Breaking The Memory Bottleneck In AI And HPC SoCs

Embedded memory IP adapted for RibbonFET and PowerVia designs

By GPU Data Hub Desk · GPU Data Hub·Editorial standards·19 Oct 2026, 07:03

Chip architects are refining embedded memory intellectual property to overcome data bottlenecks in high-performance computing and artificial intelligence processors. The design work centres on tailoring memory structures for advanced semiconductor processes, specifically RibbonFET and backside power delivery via PowerVia. These adjustments aim to ensure memory subsystems keep pace with accelerated logic advancements.

Original source

This summary was written by the GPU Data Hub desk from reporting published by Semiconductor Engineering on 19 Oct 2026, 07:03. Read the full article at the original publisher.

Read Original Source →

Topics

embedded memory SoCs RibbonFET PowerVia HPC

LinkedIn X

More from GPU Data Hub

Latest →

Why Physical AI Is So Hard To Engineer

Engineers developing physical artificial intelligence systems face complex silicon design hurdles due to the convergence of distinct hardware domains. Autonomous platforms require the simultaneous integration of compute logic, sensing components, and mechanical actuators on tight performance budgets. Balancing these varied requirements poses unique complications for next-generation edge device architecture.

Semiconductors·2h ago·Semiconductor Engineering

Shift-Left DE-Via

A shift-left methodology for design-rule-aware via placement aims to produce signoff-ready physical layouts with reduced risk of electrical shorts. By integrating Calibre-aware checks earlier into the design flow, engineers can detect and resolve fabrication vulnerabilities sooner. The technique seeks to improve manufacturing yield and layout reliability in advanced node implementation.

Semiconductors·2h ago·Semiconductor Engineering

China's CXMT sees shares drop over 6% as investors look past chipmaker's profit turnaround

Shares in Chinese memory chip manufacturer CXMT dropped by more than 6% following the release of its financial guidance for the initial nine months of 2026. Despite projections indicating an operational turnaround, equity investors appeared cautious regarding the company's prospective trajectory.

Semiconductors·7h ago·CNBC Technology

Nanoelectromechanical Switches Target More Efficient Embedded DRAM (UC Berkeley, Samsung, Intel)

Researchers from Samsung Electronics, Intel, and UC Berkeley have outlined an embedded dynamic memory architecture utilising nanoelectromechanical switches. The collaborative work seeks to overcome data retention limits and tackle persistent scaling barriers associated with legacy on-chip memory structures.

Semiconductors·10h ago·Semiconductor Engineering

A Composable Thermal Solver For Heterogeneous 2.5D/3D-IC Packages (Cadence, Notre Dame)

Cadence Design Systems and the University of Notre Dame have published details on an AI-assisted iterative solver intended for multi-die thermal simulation. The framework aims to drastically reduce the compute runtime traditionally required to model heat transfer across intricate 2.5D and 3D integrated packages.

Semiconductors·10h ago·Semiconductor Engineering

Thermal Side Channels Create New Security Risks For Autonomous Vehicles (Missouri S&T)

Academics from the Missouri University of Science and Technology have compiled a study detailing how thermal side channels pose emerging security threats to self-driving systems. The research categorises multiple existing works across areas including telemetry leakage, model extraction, and sensor perception interference. The findings provide a framework assessing these hardware vulnerabilities based on factors such as required privileges and deployment feasibility.

Semiconductors·11h ago·Semiconductor Engineering