XenoSpectrum — Latest News
-
テクノロジーIntel May In-House 80-90% of Nova Lake Tiles, as 18A Improvements Seen Reducing TSMC Reliance
KeyBanc projects that Intel will manufacture 80-90% of Nova Lake tiles in-house on 18A. We examine the 18A improvements confirmed in official disclosures and the conditions under which reduced TSMC reliance would affect gross margin and supply capacity.
-
テクノロジーTYLsemi Builds Custom AI Chips from Parts, Aims to Cut Development Time by Up to 50% with $43 Million Raise
US-based TYLsemi has unveiled a chiplet and volume-production support model that standardizes the surrounding functions of AI chip development. By providing reusable dies that handle I/O and power delivery, the company says customers can focus solely on their proprietary compute circuit designs, enabling significant reductions in development time and cost.
-
テクノロジーGoogle Images Moves Toward Discovery "Before You Search," Bringing a Personalized Gallery to US Desktop
Google is introducing a personalized gallery to desktop Google Images in the US that updates based on user interests. AI Overviews also gains image generation powered by Nano Banana 2 Lite.
-
テクノロジーLucid Fully Denies Bankruptcy Consideration Report, Cites $4.7B Liquidity and $1.4B Cash Burn
Lucid denied reports of considering bankruptcy or going private, but its roughly $4.7 billion in liquidity relies heavily on a PIF-affiliated credit facility and a recent capital raise. Over $1.4 billion in Q1 cash burn and production improvement will be tested at the August 4 earnings report.
-
テクノロジーWSL 2 Gets cgroup Isolation to Prevent Total Collapse Under Heavy Load
Microsoft has merged into master a design that isolates WSL 2's user processes with cgroup v2, preserving resources for core processes even under heavy load. The change is not yet in the latest pre-release, leaving distribution timing and cgroup v1 compatibility as open questions.
// TECHNOLOGY
-
テクノロジーChina's DF1000 Takes On AI Training at 14nm, Shifting the Competitive Bottleneck to 3D Stacking
Shanghai-based Dongfang Suanxin has unveiled the DF1000, an AI chip that combines a 14nm process with 3D stacking technology against the backdrop of US export controls. Its design vertically stacks logic circuits and DRAM to achieve high memory bandwidth, but mass-production yield and building a practical software ecosystem remain challenges ahead.
-
テクノロジーSamsung Recruits Across HBM4-to-HBM5 Design, Bonding, and Qualification Simultaneously
Samsung has launched an experienced-hire recruitment drive spanning HBM design, packaging, and customer qualification. With an eye on scaling up HBM4 production and developing next-generation products, the company aims to strengthen coordination across processes to shorten development timelines and improve mass-production yields, boosting its competitiveness in the AI memory market.
-
テクノロジーSamsung's Operating Profit Jumps 19x, Yet Stock Falls 10% as New Memory Plants Won't Catch Up Even by 2034
The memory semiconductor market keeps posting record earnings on the AI boom, yet Samsung's stock fell instead. With more than three years needed from groundbreaking to mass production at new fabs, and SK hynix targeting full-scale capacity expansion only by 2034, this piece examines the supply lag and compares the current surge quantitatively against the two previous boom-and-bust cycles.
-
テクノロジーOpenAI Expands Usage Limits for GPT-5.6 Sol, Anthropic Also Extends Fable 5 Again
OpenAI and Anthropic are rolling out measures to ease usage restrictions on their top-tier models and improve work efficiency within fixed-rate plans. OpenAI is curbing consumption through reasoning optimizations, while Anthropic is temporarily expanding access to its premium model—intensifying competition on the practical front.
// SCIENCE
-
サイエンスRecreating the 'Stealing Energy from a Black Hole' Thought Experiment on a Tabletop: How an Electronic Circuit Overturned the Principles of Wave Amplification
By applying black hole energy extraction theory to create synthetic rotation within an electronic circuit, researchers have succeeded in amplifying electromagnetic waves. This new principle promises to solve the power consumption and heat dissipation challenges plaguing conventional amplifiers, positioning it as a revolutionary technology to underpin next-generation ultra-high-speed optical communication infrastructure.
-
サイエンスQuantum Computers Overcome Errors Through "Self-Learning": Google's New Technique That Never Stops Computation
To correct drift in control parameters that occurs during quantum computation, a Google research team has developed a method that leverages quantum error correction detection signals for reinforcement learning. By using local error rates as a metric to autonomously adjust tens of thousands of parameters, continuous calibration becomes possible without ever halting computation.
-
サイエンスNew Structure Eliminates "Electron Traffic Jams" to Zero — KAIST Nullifies Semiconductor Contact Resistance with 2D Material
A Korean research team has developed a technology that eliminates contact resistance at the junction with metal electrodes by forming continuous semimetallic and semiconducting regions within a single platinum diselenide thin film. Atomic-level observation proves the absence of electron traffic jams, contributing to lower power consumption in next-generation chips.
-
サイエンスToward Heat-Free "Single-Atom Memory": University of Tokyo Demonstrates New Principle for Reading and Writing Data with Zero Electric Current
A research group at the University of Tokyo has developed, for the first time in the world, a technology that reads and writes the magnetic information of a single atom using only quantum mechanical "force," without using any electric current whatsoever. Because this method can prevent heat generation from Joule heating as well as malfunctions, it represents a groundbreaking achievement toward realizing the ultimate high-density memory.