Vertical DRAM Stacking Expands Thermal Design Space Beyond HBM-on-GPU (imec)
Researchers at imec evaluated a 3D memory architecture that vertically stacks DRAM dies directly above a GPU, incorporating cooling cavities between layers. This approach serves as a thermal alternative to traditional 2.5D HBM and direct-stacked HBM-on-GPU, using simulations to define the thermal design envelope for various stack heights.
Semiconductor Engineering · 27 October 2026 · Read the original →
Self-Heating And Parasitics Challenge Multi-Tier CFET Scaling (TU Munich, IIT Kanpur)
A study by TU Munich and IIT Kanpur analyzes the impact of self-heating effects (SHEs) and parasitics in multi-tier Complementary Field-Effect Transistor (CFET) designs. The research specifically examines how vertically stacking multiple nanosheet devices introduces middle-of-line (MOL) and back-end-of-line (BEOL) induced parasitic challenges.
Semiconductor Engineering · 27 October 2026 · Read the original →
The AI Packaging Challenge: 5 Reasons Panels Are Gaining Momentum
Lam Research is advocating for panel-level processing as a solution to AI chip packages exceeding traditional wafer size limits. The shift involves moving from circular silicon wafers to large-format rectangular panels to scale advanced packaging for the next decade of high-performance compute.
EE Times · 27 October 2026 · Read the original →
Micron plans to use wafer bonding and High-NA EUV for next-gen DRAM
Micron has detailed its roadmap for next-generation DRAM, which will utilize wafer bonding and High-NA EUV lithography. This strategy involves separately manufacturing the CMOS logic and the memory array before bonding them, allowing for more aggressive scaling and optimized process nodes for each component.
Tom's Hardware · 27 October 2026 · Read the original →
Two Dollars and a Test Suite
An analysis of the Arm-Qualcomm legal verdict reveals that the core dispute centers on the control of compliance test suites. The ruling suggests that the right to ship a custom silicon implementation is contingent upon passing specific IP-vendor-controlled tests to ensure architectural compatibility.
EE Times · 27 October 2026 · Read the original →
Cadence reveals Maestro, fine-tuned AI models built to run specialized chip design agents, with early access coming for 'select customers' in Q4 2026
Cadence has announced Maestro, a suite of fine-tuned AI models designed to act as specialized agents for chip design tasks. The system is intended to automate complex EDA workflows, with early access for select customers beginning in Q4 2026.
Tom's Hardware · 27 October 2026 · Read the original →
Intel Advances Arizona Fab 62, High-NA EUV
Intel has begun installing production tools at its Fab 62 in Chandler, Arizona, part of a $32 billion investment, with operations expected by late 2028. The company also confirmed the integration of High-NA EUV lithography into high-volume manufacturing.
digitimes.com · 27 October 2026 · Read the original →
Physics remains uncooperative, but at least the roadmaps stay optimistic.
7 stories, every Tuesday
Published here every week. Follow by RSS to get it as it lands.