
#Technology #Thread #Semiconductor #Manufacturing #Data The Semiconductor Capturing Data: 1/ - Every Chip Tells A Story. - But If You Do Not Capture The Data At Each Stage Of Semiconductor Manufacturing, You Are Missing The Entire Plot.
AI Hardware Spending Surges 31%, Software 15%
AI spending on hardware is running at an annualized clip of 31% the past five quarters, while software is 15%. Both are well above longer trends. https://t.co/jpzfvBjcKR
Apple Explores Using Intel and Samsung to Build Main Device Chips in the US
Apple is in early talks with Intel and Samsung to manufacture its primary device processors in the United States, adding a domestic alternative to its long‑standing reliance on Taiwan Semiconductor Manufacturing Co. (TSMC). Intel would provide its U.S. foundry capacity,...
Intel's Logic Deals Rise to Meet Capacity Demand
$INTC logic deals are coming. It’s not an indictment on $TSM, it’s purely capacity driven. Intel will get its chance to shine here as it will offer critical wafers to meet rising demand. 👏🏻
China Aims 70% Domestic Advanced Wafers by 2026
Exclusive: China targets 70% advanced domestic silicon wafer use by 2026 Local leaders led by Eswin drive major expansion for self-sufficiency milestone https://t.co/GCoRLhlgwu via @NikkeiAsia
NVIDIA Adds 12 GB RTX 5070 Mobile GPU, Boosting VRAM 50% without Architecture Change
NVIDIA unveiled a 12‑GB version of its GeForce RTX 5070 mobile GPU, expanding video memory by 50% while leaving the core design untouched. The move responds to a shortage of 16‑Gb G7 memory chips and gives laptop makers a clearer product...
CoWoS Evolves Into TSMC’s High‑Value Ecosystem Engine
CoWoS Is No Longer Just Packaging: It Is TSMC’s New Value-Creation Engine If CoWoS really has a value-added rate close to 50%, it reflects one thing: TSMC is not simply selling capacity. It is selling a closed high-value manufacturing ecosystem. https://t.co/7w2GYbvF7t
Intel Arc Pro B70 GPU Doubles Performance Over B580, Challenges RTX 5060 Ti
Independent testing by Expreview finds Intel's new Arc Pro B70 workstation GPU, equipped with 32 GB of VRAM, outpaces the previous Arc B580 by roughly 32.5% in raster workloads and closes the performance gap with Nvidia's RTX 5060 Ti. The results...
Samsung Shifts Focus to AI‑Driven Data‑Center Memory and Storage
Samsung Electronics announced it will prioritize memory and storage production for AI‑focused data centers, marking a strategic pivot from consumer devices. The move reflects soaring AI demand and positions Samsung as a key supplier for big‑data workloads.
Tokyo Researchers Build 25‑nm Memory Chip That Improves With Size
Professor Yutaka Majima’s team at Science Tokyo has demonstrated a 25‑nanometer ferroelectric memory device that performs better as it gets smaller, overturning a long‑standing scaling rule. The breakthrough could slash energy use and heat in smartphones, wearables and AI hardware,...
Atomera Inc (ATOM) Q1 2026 Earnings Call Transcript
Atomera reported FY 2025 revenue of $65,000, GAAP loss $20.2 million and non‑GAAP loss $16.1 million, with operating expenses rising modestly. The company confirmed definitive silicon data showing its Mears Silicon Technology (MST) can be deposited in gate‑all‑around nanosheet structures and outperforms...
YMTC Expands NAND and DRAM Ambitions with New Fabs Despite U.S. Sanctions Pressure
Yangtze Memory Technologies Corp. (YMTC) is launching three new fabs, including Phase 3 in Wuhan slated for mass‑production of cutting‑edge NAND in late 2026, with two additional 100,000‑wafer‑per‑month lines planned for 2027. Over half of Phase 3’s equipment is sourced from Chinese...
Blaize and Winmate Forge Strategic Partnership to Accelerate Edge AI Integration in Ruggedized Systems
Blaize Holdings and Taiwan‑based Winmate have signed a strategic partnership to embed Blaize’s Graph Streaming Processor (GSP) AI accelerators into Winmate’s ruggedized hardware platforms. The deal targets roughly $15 million in revenue during the first year and includes a three‑year term...
Atomera and Synopsys Expand Collaboration to Accelerate GaN Modeling in RF, Power Devices
Atomera Inc. has expanded its partnership with Synopsys Inc. to develop gallium nitride (GaN) device modeling for RF and power applications. The collaboration will leverage Synopsys’ Sentaurus TCAD tools and Atomera’s MSTcad to create a calibrated GaN workflow, marketing assets,...
Apple's Tim Cook Says AI-Fueled 'RAMmageddon' Will Make Your Next Laptop and Phone a Lot Pricier
Apple CEO Tim Cook warned that soaring AI‑driven demand for RAM is inflating component costs, a trend he called “RAMmageddon.” He said memory expenses rose sharply in the March quarter and will climb further in the June quarter, eroding gross...
Ottawa Plans to Spin Off Federal Semiconductor Facility Into “Commercial Entity”
The Canadian government announced it will spin off the National Research Council’s Canadian Photonics Fabrication Centre (CPFC) into a commercial entity. The move, unveiled by Industry Minister Mélanie Joly at the CHIPS NORTH conference, aims to draw private capital and...
All Major Memory Chip Makers Acknowledge Severe RAM Crisis
Micron has warned on the gravity of the RAM crisis, meaning that every major memory chip maker has now done so. https://t.co/XSOFwlwhiA