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.
Samsung’s Profit Surges On AI-Related Gains, Japan Moves to the Industrial Deployment of AI
Samsung projected a first‑quarter operating profit of about $38.7 billion, driven by soaring demand for AI‑optimized DRAM and HBM memory. South Korean chip makers, led by SK Hynix’s 60% HBM share, saw stock gains as memory prices are forecast to jump 58‑63% quarter‑over‑quarter. In Japan, Microsoft announced a $10 billion AI investment to build domestic compute capacity and train one million workers amid a severe labor shortage. The episode also covered cloud‑cost waste, AI‑driven observability at Grafana, and impress.ai’s AI‑powered hiring efficiencies.
Anthropic Builds Custom AI Chips to Ease Shortage
JUST IN: Anthropic is developing its own AI chips to combat the escalating shortage fueling their advanced systems.
Panmnesia Wins Government Project to Develop AI Accelerator Link Controller
Panmnesia announced on April 8 that it secured a South Korean government‑backed project to develop AI accelerator link controllers and switches using open‑standard interconnects. The initiative, part of the K‑Cloud AI semiconductor program, focuses on UALink and Ethernet technologies, with silicon...
AWS's Trainium Powers Anthropic, Sparks Amazon Gemini Moment
$AMZN is about to have its $GOOG Gemini moment. AWS CEO : “All Anthropic models out today were trained on Trainium.” Anthropic Mythos was most likely trained on Trainium. Trainium chips are sold out
A New Memory Chip Survives 700°C and Could Enable AI in Space
Researchers at the University of Southern California have demonstrated a memristor memory chip that functions at 700 °C (1,300 °F) without degradation. The device uses a tungsten electrode, hafnium‑oxide insulator and a graphene interlayer that blocks tungsten filament formation. It retains data...
Compute Domains & Multi-Node NVLink in Kubernetes: Scaling GPU Workloads
NVIDIA’s ComputeDomains add a Kubernetes‑native layer that dynamically creates and tears down multi‑node NVLink communication groups for GPU workloads. By extending the Dynamic Resource Allocation driver, the feature makes cross‑node bandwidth a schedulable resource rather than a static configuration. This...
Photovoltaic Driver Streamlines EV Power Designs
Vishay introduced the VODA1275, a photovoltaic MOSFET driver designed for high‑voltage automotive applications. It offers a 20 V open‑circuit output, 20 µA short‑circuit current, and an 80 µs turn‑on time—about three times faster than rival parts. The AEC‑Q102‑qualified device targets EV pre‑charge circuits,...
RISC-V SoC Supports Voice-Enabled IoT Devices
Espressif Systems began sampling its new ESP32‑S31, a dual‑core RISC‑V SoC that bundles Wi‑Fi 6, Bluetooth 5.4, Thread, Zigbee and Ethernet. Running at 320 MHz, the chip delivers 6.86 CoreMark/MHz, a 128‑bit SIMD path, 512 KB SRAM and up to 8‑bit DDR PSRAM for edge...
Coherent Advances Silicon Carbide Thick Epitaxy Capabilities for High-Voltage AI Datacenter and Industrial Power Applications Up to 10kV
Coherent Corp announced new thick silicon‑carbide (SiC) epitaxy platforms on 150 mm and 200 mm wafers that support power devices up to 10 kV, with demonstrated capability beyond that threshold. The technology targets high‑efficiency, high‑power‑density converters for AI‑intensive datacenters and industrial electrification such...
SiFive Announces $400M Series G Round
SiFive announced a $400 million Series G financing that lifts its valuation to $3.65 billion. The round was led by Atreides Management and attracted marquee backers such as NVIDIA, Apollo Global Management, Point72 Turion and T. Rowe Price. SiFive says the capital will speed development...
Infineon Talks Powering AI and Infrastructure
Infineon highlighted its strategy to power AI across the electrical grid and data‑center cores, emphasizing power electronics as a critical enabler for automotive, industrial and consumer applications. The company showcased its silicon‑carbide (SiC) and gallium‑nitride (GaN) product portfolio, along with...
At APEC 2026, AOS Showcases Its Expanding Portfolio with Advanced Controllers, Power Stages, and Protection Solutions
At APEC 2026, AOS unveiled a suite of new power‑management ICs aimed at AI‑centric workloads. The lineup includes the 16‑phase AOZ73216QI GPU controller, Intel‑compatible CPU controllers supporting up to nine phases, and compact Smart Power Stages for high‑performance compute. AOS...
Hypertec Becomes Key Partner for Nvidia in Canada
Hypertec Group’s Ciara division has been named Nvidia’s first original equipment manufacturer (OEM) partner in Canada, granting the Montreal‑based firm early access to GPU silicon, engineering support, and joint marketing. The partnership elevates Hypertec’s visibility and credibility, allowing it to...
CRDO As An AI Play
Credo Technology Group (CRDO) is shifting from a niche SerDes and active cable supplier to a full‑stack AI connectivity fabric architect. The company aims to close the reliability gap in massive GPU clusters by delivering vertically integrated interconnect solutions that...
Group‐III Nitride‐Based Wide‐Spectrum Multifunctional Synapses for Encrypted Light Communication and Image Recognition
Researchers have engineered InGaN core‑shell nanorod synapses that combine wide‑spectrum photodetection with stable photo‑electric memory. The devices achieve a peak responsivity of 31.47 A/W and sub‑250 µs response times under 810 nm illumination, while delivering tunable synaptic plasticity at 365 nm UV light. By...
WEKA Claims Nvidia CMX Support Plays to Its Strengths
Nvidia’s GTC 2026 announcement introduced the CMX KV‑cache extension for RDMA‑connected SSDs, prompting industry speculation that it could erode WEKA’s advantage with local SSDs in GPU servers. WEKA counters that its NeuralMesh client and server software already supports Nvidia Grace...
YieldHUB Expands Its Impact with New Technology and a New Website
YieldHUB has launched a redesigned website and introduced YieldHUB Live, a real‑time manufacturing intelligence layer for semiconductor test floors. The new portal groups solutions by product lifecycle, device architecture and user role, while the live platform delivers continuous visibility, anomaly...
Siemens Accelerates AI Chip Verification to Trillion‑cycle Scale with NVIDIA Technology
Siemens and NVIDIA announced that Siemens’ Veloce proFPGA CS hardware‑assisted verification platform can capture tens of trillions of pre‑silicon design cycles in just a few days. The breakthrough combines Siemens’ scalable FPGA‑based architecture with NVIDIA’s performance‑optimized chip designs, dramatically accelerating AI/ML system‑on‑chip...
Intel: In-House Fabrication and Market Challenges
Intel continues to manufacture its chips in‑house, a rare model among major semiconductor firms, while facing mounting competition from fabless rivals like TSMC, AMD, and Samsung. The loss of Apple as a customer in 2020 and a delayed Ohio megafab,...
Even Nvidia’s Own Research Teams Can’t Get Enough GPUs Amid the Race for AI Computing Power
Nvidia’s own research groups are scrambling for GPUs, highlighting a sector‑wide shortage of the $30,000‑plus chips that power AI model training. At the HumanX conference, applied‑deep‑learning lead Bryan Catanzaro confirmed that even internal teams must fight for compute allocations. The scarcity...
Nvidia N1 Engineering Sample Motherboard Leak Shows 128GB LPDDR5X Configuration
Nvidia’s upcoming N1 ARM‑based processor was spotted on a Chinese marketplace as an engineering‑sample motherboard, indicating early validation of the chip for thin‑and‑light devices. The board packs eight LPDDR5X modules for a total of 128 GB memory running at 8533 MT/s, and...
Intel's EMIB-T Packaging Technology Set for Fab Rollout This Year — as TSMC CoWoS Capacity Remains Limited,EMIB-T Is Preparing for...
Intel announced that its next‑generation EMIB‑T advanced‑packaging technology will enter fab production this year, positioning the company to capture AI‑accelerator demand as TSMC’s CoWoS‑L capacity stays oversubscribed. EMIB‑T adds through‑silicon vias to the existing embedded bridge, enabling HBM4‑class power delivery...
NXP Expands Arteris NoC Deployment to Scale Edge AI Architectures
NXP announced an expanded deployment of Arteris’s NoC and cache‑coherent IP suite—including FlexNoC®, Ncore®, CodaCache® and the Magillem® integration platform—across its upcoming edge‑AI silicon. The move targets the growing need for deterministic latency, high bandwidth, and safety‑critical isolation in heterogeneous...
Taming Skyrmions: Atom-Thin Magnets Point to Ultra-Dense, Low-Power Memory
Researchers at Argonne National Laboratory used cryogenic Lorentz transmission electron microscopy to directly image magnetic domains and skyrmion evolution in atom‑thin Fe₃GeTe₂ (FGT). The study shows that sample thickness and applied magnetic field precisely control skyrmion size, density, and reversal...
Rising Memory Costs Threaten AI ROI and Market Confidence
While we talk about memory prices 📈, and while that is true (great for memory names), I am worried about what this does to capex costs and in return on AI $$. If the cost to serve AI goes...
Nvidia Invests In CPU Chip Startup SiFive
Nvidia disclosed a strategic stake in SiFive, the RISC‑V processor startup, as part of a $400 million oversubscribed financing round that values SiFive at $3.65 billion. The new capital will accelerate SiFive’s development of custom RISC‑V CPUs and AI IP aimed at...
Tenstorrent Launches $9,999 AI Workstation with 384 GB RAM
New Tenstorrent AI workstations are expected to cost US $9,999, with 384 GB of memory, designed to run cutting-edge AI models at your desk. @Nvidia has released its own AI workstation with 748 GB. https://spectrum.ieee.org/ai-workstation-looks-like-pcs
SiFive Raises $400M To Double Down On High Performance RISC-V For Data Centers
SiFive announced a $400 million Series G financing round to accelerate its high‑performance RISC‑V offerings for data‑center workloads. The round was oversubscribed, with lead investors including NVIDIA and Apollo Global Management. Proceeds will fund new CPU core designs, accelerators, and system IP,...
World Briefs | Intel and Google Focus on AI CPUs in Expanded Partnership
Intel and Google have broadened their alliance, with Google committing to deploy Intel’s latest Xeon 6 processors for AI inference and general‑purpose workloads. In Brazil, the government plans to appeal a court decision that halted a newly imposed 12% oil export...
RISC-V Optimized Strnlen Implementation For Linux 7.1 Yields Big Speed-Up
A hand‑optimized RISC‑V implementation of the kernel’s strnlen() function is slated for Linux 7.1. Developed by Feng Jiang of KylinOS, the assembly version includes a generic path and a Zbb‑enabled variant, delivering up to a 427.5% speed increase in benchmarks. The...
Tuurny Deploys AI Robotics to Harvest RAM Chips, Tackling US Chip Shortage
Tuurny unveiled its Nantul robotic cell, capable of pulling up to 300 RAM integrated circuits per hour from domestic e‑waste, and secured a six‑figure contract with UK processor Areera. The system promises a new on‑shore source of memory chips while...
Intel and Google Team up for Next‑gen AI Cloud Infrastructure
INTEL $INTC AND GOOGLE $GOOGL PARTNERSHIP Intel and Google just announced a multiyear collaboration to "advance the next generation of AI and cloud infrastructure"
Mythos Launches Blackwell AI, GPUs Promise 2‑3×
Get ready for the acceleration 🚀 Mythos is JUST the first Blackwell class AI Vera Rubin GPUs land in months → 2-3x better models by December. Feynman in 2028 → another 2-3x leap. Mythos 10T → 100T–500T+ class by end-2028
HRL’s T3L 40nm GaN-on-SiC Technology Achieves Manufacturing Readiness Level 6
HRL Laboratories announced that its 40 nm T3L gallium‑nitride on silicon‑carbide (GaN‑on‑SiC) technology has attained Manufacturing Readiness Level 6, confirming production‑grade manufacturability under U.S. Department of War standards. The milestone paves the way for high‑volume manufacturing through a partnership with MACOM Technology...
Google Partners with Intel for SmartNIC‑powered Datacenters
Google wants more Intel inside ... its datacenters, taps Chipzilla for more SmartNICs https://t.co/NWjnTCSr9z https://t.co/IGYOky9NxI
AI Compute Shortage Far Worse Than Expected
I was not emphatic enough about just how short we are of compute, memory, energy, etc. The demand curve for AI has been greatly understated. And the estimates aren’t bullish enough. Watch. 👀 We do not have even close to enough to...
Chip Can Project Video the Size of a Grain of Sand
Researchers from MIT, the University of Colorado, Sandia National Laboratories and MITRE have unveiled a sub‑0.1 mm² photonic chip that can steer light with unprecedented speed. The device uses voltage‑actuated metallic cantilevers to project up to 68.6 million light spots per second...
Amazon Graviton Surges; CPU Bottleneck Drives ARM AGI Push
$AMZN Graviton is soaring. As we've said, the CPU is the next great bottleneck. $ARM is on to something its AGI CPU.
Nvidia N
Alleged images of the long-awaited Nvidia N1/N1X SoC surface on laptop motherboard — board features 128 GB of LPDDR5X memory alongside 8+6+2 phase VRM https://t.co/fFeae7JJ8r
Advanced iBGA Image Sensor Encapsulation
DELO introduced a PFAS‑free encapsulation material designed for automotive image sensors used in advanced driver‑assistance and autonomous systems. The formulation delivers low warpage, fine‑pitch flow control, and automotive‑grade reliability, addressing the harsh temperature, vibration, and moisture conditions of vehicle environments....
Accelerator Mismatch Turns Datacenters Into
Historically this doesn’t work In chips we call it dead silicon Predicting dead datacenter Ratioing accelerators to the main computer is fragile Models change, size changes and you have a brick Possibly $$$
Liquid Cooling Unlocks AI-Scale Compute Capacity
Part 2 of the future of the datacenter series. 800 vDC + liquid cooling becomes a co-design situation with compute infra decisions. The key is the compute capacity it unlocks. Liquid Cooling: The Thermal Prerequisite for AI Infrastructure...
Amazon CEO Says Chip Business 'On Fire' As AWS Steps Up Challenge To Nvidia
Amazon’s cloud unit reported a $15 billion annual AI revenue run rate and a custom‑chip business now exceeding $20 billion, double the Q4 figure. CEO Andy Jassy highlighted the rapid growth of its Trainium 3 accelerator, which promises better price‑performance than Nvidia GPUs....
CPU Designs Race Ahead, yet Trail AI Breakthroughs
Processor architectures are evolving faster than ever, but they still lag the pace of AI development https://t.co/KPtUjyF6PQ #processors #CPU @Arm #AgenticAI @Synopsys @unisouthampton @arteris_noc @quadric_io #NPU @Keysight @Cadence @AlphaDesignAI https://t.co/3Nu81L1qgc
Eliminate Davis‑Bacon to Unblock the CHIPS Act
Great detailed piece on how Davis-Bacon was a problem for the CHIPS Act, as some of us predicted it would be. But the obvious answer is to get rid of Davis-Bacon entirely, which is not the recommendation for some reason....
Nvidia Rubin GPUs May Be Delayed, Slowing the Next Phase of AI Infrastructure
Nvidia’s next‑gen Rubin GPUs, slated for late‑2024 shipment, face supply delays due to HBM4 memory validation, CX9 interconnect rollout, and higher power and cooling demands. TrendForce now projects Rubin to represent only 22% of Nvidia’s 2026 shipments, down from an...
Amazon and Nvidia Share Israeli AI Chip Roots
For those of you blown away by $AMZN announcement on Trainium4 and $NVDA growth, they have something important in common…🇮🇱🇮🇱
Intel Missed Chance by Denying Gelsinger's Foundry Investment
Can you imagine if Intel's board let @PGelsinger invest in the foundry capacity he asked them for? I agree, it seemed crazy at the time, but hindsight is 20/20.
HighPoint Announces Rocket 1604L Compact PCIe Gen5 X16 Retimer AIC for AI and Industrial Edge
HighPoint Technologies unveiled the Rocket 1604L, a 4‑M.2 PCIe Gen5 x16 add‑in card powered by Astera Labs' Gen5 retimer. The compact 167 mm module delivers full 32 GT/s bandwidth with near‑zero latency, targeting AI edge and high‑velocity compute workloads. It integrates a Smart Firmware Layer...
TSMC and Top Memory Makers Set Prices, Customers Agree
TSMC, and the three memory names right now, are the primary beneficiaries of a name your price market dynamic. Meaning they can literally name their price and customers will pay it.