Chip Industry Week In Review

Lam Research and ASE expansions; data center shifts; CPO system architecture initiative; 2D-transistor interface solution; USENIX HW security; patent suits; image sensors; programmable AI memtransistor; Nvidia’s new plan; earnings. UC Berkeley experts argue that AI progress does not require ever-larger data centers. Smaller open-source models are becoming capable enough for many tasks while using far less compute, energy, and water — and increasingly they could run on local HW. Data-center network fabrics are undergoing another architectural shift as optical interconnects assume a larger role in AI-cluster connectivity. The push toward 1-megawatt racks is forcing fundamental changes in data center architecture, including cooling, power delivery, rack design, and 3D-IC packaging. CSISrecommends tying data center incentives to measurable community benefits, prioritizing brownfields and existing industrial or federal sites, and locating new compute capacity near universities, national labs and advanced-manufacturing hubs. CPO is taking off as AI data centers leverage every possible option for improving performance and reducing power, but keeping pace with demand is a challenge as CPO shifts from lab instrumentation to production ATEs on a factory floor. Capacity ASE’s subsidiary SPIL broke ground on a nearly US$3.1B advanced packaging and test plant in Douliu, Taiwan. The 6-hectare facility is expected to add CoWoS capacity for AI chips, with first-phase production targeted for 2028 and more than 2,200 jobs planned at full buildout. Lam Research will spend $3B over 5 years to expand its R&D lab network, increasing its experiment capacity by more than 50%. China is rapidly gaining ground across the semiconductor supply chain, with legacy-chip capacity projected to approach half of global output by 2030 and domestic equipment makers expanding even as access to leading-edge AI chips remains constrained, predicts Rhodium Group. Notable deals Lam Research and NY Creates are partnering to train about 3,500 university students in semiconductor process integration over the next five years. The program will deploy Lam Reserach’s SEMulator3D virtual fabrication software across colleges and universities in NY Creates’ Northeast workforce-development network. Sony Semiconductor and TSMCinked a deal on their next-gen image sensor joint venture in Japan. Sony will contribute about ¥465B (~$2.9B) and control the venture, while TSMC will contribute about ¥282B (~$1.8B). The Kumamoto operation is expected to begin volume production of smartphone image sensors in 2029. Required regulatory approvals are still pending. Quantinuum and Oracle partnered to accelerate the adoption of hybrid quantum compute, with Oracle customers able to combine its graphics-processing and HPC services with access to Quantinuum’s 98-qubit Helios quantum computer at an Oracle cloud facility. Memory Counterpoint says enterprise SSDs accounted for 48% of global NAND bits shipped in Q2, versus 26% a year ago, and YMTC moved into third place in NAND shipments with a 14% share. Kearney says the memory makers are deliberately limiting NAND investment and instead prioritizing higher-return HMB and DRAM, prolonging tight NAND supply. Kioxia and Sandisk released high-performance 2 Tb QLC 3D flash memory, featuring a CMOS directly Bonded to Array (CBA) architecture, improving performance without requiring a new NAND cell stack. SK hynix is restarting investment in its second Dalian Chinese NAND fab, reports Seoul Economic Daily. Point2 Technology extended its Series B to a total of $136M, with participation from Arm, for its RF‑based interconnect solutions for AI data‑center infrastructure. The South Korean government is creating a ~$3.5B semiconductor fund targeting promising materials, components, equipment, and fabless companies, alongside $3.5B in trade financing for export-oriented suppliers and a $700M 10-year program supporting joint chip development, validation, and production by a variety of semiconductor companies. Advancements KAIST developed a “programmable dynamic memtransistor, featuring a dual-functional gate stack that enables HW-level control of dynamic behavior without constant biasing or input preprocessing.” The researchers claim reduced prediction errors by up to 40-fold versus conventional devices. QuTech and TU Delft researchers demonstrated a flip-chip architecture that integrates superconducting qubits using electroplated indium bump bonds, achieving quality factors around 10⁶ and showing a potential path toward scalable 3D quantum-chip integration. Of particular note, Applied Materials expects revenue from its advanced-packaging equipment business to grow more than 70% in 2026, driven by investment in HBM and 3D chiplet stacking. Patent infringement Navitas Semiconductor filed a patent infringement lawsuit against Renesas Electronics, alleging that Renesas’ GaN products, including its SuperGaN power semiconductors, infringe four Navitas patents. Seoul Semiconductor and affiliate Seoul Viosys won a permanent injunction from the Delhi High Court barring Ornate Agencies from selling products that infringe Seoul’s opto-semiconductor patents. Siemens plans to invest more than $200M in new manufacturing capacity in Georgia and Texas to expand production of electrical infrastructure for data centers, semiconductor fabs, and other industrial markets. Indium opened a new manufacturing facility in Guadalajara, Mexico to expand production for electronics and semiconductor customers in the Americas. UC Berkeleywill joinApplied Materials’ EPIC Center in Silicon Valley as a research partner. Asia NXPbroke ground on its assembly and test site in Malaysia. South Korea plans to complete relocation and temporary dispersal of military facilities at the Gwangju Air Base by the second half of 2028, clearing the site for a new semiconductor industrial complex. In-Depth Semiconductor Engineering published 2 newsletters, 2 special reports, 4 articles, and an opinion piece this week: Onto Innovation completed its acquisition of a 27% stake in Rigaku for approximately $720M, deepening a partnership announced in April to develop X-ray-based process control technologies for semiconductor manufacturing. Semtech is selling its cellular IoT module business to Compal for $62M so it can focus on data center and LoRa connectivity. Fundings Soctera raised $4M to develop its power amplifiers for mil/aero, halving thermal resistance and reducing GaN content by co-optimizing the semiconductor stack electrically and thermally. Why Wall Street and Nvidia Are Building an Exotic Money Pipeline for the AI Boom (WSJ) New Technologies Design Vertical integration is becoming more common, but software iteration time is now much faster than hardware, making co-design a catch-up game for hardware. New tools are being developed to improve this situation. Googlejoined the OpenROAD Initiative aimed at open-source silicon innovation. Samsung adopted Anthropic‘s Claude Code and significantly shortened timelines for functional verification of customized SoCs and for early-stage software development for semiconductor projects, according to ChosunBiz. Microsoft will reportedly unveil its new Maia 300 accelerator in September, but TSMC capacity may be a hurdle, per the Information via Reuters. AI data centers OpenLight’s photonic PDK is available for Tower‘s PH18DA InP-on-Si photonics platform in Cadence’s EDA tools, for applications such as optical interconnects, AI infrastructure, and sensing. Sivers Semiconductors and SemiNex announced a $3.4M program for high-power, multi-wavelength InP light sources needed for CPO and emerging scale-up and scale-out architectures. The Open Compute Project is collaborating with Google, Microsoft, and Nvidia to establish 800V DC as the open, standardized power architecture for AI data centers, accelerating the industry transition to LVDC (low-voltage DC). Research LBNL researchers demonstrated ferroelectric behavior in ultra-thin titanium dioxide (TiO₂) films, a material that normally behaves as a conventional dielectric. Because ferroelectric materials can switch at lower voltages, the discovery could provide a new materials path for reducing power consumption in future logic and memory devices while remaining compatible with silicon-based semiconductor manufacturing. An Argonne-led team found that changing the geometric arrangement of nanomagnets can tune whether they settle into predictable or probabilistic states, offering potential ultra-low power logic and AI hardware applications. Washington State and UW-Madison researchers created an AI-based power-delivery optimization method for 2.5D chiplet systems that predicts voltage droop on individual chiplets and adjusts operation before violations occur. High-Density Chiplet Integration using Face-Down COW Processes with Bumpless Interconnects for Heterogeneous 3D Systems (Science Tokyo) Security Attacks, vulnerabilities Taiwanese officials said an “unprecedented” AI-assisted attack compromised at least 85 government accounts and stole more than 2,500 personnel records after autonomous agents mapped 21 government systems, the Financial Times and the Dream Group reported. UCSD and Oberlin researchers developed a coin-sized hardware implant that, once physically installed in an externally accessible Boeing 737 port, can inject commands into the aircraft’s avionics network and potentially alter its autopilot flight plan, with remote control possible over Wi-Fi, Wired reports. Delta Air is investigating an incident where an “evil twin” hoax WiFi network appeared to passengers on board a flight to Atlanta. Attackers did not compromise any of the plane’s flight safety systems, company and FAA officials told the media. AI security startup CloudSEK said it has identified more than 2,500 organizations potentially exposed by an AI supply chain attack earlier this year involving LiteLLM, an open-source AI gateway tool. CISA issued new guidance on countering attacks from the Gunra ransomware-as-a-service variant, alongside several other national security agencies in the U.S. and South Korea. CISA’s alerts for the week are here. Cybersecurity programs, releases BTQ and ITRI validated the QCIM PQC architecture and core IP in a hardware design environment, an early step toward implementing NIST-standardized post-quantum algorithms in CIM hardware. NIST is seeking feedback on modernizing the National Vulnerability Database with AI-enabled automation, including its new V-etalon tool for enriching vulnerability data and updates to better represent hardware vulnerabilities. DEF CON Franklin and the National Rural Water Association have partnered to develop a new program providing cybersecurity services to municipal water systems, on the heels of a series of Iran-linked cyber-attacks on local U.S. water systems. Hospital leaders in the U.S. have formed the Health AI Cybersecurity Work Group to develop industry-wide frameworks for defending against frontier AI cyber attacks. New security research New hardware-related security research from this week’s USENIX Security Symposium in Baltimore is now available, including: Kinetic Technologies announced the KTS1642Q, an AEC-Q100 qualified automotive load switch. The switch is designed to help automotive electronic systems withstand the demanding electrical conditions defined by ISO 16750-2 and ISO 7637-2. LX Semicon has started mass production for the LX61101, an automotive MCU for Hyundai and Kia vehicles. The product will be the first South Korean-produced MCU for either carmaker. TrendForce says all-solid-state batteries have moved beyond lab proof-of-concept toward TRL 5–6, with Toyota, Honda, Nissan and Samsung SDI among companies ahead in pilot production. Auto deals and funding The UK government has committed £65M in investment, with equal industry matching, to support advanced AV and EV technology development in the country. Archer signed a definitive agreement with Boeing to acquire its Wisk Aero, Insitu and SkyGrid subsidiaries through a new strategic investment partnership. The autonomous-flying shake-up will also yield Boeing a large stake in Archer going forward. The White House announced more than $3B in federal funding for a slew of domestic mineral mining and lithium-ion battery projects. The funding includes a conditional $1.4B loan to California battery-maker Sila Nanotechnologies. A new California state program offering $3,500 rebates for first-time EV buyers opened up, with consumers exhausting the allotment carved out for Tesla vehicles in just five days. Improving Yield Through Shared Data: Increasing complexity due to advanced packaging, multi-die assemblies, and more devices under test is having an impact on yield, which in turn slows time to market and impacts overall chip costs. What’s needed is a way to share data that previously was siloed by chipmakers, fabs, and OSATs. Jayant D’Souza at Siemens EDA talks about the underlying drivers for sharing data, how it can be done securely, and why this is suddenly getting so much attention. Quantum Pasqal and Aeponyx generated four optical traps through a single PIC and used them to hold four individual rubidium atoms, a step toward replacing bulky free-space optics with scalable chip-based photonics for neutral-atom quantum computers. TU Delft and ICN2 researchers integrated a high-field superconducting material with a buried Ge/SiGe quantum well using a lithography-compatible process that preserved high carrier mobility after a 500°C anneal. The approach could support tighter integration of superconducting and semiconductor devices for scalable quantum HW. U. of Michigan and LANL researchers will develop AI agents that automate HW-SW co-design, with the goal of accelerating development of specialized chips and computing systems for scientific workloads under a new DOE award. Mahindra U. launched a semiconductor HackerFab in India to give students hands-on experience building fabrication equipment including maskless lithography, sputtering, spin-coating and furnace systems. Vietnam launched an ambitious national AI workforce initiative targeting 10 million workers trained in basic AI skills by 2030. Reducing variation in manufacturing, monitoring behavior over time, and targeting specific workloads can have a big impact on power, performance, and area/cost. This website uses cookies to improve your experience while you navigate through the website. The cookies that are categorized as necessary are stored on your browser as they are essential for the working of basic functionalities of the website. We also use third-party cookies that help us analyze and understand how you use this website. We do not sell any personal information. By continuing to use our website, you consent to our Privacy Policy. If you access other websites using the links provided, please be aware they may have their own privacy policies, and we do not accept any responsibility or liability for these policies or for any personal data which may be collected through these sites. Please check these policies before you submit any personal information to these sites. Necessary cookies are absolutely essential for the website to function properly. This category only includes cookies that ensures basic functionalities and security features of the website. These cookies do not store any personal information. Any cookies that may not be particularly necessary for the website to function and is used specifically to collect user personal data via analytics, ads, other embedded contents are termed as non-necessary cookies. It is mandatory to procure user consent prior to running these cookies on your website. Leave a Reply