# Semiconductor Engineering

Deep Insights For Chip Engineers

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## AI Accelerators Drive Changes In Power Delivery Networks

DevFeed: [AI Accelerators Drive Changes In Power Delivery Networks](<https://devfeed.tech/articles/ai-accelerators-drive-changes-in-power-delivery-networks-143949.md>)

Original publisher: [Read original article](<https://semiengineering.com/ai-accelerators-drive-changes-in-power-delivery-networks/>)

Author: Laura Peters

Published: 2026-11-19T08:05:17Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [systems](<https://devfeed.tech/topics/systems.md>), [tsmc](<https://devfeed.tech/topics/tsmc.md>)

Tags: [800v-dc](<https://devfeed.tech/tags/800v-dc.md>), [ai-accelerators](<https://devfeed.tech/tags/ai-accelerators.md>), [amkor](<https://devfeed.tech/tags/amkor.md>), [ase](<https://devfeed.tech/tags/ase.md>), [backside-power-delivery](<https://devfeed.tech/tags/backside-power-delivery.md>), [backside-power-delivery-networks](<https://devfeed.tech/tags/backside-power-delivery-networks.md>), [chiplets](<https://devfeed.tech/tags/chiplets.md>), [empower](<https://devfeed.tech/tags/empower.md>), [intel-foundry](<https://devfeed.tech/tags/intel-foundry.md>), [manufacturing-packaging-and-materials](<https://devfeed.tech/tags/manufacturing-packaging-and-materials.md>), [marvell](<https://devfeed.tech/tags/marvell.md>), [osats](<https://devfeed.tech/tags/osats.md>), [power-delivery-networks](<https://devfeed.tech/tags/power-delivery-networks.md>), [power-loss](<https://devfeed.tech/tags/power-loss.md>), [proteantecs](<https://devfeed.tech/tags/proteantecs.md>), [samsung](<https://devfeed.tech/tags/samsung.md>), [synopsys](<https://devfeed.tech/tags/synopsys.md>), [top-stories](<https://devfeed.tech/tags/top-stories.md>), [tsmc](<https://devfeed.tech/tags/tsmc.md>), [umc](<https://devfeed.tech/tags/umc.md>), [voltage-droop](<https://devfeed.tech/tags/voltage-droop.md>)

### AI overview

AI accelerators' low operating voltages and currents approaching or exceeding 1,000 A make power delivery a performance and efficiency constraint: even milliohm-level resistance causes substantial voltage drop, while workload transients prompt power-wasting voltage guardbands. The article describes moving regulation closer to the processor--from under the PCB to the package or interposer--using distributed integrated voltage regulators, compact capacitors, and inductors. Shorter paths and faster control loops can reduce losses and improve voltage response, but integration adds thermal, stress, noise, packaging, and manufacturing challenges. The 800 V DC data-center approach steps power down through intermediate voltages toward the chip. Solutions vary by provider, and require early co-design across dies, packages, boards, power components, and thermal paths. The excerpt also emphasizes simulation and supply-chain model exchange to assess power integrity before designs are finalized.

### Source excerpt

Moving voltage regulators closer to processors can reduce power loss, but adds challenges for advanced packaging, thermal management, and system-level design. The post AI Accelerators Drive Changes In Power Delivery Networks appeared first on Semiconductor Engineering.

## From Metrology To Autonomy: Scaling Yield, Quality, And Innovation

DevFeed: [From Metrology To Autonomy: Scaling Yield, Quality, And Innovation](<https://devfeed.tech/articles/from-metrology-to-autonomy-scaling-yield-quality-and-innovation-143953.md>)

Original publisher: [Read original article](<https://semiengineering.com/from-metrology-to-autonomy-scaling-yield-quality-and-innovation/>)

Author: Lori Scott

Published: 2026-11-19T08:02:31Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [AI Integration Strategies](<https://devfeed.tech/topics/ai-integration-strategies.md>), [AI-Ready Data](<https://devfeed.tech/topics/ai-ready-data.md>)

Tags: [2-5d](<https://devfeed.tech/tags/2-5d.md>), [2026](<https://devfeed.tech/tags/2026.md>), [3d](<https://devfeed.tech/tags/3d.md>), [advanced-packaging](<https://devfeed.tech/tags/advanced-packaging.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-workflow](<https://devfeed.tech/tags/agentic-workflow.md>), [ai](<https://devfeed.tech/tags/ai.md>), [analytics](<https://devfeed.tech/tags/analytics.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [automatic-defect-classification](<https://devfeed.tech/tags/automatic-defect-classification.md>), [autonomy](<https://devfeed.tech/tags/autonomy.md>), [blogs-md](<https://devfeed.tech/tags/blogs-md.md>), [data](<https://devfeed.tech/tags/data.md>), [foundry-focus](<https://devfeed.tech/tags/foundry-focus.md>), [innovation](<https://devfeed.tech/tags/innovation.md>), [intel-foundry](<https://devfeed.tech/tags/intel-foundry.md>), [intelligence](<https://devfeed.tech/tags/intelligence.md>), [manufacturing](<https://devfeed.tech/tags/manufacturing.md>), [manufacturing-packaging-and-materials](<https://devfeed.tech/tags/manufacturing-packaging-and-materials.md>), [metrology](<https://devfeed.tech/tags/metrology.md>), [quality](<https://devfeed.tech/tags/quality.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [semiconductor](<https://devfeed.tech/tags/semiconductor.md>), [semiconductor-manufacturing](<https://devfeed.tech/tags/semiconductor-manufacturing.md>), [stack](<https://devfeed.tech/tags/stack.md>), [trusted-data](<https://devfeed.tech/tags/trusted-data.md>), [yield](<https://devfeed.tech/tags/yield.md>)

### AI overview

Intel Foundry's account argues that factory autonomy depends on trusted measurements, interoperable systems, and advanced test infrastructure alongside AI. Continuous metrology and predictive process control are intended to catch yield risks earlier across wafer fabrication and increasingly complex packaging. Advanced Chiplet Test services combine wafer-to-system coverage, known-good die selection, predictive die classification, optical testing, and closed-loop yield analytics to inform manufacturing. For defect classification, a human-in-the-loop agentic workflow assists with log analysis, knowledge retrieval, root-cause investigation, and recipe review; human approval checkpoints preserve oversight and auditability. Research also combines AI with heat-transfer physics to predict device temperatures using less training data. The account identifies vendor-locked automation and data silos as limits, and points to hardware-independent architectures, connected data, edge computing, and shared industry standards as foundations for scaling. These are described as approaches and ongoing research, not quantified outcomes.

### Source excerpt

Operationalizing intelligence across the manufacturing stack begins with trusted data. The post From Metrology To Autonomy: Scaling Yield, Quality, And Innovation appeared first on Semiconductor Engineering.

## Detachable Fiber Connectors Pose New Challenges For Photonics

DevFeed: [Detachable Fiber Connectors Pose New Challenges For Photonics](<https://devfeed.tech/articles/detachable-fiber-connectors-pose-new-challenges-for-photonics-143952.md>)

Original publisher: [Read original article](<https://semiengineering.com/detachable-fiber-connectors-pose-new-challenges-for-photonics/>)

Author: Bryon Moyer

Published: 2026-11-19T08:01:36Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [communications](<https://devfeed.tech/topics/communications.md>), [Chip design](<https://devfeed.tech/topics/chip-design.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [3](<https://devfeed.tech/tags/3.md>), [about](<https://devfeed.tech/tags/about.md>), [aim-photonics](<https://devfeed.tech/tags/aim-photonics.md>), [amkor](<https://devfeed.tech/tags/amkor.md>), [ayar-labs](<https://devfeed.tech/tags/ayar-labs.md>), [but](<https://devfeed.tech/tags/but.md>), [challenges](<https://devfeed.tech/tags/challenges.md>), [co-packaged-optics](<https://devfeed.tech/tags/co-packaged-optics.md>), [connectors](<https://devfeed.tech/tags/connectors.md>), [cpo](<https://devfeed.tech/tags/cpo.md>), [designs](<https://devfeed.tech/tags/designs.md>), [detachable](<https://devfeed.tech/tags/detachable.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [fiber](<https://devfeed.tech/tags/fiber.md>), [fiber-attach](<https://devfeed.tech/tags/fiber-attach.md>), [first](<https://devfeed.tech/tags/first.md>), [intel-foundry](<https://devfeed.tech/tags/intel-foundry.md>), [lightmatter](<https://devfeed.tech/tags/lightmatter.md>), [manufacturing](<https://devfeed.tech/tags/manufacturing.md>), [manufacturing-packaging-and-materials](<https://devfeed.tech/tags/manufacturing-packaging-and-materials.md>), [mitsubishi-chemical](<https://devfeed.tech/tags/mitsubishi-chemical.md>), [multiple](<https://devfeed.tech/tags/multiple.md>), [near-packaged-optics](<https://devfeed.tech/tags/near-packaged-optics.md>), [npo](<https://devfeed.tech/tags/npo.md>), [ny-creates](<https://devfeed.tech/tags/ny-creates.md>), [photonics](<https://devfeed.tech/tags/photonics.md>), [semiconductor](<https://devfeed.tech/tags/semiconductor.md>), [semiconductor-engineering](<https://devfeed.tech/tags/semiconductor-engineering.md>), [senko](<https://devfeed.tech/tags/senko.md>), [standardization](<https://devfeed.tech/tags/standardization.md>), [stl](<https://devfeed.tech/tags/stl.md>), [synopsys](<https://devfeed.tech/tags/synopsys.md>), [top-stories](<https://devfeed.tech/tags/top-stories.md>), [university-of-toronto](<https://devfeed.tech/tags/university-of-toronto.md>), [volume](<https://devfeed.tech/tags/volume.md>)

### AI overview

As optics move closer to processors and switches, detachable fiber connections could reduce long, power-hungry electrical links while making photonics packages easier to service. Current designs include edge or vertical coupling, pigtails, and package-mounted connectors; vertical coupling can use more chip area but may conflict with thermal hardware, while pigtails complicate tray handling and automated assembly. Connections also need precise optical alignment, reliable latching, and test methods suited to production. A reported Senko test connector achieved 10,000 matings with little variation in connection loss without active alignment. Permanently fused fibers impair replacement when a chip or fiber fails. The excerpt describes multiple designs but no standardized interface or proven high-volume approach; compatibility with automated handling, repeatable low-cost termination, testing, and differing customer layouts remain challenges.

### Source excerpt

Multiple designs are emerging for CPO and NPO, but standardization and high-volume manufacturing are still not ready. The post Detachable Fiber Connectors Pose New Challenges For Photonics appeared first on Semiconductor Engineering.

## Can AI Agents Really Shift Chip Design Left?

DevFeed: [Can AI Agents Really Shift Chip Design Left?](<https://devfeed.tech/articles/can-ai-agents-really-shift-chip-design-left-140410.md>)

Original publisher: [Read original article](<https://semiengineering.com/can-ai-agents-really-shift-chip-design-left/>)

Author: Ed Sperling

Published: 2026-11-18T08:01:53Z

Content type: news

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>), [knowledge-engineering](<https://devfeed.tech/topics/knowledge-engineering.md>), [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>)

Tags: [adoption](<https://devfeed.tech/tags/adoption.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-companies](<https://devfeed.tech/tags/ai-companies.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [automate](<https://devfeed.tech/tags/automate.md>), [axiomise](<https://devfeed.tech/tags/axiomise.md>), [chip](<https://devfeed.tech/tags/chip.md>), [chip-design](<https://devfeed.tech/tags/chip-design.md>), [decades](<https://devfeed.tech/tags/decades.md>), [eda](<https://devfeed.tech/tags/eda.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [flow](<https://devfeed.tech/tags/flow.md>), [hiveware](<https://devfeed.tech/tags/hiveware.md>), [integration](<https://devfeed.tech/tags/integration.md>), [round-tables](<https://devfeed.tech/tags/round-tables.md>), [semiconductor](<https://devfeed.tech/tags/semiconductor.md>), [semiconductor-engineering](<https://devfeed.tech/tags/semiconductor-engineering.md>), [shift-left](<https://devfeed.tech/tags/shift-left.md>), [silvaco](<https://devfeed.tech/tags/silvaco.md>), [synopsys](<https://devfeed.tech/tags/synopsys.md>), [system-level-design](<https://devfeed.tech/tags/system-level-design.md>), [tools](<https://devfeed.tech/tags/tools.md>), [top-stories](<https://devfeed.tech/tags/top-stories.md>)

### AI overview

AI agents could automate more chip design work through a hierarchy of specialized agents and orchestration, but how to integrate them into a coherent flow remains unsettled. Panelists say agents may help with many challenges, while capturing decades of specialized engineering skills is still a gap; early manufacturability and multi-physics effects also need checking. Training and operation depend on access to proprietary design data: Samsung reportedly used its own LLM with company data kept in-house, while Cognichip hired chip designers and built chips to train its models. Token costs may limit adoption, especially for smaller companies. These are excerpts from a panel discussion, and the speakers differ on how quickly methodologies and simulation practices may change.

### Source excerpt

Agentic AI promises to automate more of the chip flow, but integration with EDA tools and capturing decades of engineering expertise are non-trivial. The post Can AI Agents Really Shift Chip Design Left? appeared first on Semiconductor Engineering.

## Blog Review: Nov. 18

DevFeed: [Blog Review: Nov. 18](<https://devfeed.tech/articles/blog-review-nov-18-140409.md>)

Original publisher: [Read original article](<https://semiengineering.com/blog-review-nov-18-3/>)

Author: Jesse Allen

Published: 2026-11-18T08:01:49Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [trust](<https://devfeed.tech/topics/trust.md>), [Chip design](<https://devfeed.tech/topics/chip-design.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [pcie](<https://devfeed.tech/topics/pcie.md>), [human computer interaction](<https://devfeed.tech/topics/human-computer-interaction.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-data-center](<https://devfeed.tech/tags/ai-data-center.md>), [arm](<https://devfeed.tech/tags/arm.md>), [blog](<https://devfeed.tech/tags/blog.md>), [blog-reviews](<https://devfeed.tech/tags/blog-reviews.md>), [business](<https://devfeed.tech/tags/business.md>), [cadence](<https://devfeed.tech/tags/cadence.md>), [co-design](<https://devfeed.tech/tags/co-design.md>), [eu](<https://devfeed.tech/tags/eu.md>), [keysight](<https://devfeed.tech/tags/keysight.md>), [news](<https://devfeed.tech/tags/news.md>), [on-device](<https://devfeed.tech/tags/on-device.md>), [pcie](<https://devfeed.tech/tags/pcie.md>), [rack-scale](<https://devfeed.tech/tags/rack-scale.md>), [review](<https://devfeed.tech/tags/review.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [semi](<https://devfeed.tech/tags/semi.md>), [semiconductor](<https://devfeed.tech/tags/semiconductor.md>), [siemens](<https://devfeed.tech/tags/siemens.md>), [synopsys](<https://devfeed.tech/tags/synopsys.md>), [system-level-design](<https://devfeed.tech/tags/system-level-design.md>), [trends](<https://devfeed.tech/tags/trends.md>), [trust](<https://devfeed.tech/tags/trust.md>)

### AI overview

The review highlights system-level challenges in scaling AI hardware: securing sensitive data, models, keys, and workloads across PCIe; co-designing silicon, interfaces, memory, cooling, and power for efficient AI racks; and managing the complexity of heterogeneous architectures, 3D-ICs, and chiplets. Other features cover AI data center technologies for training and inference, gaze-based smart-glasses interactions using on-device AI for fast sensor-dependent tasks while larger workloads run where more compute is available, and European semiconductor trends and collaboration. The accompanying low-power newsletter adds perspectives on shifting memory pressure, using silicon evidence to guide hardware tools and decisions, voltage scaling limits under energy and thermal constraints, memory-hierarchy sizing, and 8-bit floating-point formats.

### Source excerpt

PCIe and trust; rack-scale co-design; heterogeneous scaling; AI data center primer; on-device balancing; EU semi trends. The post Blog Review: Nov. 18 appeared first on Semiconductor Engineering.

## Defect-Aware Chiplet Matching For Heterogeneous Integration Yield (ASU, Ampere)

DevFeed: [Defect-Aware Chiplet Matching For Heterogeneous Integration Yield (ASU, Ampere)](<https://devfeed.tech/articles/defect-aware-chiplet-matching-for-heterogeneous-integration-yield-asu-ampere-138403.md>)

Original publisher: [Read original article](<https://semiengineering.com/defect-aware-chiplet-matching-for-heterogeneous-integration-yield-asu-ampere/>)

Author: Technical Paper Link

Published: 2026-11-17T17:44:39Z

Content type: news

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [advanced-packaging](<https://devfeed.tech/tags/advanced-packaging.md>), [ampere](<https://devfeed.tech/tags/ampere.md>), [ampere-computing](<https://devfeed.tech/tags/ampere-computing.md>), [arizona-state-university](<https://devfeed.tech/tags/arizona-state-university.md>), [chiplet](<https://devfeed.tech/tags/chiplet.md>), [chiplet-matching](<https://devfeed.tech/tags/chiplet-matching.md>), [chiplets](<https://devfeed.tech/tags/chiplets.md>), [d2d](<https://devfeed.tech/tags/d2d.md>), [defect](<https://devfeed.tech/tags/defect.md>), [defects](<https://devfeed.tech/tags/defects.md>), [die-to-die-interconnects](<https://devfeed.tech/tags/die-to-die-interconnects.md>), [heterogeneous-integration](<https://devfeed.tech/tags/heterogeneous-integration.md>), [interconnect](<https://devfeed.tech/tags/interconnect.md>), [manufacturing](<https://devfeed.tech/tags/manufacturing.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [packaging](<https://devfeed.tech/tags/packaging.md>), [published](<https://devfeed.tech/tags/published.md>), [repair](<https://devfeed.tech/tags/repair.md>), [soc](<https://devfeed.tech/tags/soc.md>), [system-on-chip](<https://devfeed.tech/tags/system-on-chip.md>), [technical-papers](<https://devfeed.tech/tags/technical-papers.md>), [test-measurement-analytics-tech-papers](<https://devfeed.tech/tags/test-measurement-analytics-tech-papers.md>), [topology](<https://devfeed.tech/tags/topology.md>), [wafer-level-test](<https://devfeed.tech/tags/wafer-level-test.md>), [yield](<https://devfeed.tech/tags/yield.md>)

### AI overview

DT-MATCH aims to maximize functional SoCs assembled from defective but repairable chiplets while accounting for lane-level defects, limited spares, and die-to-die topology and compatibility constraints. Its ILP approach uses batching to approximate the exact formulation and achieves better solution quality in experiments on industrial designs. Its GRASP approach runs faster without guaranteeing optimal yield, while producing solution quality reported as comparable to ILP. Both use preprocessing to reduce problem size, offering a practical tradeoff between solution quality and runtime.

### Source excerpt

Researchers at Arizona State University and Ampere Computing published "DT-MATCH: Defect- and Topology-Aware Chiplet Matching for Yield Optimization in Heterogeneous Integration." Abstract "Heterogeneous integration (HI) and advanced packaging enable chiplet-based system-on-chip (SoC) designs; however, they also introduce new yield challenges arising from manufacturing defects and complex die-to-die (D2D) interconnect constraints. In particular, lane-level defects, limited... " read more The post Defect-Aware Chiplet Matching For Heterogeneous Integration Yield (ASU, Ampere) appeared first on Semiconductor Engineering.

## Backside Clock Mesh Reduces Skew And Power In Backside Power Delivery (KAIST)

DevFeed: [Backside Clock Mesh Reduces Skew And Power In Backside Power Delivery (KAIST)](<https://devfeed.tech/articles/backside-clock-mesh-reduces-skew-and-power-in-backside-power-delivery-kaist-138402.md>)

Original publisher: [Read original article](<https://semiengineering.com/backside-clock-mesh-reduces-skew-and-power-in-backside-power-delivery-kaist/>)

Author: Technical Paper Link

Published: 2026-11-17T17:35:37Z

Content type: news

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>)

Tags: [architectures](<https://devfeed.tech/tags/architectures.md>), [backside-clock-mesh](<https://devfeed.tech/tags/backside-clock-mesh.md>), [backside-power-delivery](<https://devfeed.tech/tags/backside-power-delivery.md>), [bspdn](<https://devfeed.tech/tags/bspdn.md>), [clock](<https://devfeed.tech/tags/clock.md>), [clock-distribution-network](<https://devfeed.tech/tags/clock-distribution-network.md>), [clock-mesh](<https://devfeed.tech/tags/clock-mesh.md>), [clock-skew](<https://devfeed.tech/tags/clock-skew.md>), [design-verification](<https://devfeed.tech/tags/design-verification.md>), [ir-drop](<https://devfeed.tech/tags/ir-drop.md>), [kaist](<https://devfeed.tech/tags/kaist.md>), [korea-advanced-institute-of-science-and-technology](<https://devfeed.tech/tags/korea-advanced-institute-of-science-and-technology.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [nano-tsv](<https://devfeed.tech/tags/nano-tsv.md>), [physical-design](<https://devfeed.tech/tags/physical-design.md>), [power-delivery](<https://devfeed.tech/tags/power-delivery.md>), [power-performance](<https://devfeed.tech/tags/power-performance.md>), [published](<https://devfeed.tech/tags/published.md>)

### AI overview

KAIST researchers propose placing the clock mesh on backside metal layers in a backside power delivery network architecture, while keeping the rest of the clock distribution network on the frontside. Because the backside metal is shared with the power network, mesh receiver placement must account for keep-out zones between clock and power nTSVs. The authors say a potential-field method solves this placement problem. The excerpt provides no quantitative results on skew or power, despite the headline's claim of reductions.

### Source excerpt

Researchers at Korea Advanced Institute of Science and Technology (KAIST) published "Clock Distribution Network with Backside Clock Mesh in BSPDN Architecture." Abstract Excerpt "We propose a clock distribution network (CDN) in which the clock mesh is located on backside metal (BM) layers, while the rest of the network remains on the frontside. Since the backside... " read more The post Backside Clock Mesh Reduces Skew And Power In Backside Power Delivery (KAIST) appeared first on Semiconductor Engineering.

## Chip Industry Technical Paper Roundup: Nov. 17

DevFeed: [Chip Industry Technical Paper Roundup: Nov. 17](<https://devfeed.tech/articles/chip-industry-technical-paper-roundup-nov-17-137213.md>)

Original publisher: [Read original article](<https://semiengineering.com/chip-industry-technical-paper-roundup-nov-17/>)

Author: Linda Christensen

Published: 2026-11-17T08:01:08Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>)

Tags: [2d](<https://devfeed.tech/tags/2d.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-chip-smuggling](<https://devfeed.tech/tags/ai-chip-smuggling.md>), [cache](<https://devfeed.tech/tags/cache.md>), [chonnam-national-university](<https://devfeed.tech/tags/chonnam-national-university.md>), [csiro](<https://devfeed.tech/tags/csiro.md>), [e-beam](<https://devfeed.tech/tags/e-beam.md>), [ecc](<https://devfeed.tech/tags/ecc.md>), [ferroelectric-capacitors](<https://devfeed.tech/tags/ferroelectric-capacitors.md>), [georgia-tech](<https://devfeed.tech/tags/georgia-tech.md>), [hanyang-university](<https://devfeed.tech/tags/hanyang-university.md>), [hbm4](<https://devfeed.tech/tags/hbm4.md>), [hzo](<https://devfeed.tech/tags/hzo.md>), [low-power](<https://devfeed.tech/tags/low-power.md>), [max-planck-institute-for-security-and-privacy](<https://devfeed.tech/tags/max-planck-institute-for-security-and-privacy.md>), [metal-semiconductor-contacts](<https://devfeed.tech/tags/metal-semiconductor-contacts.md>), [national-taiwan-university](<https://devfeed.tech/tags/national-taiwan-university.md>), [national-university-of-singapore](<https://devfeed.tech/tags/national-university-of-singapore.md>), [nexstrom](<https://devfeed.tech/tags/nexstrom.md>), [nycu](<https://devfeed.tech/tags/nycu.md>), [partitioning](<https://devfeed.tech/tags/partitioning.md>), [postech](<https://devfeed.tech/tags/postech.md>), [protection](<https://devfeed.tech/tags/protection.md>), [research](<https://devfeed.tech/tags/research.md>), [research-lphp](<https://devfeed.tech/tags/research-lphp.md>), [ruhr-university-bochum](<https://devfeed.tech/tags/ruhr-university-bochum.md>), [rwth-aachen-university](<https://devfeed.tech/tags/rwth-aachen-university.md>), [samsung-electronics](<https://devfeed.tech/tags/samsung-electronics.md>), [security](<https://devfeed.tech/tags/security.md>), [semiconductor-research](<https://devfeed.tech/tags/semiconductor-research.md>), [seoul-national-university](<https://devfeed.tech/tags/seoul-national-university.md>), [singapore-university-of-technology-and-design](<https://devfeed.tech/tags/singapore-university-of-technology-and-design.md>), [texas-a-m-university](<https://devfeed.tech/tags/texas-a-m-university.md>), [tsmc](<https://devfeed.tech/tags/tsmc.md>), [uc-riverside](<https://devfeed.tech/tags/uc-riverside.md>), [university-of-florida](<https://devfeed.tech/tags/university-of-florida.md>), [university-of-hong-kong](<https://devfeed.tech/tags/university-of-hong-kong.md>), [unsw](<https://devfeed.tech/tags/unsw.md>)

### AI overview

The roundup lists recent semiconductor research on HBM4 controller partitioning, high-κ dielectrics for low-power 2D CFETs, hardware-based location checks to address AI chip smuggling, embedded ECC for AI weight protection, e-beam evaporation for metal-semiconductor contacts, sub-volt HZO ferroelectric capacitors, and randomized-cache hardware security. The excerpt names these topics but provides no methods or results.

### Source excerpt

HBM4 controller partitioning; single-crystal high-κ dielectrics for low-power 2D CFETs; HW-based location checks for AI chip smuggling; embedded ECC for AI weight protection; optimized e-beam evaporation for metal-semiconductor contacts; sub-volt HZO ferroelectric capacitors; randomized-cache HW security. The post Chip Industry Technical Paper Roundup: Nov. 17 appeared first on Semiconductor Engineering.

## Reducing High-Bandwidth Flash Latency In GPU Systems (Georgia Tech, Nvidia, Stanford)

DevFeed: [Reducing High-Bandwidth Flash Latency In GPU Systems (Georgia Tech, Nvidia, Stanford)](<https://devfeed.tech/articles/reducing-high-bandwidth-flash-latency-in-gpu-systems-georgia-tech-nvidia-stanford-136674.md>)

Original publisher: [Read original article](<https://semiengineering.com/reducing-high-bandwidth-flash-latency-in-gpu-systems-georgia-tech-nvidia-stanford/>)

Author: Technical Paper Link

Published: 2026-11-16T23:52:56Z

Content type: news

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [performance-engineering](<https://devfeed.tech/topics/performance-engineering.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [systems](<https://devfeed.tech/topics/systems.md>), [NVIDIA Research](<https://devfeed.tech/topics/nvidia-research.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-ml-dl](<https://devfeed.tech/tags/ai-ml-dl.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [batch](<https://devfeed.tech/tags/batch.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [dram](<https://devfeed.tech/tags/dram.md>), [first](<https://devfeed.tech/tags/first.md>), [flash](<https://devfeed.tech/tags/flash.md>), [georgia](<https://devfeed.tech/tags/georgia.md>), [georgia-tech](<https://devfeed.tech/tags/georgia-tech.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [hardware-software-co-design](<https://devfeed.tech/tags/hardware-software-co-design.md>), [hbf](<https://devfeed.tech/tags/hbf.md>), [hbm](<https://devfeed.tech/tags/hbm.md>), [high-bandwidth-flash](<https://devfeed.tech/tags/high-bandwidth-flash.md>), [inference](<https://devfeed.tech/tags/inference.md>), [language](<https://devfeed.tech/tags/language.md>), [latency](<https://devfeed.tech/tags/latency.md>), [llm-inference](<https://devfeed.tech/tags/llm-inference.md>), [memory](<https://devfeed.tech/tags/memory.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [prefetching](<https://devfeed.tech/tags/prefetching.md>), [stanford](<https://devfeed.tech/tags/stanford.md>), [stanford-university](<https://devfeed.tech/tags/stanford-university.md>), [systems](<https://devfeed.tech/tags/systems.md>), [tech](<https://devfeed.tech/tags/tech.md>), [technical-papers](<https://devfeed.tech/tags/technical-papers.md>)

### AI overview

SlipStream uses a demand-paced producer-consumer pipeline to hide High-Bandwidth Flash (HBF) latency for GPU inference while limiting tracking and buffer overhead. Software profiles the repeatable weight-page demand order of data-independent LLM kernels and inserts a staging warp that replays it with non-blocking STAGE instructions. Hardware retires each staging entry when demand first accesses its 4 KB page, rather than when the HBF read returns, and allocates buffer slots only on page return. Across seven dense models spanning 16-800 GB, a 40 MB buffer reached 92.4% of ideal HBF performance with HBM-like latency; H3 reached 43.1% with a 320 MB buffer. SlipStream's 16 MB buffer performed within 4% of its 40 MB configuration, and it reached 91.2% of ideal performance on Mixture-of-Experts models. The excerpt says buffer sizing is independent of HBF latency and workload.

### Source excerpt

Researchers at Georgia Tech, NVIDIA Research, and Stanford University published "SlipStream: Enabling HBF for GPU Systems by Mitigating the Latency and SRAM Overheads." Abstract Excerpt "Large language model (LLM) inference demands ever more memory capacity for larger models and batch sizes. High-Bandwidth Flash (HBF) replaces the DRAM dies of HBM with NAND flash and provides... " read more The post Reducing High-Bandwidth Flash Latency In GPU Systems (Georgia Tech, Nvidia, Stanford) appeared first on Semiconductor Engineering.

## Shifting HBM4 Controller Functions Closer To Memory (UC Riverside)

DevFeed: [Shifting HBM4 Controller Functions Closer To Memory (UC Riverside)](<https://devfeed.tech/articles/shifting-hbm4-controller-functions-closer-to-memory-uc-riverside-133491.md>)

Original publisher: [Read original article](<https://semiengineering.com/shifting-hbm4-controller-functions-closer-to-memory-uc-riverside/>)

Author: Technical Paper Link

Published: 2026-11-13T21:53:30Z

Content type: news

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Memory Heirarchy](<https://devfeed.tech/topics/memory-heirarchy.md>)

Tags: [2-5d-integration](<https://devfeed.tech/tags/2-5d-integration.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [advanced-packaging](<https://devfeed.tech/tags/advanced-packaging.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [base](<https://devfeed.tech/tags/base.md>), [base-die](<https://devfeed.tech/tags/base-die.md>), [control](<https://devfeed.tech/tags/control.md>), [controller](<https://devfeed.tech/tags/controller.md>), [d2d](<https://devfeed.tech/tags/d2d.md>), [die-to-die-interface](<https://devfeed.tech/tags/die-to-die-interface.md>), [dram](<https://devfeed.tech/tags/dram.md>), [functions](<https://devfeed.tech/tags/functions.md>), [hbm](<https://devfeed.tech/tags/hbm.md>), [hbm4](<https://devfeed.tech/tags/hbm4.md>), [high-bandwidth-memory](<https://devfeed.tech/tags/high-bandwidth-memory.md>), [logic](<https://devfeed.tech/tags/logic.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-controller](<https://devfeed.tech/tags/memory-controller.md>), [paper](<https://devfeed.tech/tags/paper.md>), [partitioning](<https://devfeed.tech/tags/partitioning.md>), [process](<https://devfeed.tech/tags/process.md>), [published](<https://devfeed.tech/tags/published.md>), [stack](<https://devfeed.tech/tags/stack.md>), [technical](<https://devfeed.tech/tags/technical.md>), [technical-papers](<https://devfeed.tech/tags/technical-papers.md>), [uc](<https://devfeed.tech/tags/uc.md>), [university](<https://devfeed.tech/tags/university.md>), [university-of-california-riverside](<https://devfeed.tech/tags/university-of-california-riverside.md>)

### AI overview

The University of California, Riverside study examines how HBM4 controller functions could be split between the host and the memory stack's base die, which can use a logic process for more complex control functions. It presents a quantitative comparison of four placements, ranging from a fully host-side controller to a fully base-die controller, with intermediate designs moving maintenance or scheduling functions to the base die. The supplied excerpt describes the study's setup but does not report comparative results.

### Source excerpt

Tharunya Yogananda and Wantong Li of the University of California, Riverside published a technical paper titled "Beyond Host-Side Control: HBM Controller Partitioning Across Host and HBM Base Dies." Abstract Excerpt: "Recent generations of high bandwidth memory (HBM), such as HBM4, allow the base die of the memory stack to be manufactured using a logic process... " read more The post Shifting HBM4 Controller Functions Closer To Memory (UC Riverside) appeared first on Semiconductor Engineering.

## Chip Industry Week In Review

DevFeed: [Chip Industry Week In Review](<https://devfeed.tech/articles/chip-industry-week-in-review-131345.md>)

Original publisher: [Read original article](<https://semiengineering.com/chip-industry-week-in-review-164/>)

Author: The SE Staff

Published: 2026-11-13T08:01:34Z

Content type: news

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>)

Tags: [2d](<https://devfeed.tech/tags/2d.md>), [about](<https://devfeed.tech/tags/about.md>), [accellera](<https://devfeed.tech/tags/accellera.md>), [amd](<https://devfeed.tech/tags/amd.md>), [anthropic](<https://devfeed.tech/tags/anthropic.md>), [applied-materials](<https://devfeed.tech/tags/applied-materials.md>), [arm](<https://devfeed.tech/tags/arm.md>), [asra](<https://devfeed.tech/tags/asra.md>), [aws](<https://devfeed.tech/tags/aws.md>), [breakthrough](<https://devfeed.tech/tags/breakthrough.md>), [business](<https://devfeed.tech/tags/business.md>), [business-news](<https://devfeed.tech/tags/business-news.md>), [cadence](<https://devfeed.tech/tags/cadence.md>), [canon](<https://devfeed.tech/tags/canon.md>), [cea-leti](<https://devfeed.tech/tags/cea-leti.md>), [chip](<https://devfeed.tech/tags/chip.md>), [chiplet](<https://devfeed.tech/tags/chiplet.md>), [chips](<https://devfeed.tech/tags/chips.md>), [computing-news](<https://devfeed.tech/tags/computing-news.md>), [cpus](<https://devfeed.tech/tags/cpus.md>), [csis](<https://devfeed.tech/tags/csis.md>), [diamond-quanta](<https://devfeed.tech/tags/diamond-quanta.md>), [diraq](<https://devfeed.tech/tags/diraq.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [enisa](<https://devfeed.tech/tags/enisa.md>), [eu-chips-ju](<https://devfeed.tech/tags/eu-chips-ju.md>), [expedera](<https://devfeed.tech/tags/expedera.md>), [first](<https://devfeed.tech/tags/first.md>), [flow](<https://devfeed.tech/tags/flow.md>), [fractile](<https://devfeed.tech/tags/fractile.md>), [google](<https://devfeed.tech/tags/google.md>), [google-deepmind](<https://devfeed.tech/tags/google-deepmind.md>), [hummink](<https://devfeed.tech/tags/hummink.md>), [ibm](<https://devfeed.tech/tags/ibm.md>), [imec](<https://devfeed.tech/tags/imec.md>), [industry](<https://devfeed.tech/tags/industry.md>), [infineon](<https://devfeed.tech/tags/infineon.md>), [intel](<https://devfeed.tech/tags/intel.md>), [keysight](<https://devfeed.tech/tags/keysight.md>), [lam-research-foundation](<https://devfeed.tech/tags/lam-research-foundation.md>), [lattice-semiconductor](<https://devfeed.tech/tags/lattice-semiconductor.md>), [lauterbach](<https://devfeed.tech/tags/lauterbach.md>), [llnl](<https://devfeed.tech/tags/llnl.md>), [memory](<https://devfeed.tech/tags/memory.md>), [news](<https://devfeed.tech/tags/news.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [oversight](<https://devfeed.tech/tags/oversight.md>), [process](<https://devfeed.tech/tags/process.md>), [questions](<https://devfeed.tech/tags/questions.md>), [review](<https://devfeed.tech/tags/review.md>), [week-in-review](<https://devfeed.tech/tags/week-in-review.md>)

### AI overview

The week's announcements span manufacturing, packaging, AI hardware, security, and policy. Infineon reported functional silicon-carbide power devices made on 300mm wafers with existing silicon equipment; the larger wafers can hold 2.3 times as many chips as 200mm wafers. Imec completed baseline silicon-interposer process flows with TSVs and multiple routing layers, with a PDK planned for early 2027. UCIe 4.0 adds support for integrating multiple chiplets into an on-package system, including an on-package memory link and improved manageability, security, and reliability. Google's new TPU 8tx pairs its accelerator with AMD EPYC CPU hosts for secure, confidential inference; TPU 8i Max targets large context windows and higher rack compute density. The text also reports scrutiny of Nvidia's chip-diversion oversight and investment ties, alongside a proposed 5G-based system to verify accelerator locations. Other developments include new 2D-transistor dielectric approaches and rising memory costs, which Counterpoint says now account for 54% of a low-end phone's bill of materials.

### Source excerpt

Nvidia faces questions about security and oversight; 300mm SiC breakthrough; Siemens buys Lauterbach; chiplet standards; new memory bets; imec interposer process flow; TPUs with CPUs; 2D transistors; anti-smuggling tools for chips. The post Chip Industry Week In Review appeared first on Semiconductor Engineering.

## Rethinking Memory For AI Inference

DevFeed: [Rethinking Memory For AI Inference](<https://devfeed.tech/articles/rethinking-memory-for-ai-inference-130168.md>)

Original publisher: [Read original article](<https://semiengineering.com/rethinking-memory-for-ai-inference/>)

Author: Liz Allan

Published: 2026-11-12T08:11:46Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Local AI](<https://devfeed.tech/topics/local-ai.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [CPU Cache](<https://devfeed.tech/topics/cpu-cache.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [about](<https://devfeed.tech/tags/about.md>), [abstraction](<https://devfeed.tech/tags/abstraction.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-inference](<https://devfeed.tech/tags/ai-inference.md>), [arteris](<https://devfeed.tech/tags/arteris.md>), [ayar-labs](<https://devfeed.tech/tags/ayar-labs.md>), [baya-systems](<https://devfeed.tech/tags/baya-systems.md>), [boost](<https://devfeed.tech/tags/boost.md>), [but](<https://devfeed.tech/tags/but.md>), [cadence](<https://devfeed.tech/tags/cadence.md>), [challenges](<https://devfeed.tech/tags/challenges.md>), [compute-in-memory](<https://devfeed.tech/tags/compute-in-memory.md>), [data](<https://devfeed.tech/tags/data.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [expedera](<https://devfeed.tech/tags/expedera.md>), [in-memory-compute](<https://devfeed.tech/tags/in-memory-compute.md>), [inference](<https://devfeed.tech/tags/inference.md>), [keysight-eda](<https://devfeed.tech/tags/keysight-eda.md>), [low-power-high-performance](<https://devfeed.tech/tags/low-power-high-performance.md>), [lpddr](<https://devfeed.tech/tags/lpddr.md>), [memory](<https://devfeed.tech/tags/memory.md>), [near-memory](<https://devfeed.tech/tags/near-memory.md>), [performance](<https://devfeed.tech/tags/performance.md>), [power](<https://devfeed.tech/tags/power.md>), [processing](<https://devfeed.tech/tags/processing.md>), [processing-in-memory](<https://devfeed.tech/tags/processing-in-memory.md>), [processing-near-memory](<https://devfeed.tech/tags/processing-near-memory.md>), [quadric](<https://devfeed.tech/tags/quadric.md>), [rambus](<https://devfeed.tech/tags/rambus.md>), [reduce](<https://devfeed.tech/tags/reduce.md>), [semiconductor](<https://devfeed.tech/tags/semiconductor.md>), [semiconductor-engineering](<https://devfeed.tech/tags/semiconductor-engineering.md>), [sk-hynix](<https://devfeed.tech/tags/sk-hynix.md>), [software](<https://devfeed.tech/tags/software.md>), [thermal](<https://devfeed.tech/tags/thermal.md>), [top-stories](<https://devfeed.tech/tags/top-stories.md>), [vinci](<https://devfeed.tech/tags/vinci.md>)

### AI overview

Moving AI inference compute closer to memory can reduce data movement, latency, and energy use, which matters for edge devices with limited power. The best fit depends on the workload: PIM and near-memory approaches can help low-batch decoding, while large-batch decode and prefill may become activation-communication bottlenecks. LPDDR6-PIM targets low-power applications with simple MAC operations; custom HBM logic base dies allow broader functions but face thermal limits. Adoption also requires compilers, operating systems, memory controllers, and system protocols to support offloading and scheduling. Hardware and software complexity, manufacturing cost, workload matching, and heat remain constraints, and broad commercial use is still early.

### Source excerpt

Processing data closer to memory could reduce power and boost performance, but software, thermal, and architectural challenges remain. The post Rethinking Memory For AI Inference appeared first on Semiconductor Engineering.

## Beyond Bandwidth: AI Inference Is Reshaping The Memory Wall

DevFeed: [Beyond Bandwidth: AI Inference Is Reshaping The Memory Wall](<https://devfeed.tech/articles/beyond-bandwidth-ai-inference-is-reshaping-the-memory-wall-130163.md>)

Original publisher: [Read original article](<https://semiengineering.com/beyond-bandwidth-ai-inference-is-reshaping-the-memory-wall/>)

Author: Kushal Koolwal

Published: 2026-11-12T08:08:39Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [Memory Heirarchy](<https://devfeed.tech/topics/memory-heirarchy.md>), [performance-engineering](<https://devfeed.tech/topics/performance-engineering.md>)

Tags: [a-bit-about-memory](<https://devfeed.tech/tags/a-bit-about-memory.md>), [accelerators](<https://devfeed.tech/tags/accelerators.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-assistant](<https://devfeed.tech/tags/ai-assistant.md>), [ai-inference](<https://devfeed.tech/tags/ai-inference.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-scaling](<https://devfeed.tech/tags/ai-scaling.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [blogs-lphp](<https://devfeed.tech/tags/blogs-lphp.md>), [concurrency](<https://devfeed.tech/tags/concurrency.md>), [data-movement](<https://devfeed.tech/tags/data-movement.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [inference](<https://devfeed.tech/tags/inference.md>), [kv-cache](<https://devfeed.tech/tags/kv-cache.md>), [latency](<https://devfeed.tech/tags/latency.md>), [low-power-high-performance](<https://devfeed.tech/tags/low-power-high-performance.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-pressure](<https://devfeed.tech/tags/memory-pressure.md>), [memory-wall](<https://devfeed.tech/tags/memory-wall.md>), [models](<https://devfeed.tech/tags/models.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [multimodal-models](<https://devfeed.tech/tags/multimodal-models.md>), [rambus](<https://devfeed.tech/tags/rambus.md>), [semiconductor](<https://devfeed.tech/tags/semiconductor.md>), [semiconductor-engineering](<https://devfeed.tech/tags/semiconductor-engineering.md>)

### AI overview

AI inference turns the memory wall into a system-wide trade-off among capacity, bandwidth, latency, power, utilization, and cost. KV cache avoids recomputing prior token states, but longer contexts and more concurrent requests increase the data that must be retained and moved. Its lifecycle--creation during prefill, repeated access during decode, and possible reuse, movement, compression, or eviction--makes placement a workload-dependent decision. Higher-bandwidth memory keeps data close to accelerators; larger or pooled tiers can expand capacity, but movement adds latency, bandwidth use, and energy. Moving or replicating state is worthwhile only when its benefits outweigh those costs and service-level limits. The text argues for coordinating compute, memory tiers, interconnect, and software around workload needs, since improving one memory dimension can shift pressure elsewhere.

### Source excerpt

Longer contexts, higher concurrency, multimodal models, and AI agents will continue changing where memory pressure appears. The post Beyond Bandwidth: AI Inference Is Reshaping The Memory Wall appeared first on Semiconductor Engineering.

## The Semiconductor Singularity: When AI Starts Accelerating The Silicon That Accelerates AI

DevFeed: [The Semiconductor Singularity: When AI Starts Accelerating The Silicon That Accelerates AI](<https://devfeed.tech/articles/the-semiconductor-singularity-when-ai-starts-accelerating-the-silicon-that-accelerates-ai-130171.md>)

Original publisher: [Read original article](<https://semiengineering.com/the-semiconductor-singularity-when-ai-starts-accelerating-the-silicon-that-accelerates-ai/>)

Author: Jim Shiely

Published: 2026-11-12T08:06:49Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [knowledge-engineering](<https://devfeed.tech/topics/knowledge-engineering.md>), [human review](<https://devfeed.tech/topics/human-review.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [accelerate](<https://devfeed.tech/tags/accelerate.md>), [accelerated](<https://devfeed.tech/tags/accelerated.md>), [accelerated-computing](<https://devfeed.tech/tags/accelerated-computing.md>), [acceleration](<https://devfeed.tech/tags/acceleration.md>), [accessible](<https://devfeed.tech/tags/accessible.md>), [account](<https://devfeed.tech/tags/account.md>), [act](<https://devfeed.tech/tags/act.md>), [actions](<https://devfeed.tech/tags/actions.md>), [adoption](<https://devfeed.tech/tags/adoption.md>), [after](<https://devfeed.tech/tags/after.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-in-eda](<https://devfeed.tech/tags/ai-in-eda.md>), [best-of-both-lp-hp](<https://devfeed.tech/tags/best-of-both-lp-hp.md>), [blogs-lphp](<https://devfeed.tech/tags/blogs-lphp.md>), [chip-design](<https://devfeed.tech/tags/chip-design.md>), [computational-lithography](<https://devfeed.tech/tags/computational-lithography.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [generation](<https://devfeed.tech/tags/generation.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [low-power-high-performance](<https://devfeed.tech/tags/low-power-high-performance.md>), [next](<https://devfeed.tech/tags/next.md>), [semiconductor](<https://devfeed.tech/tags/semiconductor.md>), [semiconductor-engineering](<https://devfeed.tech/tags/semiconductor-engineering.md>), [siemens-eda](<https://devfeed.tech/tags/siemens-eda.md>), [silicon](<https://devfeed.tech/tags/silicon.md>), [tools](<https://devfeed.tech/tags/tools.md>)

### AI overview

AI is reducing the effort needed to develop software, use EDA tools and analyze semiconductor designs, allowing engineers to explore more alternatives within the same schedule. The author reports recreating and expanding notebook-level computational-lithography prototypes in about 30 hours with generative AI, compared with roughly 200-300 hours previously; these were not production tools. Agentic systems can coordinate steps such as gathering SEM data, calibrating models and checking analysis results, while engineers retain judgment. The proposed workflow assigns distinct roles to physics-based simulation, qualified models and deterministic verification, specialized machine learning, generative AI, agents and accelerated computing. Trust depends on calibrated models, measured silicon, traceable data, secure access, deterministic signoff and human control. As analysis speeds up, access, IP protection, validation and authorization can become larger constraints on the overall engineering cycle.

### Source excerpt

Use evidence from silicon to improve the tools and decisions that create the next generation of hardware. The post The Semiconductor Singularity: When AI Starts Accelerating The Silicon That Accelerates AI appeared first on Semiconductor Engineering.

## Designing Edge AI SoCs At Near-Threshold: Why Foundation IP Can Help You Break Down The Power Wall

DevFeed: [Designing Edge AI SoCs At Near-Threshold: Why Foundation IP Can Help You Break Down The Power Wall](<https://devfeed.tech/articles/designing-edge-ai-socs-at-near-threshold-why-foundation-ip-can-help-you-break-down-the-power-wall-130164.md>)

Original publisher: [Read original article](<https://semiengineering.com/designing-edge-ai-socs-at-near-threshold-why-foundation-ip-can-help-you-break-down-the-power-wall/>)

Author: Andrew Appleby

Published: 2026-11-12T08:05:04Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [tinyml](<https://devfeed.tech/topics/tinyml.md>), [Sustainable Computing](<https://devfeed.tech/topics/sustainable-computing.md>)

Tags: [3nm](<https://devfeed.tech/tags/3nm.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-networking](<https://devfeed.tech/tags/ai-networking.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [area](<https://devfeed.tech/tags/area.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [benefit](<https://devfeed.tech/tags/benefit.md>), [blocks](<https://devfeed.tech/tags/blocks.md>), [blogs-lphp](<https://devfeed.tech/tags/blogs-lphp.md>), [edge-ai](<https://devfeed.tech/tags/edge-ai.md>), [energy](<https://devfeed.tech/tags/energy.md>), [foundation-ip](<https://devfeed.tech/tags/foundation-ip.md>), [ip](<https://devfeed.tech/tags/ip.md>), [ip-and-lp-in-socs](<https://devfeed.tech/tags/ip-and-lp-in-socs.md>), [logic-libraries](<https://devfeed.tech/tags/logic-libraries.md>), [low-power-high-performance](<https://devfeed.tech/tags/low-power-high-performance.md>), [low-voltage-design](<https://devfeed.tech/tags/low-voltage-design.md>), [near-threshold](<https://devfeed.tech/tags/near-threshold.md>), [performance](<https://devfeed.tech/tags/performance.md>), [power](<https://devfeed.tech/tags/power.md>), [real-time](<https://devfeed.tech/tags/real-time.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [synopsys](<https://devfeed.tech/tags/synopsys.md>), [voltage](<https://devfeed.tech/tags/voltage.md>), [voltage-scaling](<https://devfeed.tech/tags/voltage-scaling.md>)

### AI overview

Near-threshold operation around 0.4-0.5 V can improve energy per operation for edge AI SoCs, but delay and variability rise and noise margins shrink, making predictable design a cross-stack challenge. The excerpt says success depends on carefully selected and silicon-validated Foundation IP, accurate modeling, and close alignment with EDA flows. Memory can set system Vmin; a dual-rail architecture is cited as decoupling logic and memory voltage, with a 38 Mb/mm² SRAM validated at 380/540 mV in 3nm FinFET. It also describes different memory needs across MAC tiles, activation units, and shared buffers, and notes that I/O reliability can constrain system Vmin. In one high-volume AI networking SoC engagement, tuned low-voltage logic cells, custom level shifters, a dual-rail memory compiler, selected PVT corners, and application engineering support were used; the stated result was predictable execution without schedule or silicon risk. No quantitative power or performance result is provided.

### Source excerpt

Traditional voltage scaling is no longer sufficient to deliver real-time performance within tight energy and thermal limits. The post Designing Edge AI SoCs At Near-Threshold: Why Foundation IP Can Help You Break Down The Power Wall appeared first on Semiconductor Engineering.

## IP Integration Challenges Evolve

DevFeed: [IP Integration Challenges Evolve](<https://devfeed.tech/articles/ip-integration-challenges-evolve-130167.md>)

Original publisher: [Read original article](<https://semiengineering.com/ip-integration-challenges-evolve/>)

Author: Brian Bailey

Published: 2026-11-12T08:05:00Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>), [Integration testing](<https://devfeed.tech/topics/integration-testing.md>)

Tags: [about](<https://devfeed.tech/tags/about.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [accelerate](<https://devfeed.tech/tags/accelerate.md>), [added](<https://devfeed.tech/tags/added.md>), [addition](<https://devfeed.tech/tags/addition.md>), [after](<https://devfeed.tech/tags/after.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agile-analog](<https://devfeed.tech/tags/agile-analog.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-automation](<https://devfeed.tech/tags/ai-automation.md>), [analog-content](<https://devfeed.tech/tags/analog-content.md>), [axiomise](<https://devfeed.tech/tags/axiomise.md>), [but](<https://devfeed.tech/tags/but.md>), [challenges](<https://devfeed.tech/tags/challenges.md>), [chiplet](<https://devfeed.tech/tags/chiplet.md>), [chiplet-integration](<https://devfeed.tech/tags/chiplet-integration.md>), [chiplets](<https://devfeed.tech/tags/chiplets.md>), [complex](<https://devfeed.tech/tags/complex.md>), [first](<https://devfeed.tech/tags/first.md>), [ic-manage](<https://devfeed.tech/tags/ic-manage.md>), [integration](<https://devfeed.tech/tags/integration.md>), [ip](<https://devfeed.tech/tags/ip.md>), [ip-integration](<https://devfeed.tech/tags/ip-integration.md>), [ip-verification](<https://devfeed.tech/tags/ip-verification.md>), [keysight](<https://devfeed.tech/tags/keysight.md>), [low-power-high-performance](<https://devfeed.tech/tags/low-power-high-performance.md>), [more](<https://devfeed.tech/tags/more.md>), [power-and-thermal-control](<https://devfeed.tech/tags/power-and-thermal-control.md>), [semiconductor](<https://devfeed.tech/tags/semiconductor.md>), [solutions](<https://devfeed.tech/tags/solutions.md>), [synopsys](<https://devfeed.tech/tags/synopsys.md>), [top-stories](<https://devfeed.tech/tags/top-stories.md>), [verification](<https://devfeed.tech/tags/verification.md>), [version-control](<https://devfeed.tech/tags/version-control.md>), [vsia](<https://devfeed.tech/tags/vsia.md>)

### AI overview

IP integration failures often emerge between individually verified blocks, where mismatched protocol assumptions can cause deadlocks, lost credits, ordering errors, or system hangs. Complexity depends on interactions and undocumented assumptions, not simply chip size. The excerpt identifies incomplete specifications, software and firmware dependencies, power sequencing, thermal and analog constraints, chiplet interoperability, opaque or differently configured IP, and weak hardware/software version traceability as growing pressures. It notes that UCIe addresses early firmware download and link-layer interoperability, while system-level agreement on address maps, coherency, and software remains necessary. Verification may need to cover legal configurations and cross-block behavior; AI models are cited as helping reach corner cases faster. The article concludes that management mechanisms are evolving alongside the problem, with a follow-up report planned on newer solutions and AI.

### Source excerpt

Integration challenges are growing more complex, but the problem is well understood and solutions are emerging. The post IP Integration Challenges Evolve appeared first on Semiconductor Engineering.

## The Memory Challenge Of Physical AI

DevFeed: [The Memory Challenge Of Physical AI](<https://devfeed.tech/articles/the-memory-challenge-of-physical-ai-130170.md>)

Original publisher: [Read original article](<https://semiengineering.com/the-memory-challenge-of-physical-ai/>)

Author: Joe Chen

Published: 2026-11-12T08:04:39Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Physical AI](<https://devfeed.tech/topics/physical-ai.md>), [Memory Heirarchy](<https://devfeed.tech/topics/memory-heirarchy.md>), [Chip design](<https://devfeed.tech/topics/chip-design.md>), [Robotics](<https://devfeed.tech/topics/robotics.md>), [CPU Cache](<https://devfeed.tech/topics/cpu-cache.md>)

Tags: [about](<https://devfeed.tech/tags/about.md>), [accelerators](<https://devfeed.tech/tags/accelerators.md>), [act](<https://devfeed.tech/tags/act.md>), [actuator](<https://devfeed.tech/tags/actuator.md>), [adas](<https://devfeed.tech/tags/adas.md>), [ai](<https://devfeed.tech/tags/ai.md>), [automotive](<https://devfeed.tech/tags/automotive.md>), [blogs-lphp](<https://devfeed.tech/tags/blogs-lphp.md>), [cadence](<https://devfeed.tech/tags/cadence.md>), [challenge](<https://devfeed.tech/tags/challenge.md>), [data](<https://devfeed.tech/tags/data.md>), [data-movement](<https://devfeed.tech/tags/data-movement.md>), [device](<https://devfeed.tech/tags/device.md>), [everything-low-power](<https://devfeed.tech/tags/everything-low-power.md>), [gddr](<https://devfeed.tech/tags/gddr.md>), [hierarchy](<https://devfeed.tech/tags/hierarchy.md>), [latency](<https://devfeed.tech/tags/latency.md>), [low-power-high-performance](<https://devfeed.tech/tags/low-power-high-performance.md>), [lpddr](<https://devfeed.tech/tags/lpddr.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-hierarchy](<https://devfeed.tech/tags/memory-hierarchy.md>), [patterns](<https://devfeed.tech/tags/patterns.md>), [physical-ai](<https://devfeed.tech/tags/physical-ai.md>), [requirements](<https://devfeed.tech/tags/requirements.md>), [reuse](<https://devfeed.tech/tags/reuse.md>), [right](<https://devfeed.tech/tags/right.md>), [robotics](<https://devfeed.tech/tags/robotics.md>), [sensor-streaming](<https://devfeed.tech/tags/sensor-streaming.md>), [sensors](<https://devfeed.tech/tags/sensors.md>), [size](<https://devfeed.tech/tags/size.md>), [streams](<https://devfeed.tech/tags/streams.md>)

### AI overview

Physical AI performance depends on predictable data movement across compute, on-chip SRAM and cache, and external memory. Keeping frequently reused or latency-sensitive data close to processing engines can reduce transfers, energy use, bandwidth demand, and stalls; limited on-chip capacity means less-used or larger data remains in external memory. Architects should size the full hierarchy--including capacity, bandwidth, buffering, power, and integration--for each device's streams, reuse patterns, latency needs, and operating limits. The text describes LPDDR as a fit when bandwidth per watt and capacity matter, while GDDR may suit continuous image or sensor streams needing high bandwidth with fewer DRAM devices. Humanoids, drones, cameras, and vehicles impose different combinations of power, thermal, throughput, and response-predictability demands. Cadence is cited as offering LPDDR6/LPDDR5X PHY and Controller IP up to 14.4 Gbps and GDDR7 PHY IP up to 36 Gbps.

### Source excerpt

Right-size the memory hierarchy around the device's data streams, reuse patterns, latency requirements, and operating constraints. The post The Memory Challenge Of Physical AI appeared first on Semiconductor Engineering.

## Introduction To FP8 Formats: Part 1

DevFeed: [Introduction To FP8 Formats: Part 1](<https://devfeed.tech/articles/introduction-to-fp8-formats-part-1-130166.md>)

Original publisher: [Read original article](<https://semiengineering.com/introduction-to-fp8-formats-part-1/>)

Author: Siyu Mou

Published: 2026-11-12T08:03:25Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [floating-point](<https://devfeed.tech/topics/floating-point.md>), [quantization](<https://devfeed.tech/topics/quantization.md>)

Tags: [8-bit-floating-point](<https://devfeed.tech/tags/8-bit-floating-point.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [arm](<https://devfeed.tech/tags/arm.md>), [arm-v9](<https://devfeed.tech/tags/arm-v9.md>), [at-the-core](<https://devfeed.tech/tags/at-the-core.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [bf16](<https://devfeed.tech/tags/bf16.md>), [blogs-lphp](<https://devfeed.tech/tags/blogs-lphp.md>), [compute](<https://devfeed.tech/tags/compute.md>), [data-center](<https://devfeed.tech/tags/data-center.md>), [data-types](<https://devfeed.tech/tags/data-types.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [edge](<https://devfeed.tech/tags/edge.md>), [floating-point](<https://devfeed.tech/tags/floating-point.md>), [formats](<https://devfeed.tech/tags/formats.md>), [fp16](<https://devfeed.tech/tags/fp16.md>), [fp8](<https://devfeed.tech/tags/fp8.md>), [frameworks](<https://devfeed.tech/tags/frameworks.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [h100](<https://devfeed.tech/tags/h100.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [hopper](<https://devfeed.tech/tags/hopper.md>), [low-power-high-performance](<https://devfeed.tech/tags/low-power-high-performance.md>), [number-formats](<https://devfeed.tech/tags/number-formats.md>)

### AI overview

FP8 trades precision and dynamic range to improve ML throughput in data centers and reduce model size, memory use, and bandwidth on edge devices. The described interchange specification offers E4M3, which favors precision for activations and weights, and E5M2, which offers more exponent range for gradients. Arm's FPMR lets each source operand select its FP8 format and applies scaling; the text says hardware can handle its writes more efficiently than frequent FPCR updates. Arm extensions add FP8 instructions across Advanced SIMD, SVE, and SME. These use FP8 inputs with wider FP16 or FP32 accumulation to retain bandwidth benefits while supporting numerical stability. The text also identifies NVIDIA Hopper H100 support and major framework deployment.

### Source excerpt

The rationale behind 8-bit floating-point data formats, their ecosystem momentum, and hardware support. The post Introduction To FP8 Formats: Part 1 appeared first on Semiconductor Engineering.

## RISC-V For AI: Breakthrough Or Bad Idea?

DevFeed: [RISC-V For AI: Breakthrough Or Bad Idea?](<https://devfeed.tech/articles/risc-v-for-ai-breakthrough-or-bad-idea-130169.md>)

Original publisher: [Read original article](<https://semiengineering.com/risc-v-for-ai-breakthrough-or-bad-idea/>)

Author: Ann Mutschler

Published: 2026-11-12T08:01:26Z

Content type: news

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [RISC-V](<https://devfeed.tech/topics/riscv.md>), [Chip design](<https://devfeed.tech/topics/chip-design.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-accelerator](<https://devfeed.tech/tags/ai-accelerator.md>), [ai-accelerators](<https://devfeed.tech/tags/ai-accelerators.md>), [breker](<https://devfeed.tech/tags/breker.md>), [low-power-high-performance](<https://devfeed.tech/tags/low-power-high-performance.md>), [quadric](<https://devfeed.tech/tags/quadric.md>), [risc-v](<https://devfeed.tech/tags/risc-v.md>), [round-tables](<https://devfeed.tech/tags/round-tables.md>), [siemens-eda](<https://devfeed.tech/tags/siemens-eda.md>), [signature-ip](<https://devfeed.tech/tags/signature-ip.md>), [top-stories](<https://devfeed.tech/tags/top-stories.md>)

### AI overview

The excerpts show disagreement over RISC-V's role in AI: Quadric's Steve Roddy argues against using a CPU-centered, partitioned design for heavy matrix inference, citing data movement and scaling costs, while Breker's Dave Kelf describes RISC-V control processors with custom vector or matrix instructions and load/store operations that can reduce partition latency. Kelf says some such designs work, though they take substantial engineering effort. Roddy contrasts this hands-on customization with Tensilica's automated instruction, compiler, simulator, and verification flow, which carries a royalty. The other contributors emphasize system integration: Siemens EDA supports tracing custom instructions and SoC signals, and SignatureIP stresses NoC configuration and system-level checks for coherency, traffic bursts, and deadlocks. The discussion presents RISC-V as flexible for control and specialized functions, while leaving its suitability for heavy inference contested.

### Source excerpt

Custom instructions promise flexibility, but experts question whether RISC-V belongs at the heart of AI acceleration. The post RISC-V For AI: Breakthrough Or Bad Idea? appeared first on Semiconductor Engineering.

## Embedded ECC Protects AI Weights Against Bit-Flip Attacks (University of Florida)

DevFeed: [Embedded ECC Protects AI Weights Against Bit-Flip Attacks (University of Florida)](<https://devfeed.tech/articles/embedded-ecc-protects-ai-weights-against-bit-flip-attacks-university-of-florida-129706.md>)

Original publisher: [Read original article](<https://semiengineering.com/embedded-ecc-protects-ai-weights-against-bit-flip-attacks-university-of-florida/>)

Author: Technical Paper Link

Published: 2026-11-12T00:42:18Z

Content type: news

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Neural Network](<https://devfeed.tech/topics/neural-network.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-accelerators](<https://devfeed.tech/tags/ai-accelerators.md>), [ai-ml-dl](<https://devfeed.tech/tags/ai-ml-dl.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [bit-flip-attacks](<https://devfeed.tech/tags/bit-flip-attacks.md>), [critical](<https://devfeed.tech/tags/critical.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [ecc](<https://devfeed.tech/tags/ecc.md>), [embedded](<https://devfeed.tech/tags/embedded.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [error-correcting-codes](<https://devfeed.tech/tags/error-correcting-codes.md>), [hardware-security](<https://devfeed.tech/tags/hardware-security.md>), [key](<https://devfeed.tech/tags/key.md>), [machine-learning-security](<https://devfeed.tech/tags/machine-learning-security.md>), [memory-reliability](<https://devfeed.tech/tags/memory-reliability.md>), [model](<https://devfeed.tech/tags/model.md>), [model-weights](<https://devfeed.tech/tags/model-weights.md>), [neural-networks](<https://devfeed.tech/tags/neural-networks.md>), [overhead](<https://devfeed.tech/tags/overhead.md>), [paper](<https://devfeed.tech/tags/paper.md>), [parity](<https://devfeed.tech/tags/parity.md>), [quantized-neural-networks](<https://devfeed.tech/tags/quantized-neural-networks.md>), [rowhammer](<https://devfeed.tech/tags/rowhammer.md>), [security](<https://devfeed.tech/tags/security.md>), [systolic-arrays](<https://devfeed.tech/tags/systolic-arrays.md>), [technical-papers](<https://devfeed.tech/tags/technical-papers.md>), [university-of-florida](<https://devfeed.tech/tags/university-of-florida.md>), [weight](<https://devfeed.tech/tags/weight.md>), [zero](<https://devfeed.tech/tags/zero.md>)

### AI overview

SANAI is a sensitivity-guided scheme that targets critical neural-network weights against bit flips while claiming zero additional parity-storage overhead. It embeds parity across a two-dimensional weight space: spread-transform dither modulation binds parity to a secret key, and keyed interleaving disperses concentrated attacks across codewords. The supplied abstract excerpt does not provide evaluation results or further implementation details.

### Source excerpt

Researchers at the University of Florida published "SANAI: Secure ECC Distributed Keyed Parity Embedded in Weights for Protecting Neural Networks against Bit-Flip Attacks." Abstract Excerpt: "We propose SANAI, a sensitivity-guided, bit-plane-selective scheme that protects critical model weights against bit flips with zero additional parity-storage overhead. SANAI secures parity through keyed embedding across a two-dimensional weight... " read more The post Embedded ECC Protects AI Weights Against Bit-Flip Attacks (University of Florida) appeared first on Semiconductor Engineering.

## Optimized E-Beam Evaporation Reduces Metal-Semiconductor Contact Degradation (RWTH Aachen)

DevFeed: [Optimized E-Beam Evaporation Reduces Metal-Semiconductor Contact Degradation (RWTH Aachen)](<https://devfeed.tech/articles/optimized-e-beam-evaporation-reduces-metal-semiconductor-contact-degradation-rwth-aachen-129707.md>)

Original publisher: [Read original article](<https://semiengineering.com/optimized-e-beam-evaporation-reduces-metal-semiconductor-contact-degradation-rwth-aachen/>)

Author: Technical Paper Link

Published: 2026-11-11T23:00:41Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>), [Electrical engineering and computer science (EECS)](<https://devfeed.tech/topics/electrical-engineering-and-computer-science-eecs.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [carrier](<https://devfeed.tech/tags/carrier.md>), [charged](<https://devfeed.tech/tags/charged.md>), [charged-particle-bombardment](<https://devfeed.tech/tags/charged-particle-bombardment.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [deposition](<https://devfeed.tech/tags/deposition.md>), [ebe](<https://devfeed.tech/tags/ebe.md>), [electron](<https://devfeed.tech/tags/electron.md>), [electron-beam-evaporation](<https://devfeed.tech/tags/electron-beam-evaporation.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [fermi-level-pinning](<https://devfeed.tech/tags/fermi-level-pinning.md>), [flux](<https://devfeed.tech/tags/flux.md>), [manufacturing](<https://devfeed.tech/tags/manufacturing.md>), [materials](<https://devfeed.tech/tags/materials.md>), [measurements](<https://devfeed.tech/tags/measurements.md>), [metal](<https://devfeed.tech/tags/metal.md>), [metal-semiconductor-contacts](<https://devfeed.tech/tags/metal-semiconductor-contacts.md>), [n-type-silicon](<https://devfeed.tech/tags/n-type-silicon.md>), [pmma](<https://devfeed.tech/tags/pmma.md>), [published](<https://devfeed.tech/tags/published.md>), [radiation-damage](<https://devfeed.tech/tags/radiation-damage.md>), [rwth-aachen-university](<https://devfeed.tech/tags/rwth-aachen-university.md>), [schottky-diodes](<https://devfeed.tech/tags/schottky-diodes.md>), [semiconductor](<https://devfeed.tech/tags/semiconductor.md>), [semiconductor-manufacturing](<https://devfeed.tech/tags/semiconductor-manufacturing.md>), [setup](<https://devfeed.tech/tags/setup.md>), [technical-papers](<https://devfeed.tech/tags/technical-papers.md>), [tool](<https://devfeed.tech/tags/tool.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

RWTH Aachen researchers report an optimized electron beam evaporation setup that suppresses charged-particle bombardment during metal coating of 4-inch semiconductor samples. They measured charge-carrier flux in situ across tool configurations and compared platinum-silicon Schottky diodes made with and without the optimized setup. The abstract reports reduced Fermi-level pinning and higher Schottky barriers to n-type silicon with the optimized setup.

### Source excerpt

Researchers at RWTH Aachen University published "Mitigating the degradation of metal-semiconductor contacts with optimized electron beam evaporation." Abstract "An optimized electron beam evaporation (EBE) setup is presented that allows the suppression of charged particle bombardment of 4" semiconductor samples during metal coating. In-situ measurements of charge carrier flux for different tool configurations are carried out.... " read more The post Optimized E-Beam Evaporation Reduces Metal-Semiconductor Contact Degradation (RWTH Aachen) appeared first on Semiconductor Engineering.

## Variable-Length Prompts on Chimera GPNPU

DevFeed: [Variable-Length Prompts on Chimera GPNPU](<https://devfeed.tech/articles/variable-length-prompts-on-chimera-gpnpu-129315.md>)

Original publisher: [Read original article](<https://semiengineering.com/variable-length-prompts-on-chimera-gpnpu/>)

Author: Quadric.io

Published: 2026-11-11T08:02:29Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [context window](<https://devfeed.tech/topics/context-window.md>)

Tags: [capacity](<https://devfeed.tech/tags/capacity.md>), [generalpurpose-neural-processing-unit](<https://devfeed.tech/tags/generalpurpose-neural-processing-unit.md>), [gpnpu](<https://devfeed.tech/tags/gpnpu.md>), [inference](<https://devfeed.tech/tags/inference.md>), [kv-cache-interface](<https://devfeed.tech/tags/kv-cache-interface.md>), [kv-cache-traffic](<https://devfeed.tech/tags/kv-cache-traffic.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [low-power-high-performance](<https://devfeed.tech/tags/low-power-high-performance.md>), [padding](<https://devfeed.tech/tags/padding.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [quadric](<https://devfeed.tech/tags/quadric.md>), [quadric-chimera-gpnpu](<https://devfeed.tech/tags/quadric-chimera-gpnpu.md>), [range](<https://devfeed.tech/tags/range.md>), [variable](<https://devfeed.tech/tags/variable.md>), [whitepapers](<https://devfeed.tech/tags/whitepapers.md>)

### AI overview

The excerpt says fixed-length compiled inference graphs process their full configured token capacity on every run, so shorter prompts spend computation on padding; serving multiple prompt lengths efficiently may require separate binaries and qualification for each range. Quadric Chimera GPNPU is presented as using one graph for every prompt length up to a configured capacity, though the excerpt does not explain its guarantee or optimization in detail.

### Source excerpt

How a flexible binary can serve every length up to the capacity you set. The post Variable-Length Prompts on Chimera GPNPU appeared first on Semiconductor Engineering.

## Blog Review: Nov. 11

DevFeed: [Blog Review: Nov. 11](<https://devfeed.tech/articles/blog-review-nov-11-129313.md>)

Original publisher: [Read original article](<https://semiengineering.com/blog-review-nov-11-3/>)

Author: Jesse Allen

Published: 2026-11-11T08:01:21Z

Content type: news

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [Chip design](<https://devfeed.tech/topics/chip-design.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [ai security](<https://devfeed.tech/topics/ai-security.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [advanced-packaging](<https://devfeed.tech/tags/advanced-packaging.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-compute](<https://devfeed.tech/tags/ai-compute.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-scaling](<https://devfeed.tech/tags/ai-scaling.md>), [analog](<https://devfeed.tech/tags/analog.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [arm](<https://devfeed.tech/tags/arm.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [automated-driving](<https://devfeed.tech/tags/automated-driving.md>), [automotive](<https://devfeed.tech/tags/automotive.md>), [blog](<https://devfeed.tech/tags/blog.md>), [blog-reviews](<https://devfeed.tech/tags/blog-reviews.md>), [bodewell-group](<https://devfeed.tech/tags/bodewell-group.md>), [business](<https://devfeed.tech/tags/business.md>), [cadence](<https://devfeed.tech/tags/cadence.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [intel-foundry](<https://devfeed.tech/tags/intel-foundry.md>), [interconnects](<https://devfeed.tech/tags/interconnects.md>), [keysight](<https://devfeed.tech/tags/keysight.md>), [news](<https://devfeed.tech/tags/news.md>), [review](<https://devfeed.tech/tags/review.md>), [semi](<https://devfeed.tech/tags/semi.md>), [semiconductor](<https://devfeed.tech/tags/semiconductor.md>), [semiconductor-engineering](<https://devfeed.tech/tags/semiconductor-engineering.md>), [siemens](<https://devfeed.tech/tags/siemens.md>), [synopsys](<https://devfeed.tech/tags/synopsys.md>), [system-level-design](<https://devfeed.tech/tags/system-level-design.md>), [trends](<https://devfeed.tech/tags/trends.md>)

### AI overview

Interconnect bandwidth, timing, and power are key constraints on scaling AI infrastructure: the excerpt says data must reach the right processor at the right time and acceptable power cost, and that interconnect speeds must rise with compute performance. It also highlights floating nets in analog power-down circuits as a reliability risk that conventional verification may miss, and describes Adaptive Traffic Profiles as a way to assess whether interconnects can sustain real traffic. Other featured topics include Arm security features for mitigating Control Flow attacks, resilience, anti-spoofing, low Earth orbit and lunar navigation trends in positioning, navigation, and timing, and yield prediction, signal integrity, automated-driving safety, and power delivery. The excerpt reports a SEMI Market Symposium projection that AI could help drive semiconductor revenue from $1.5 trillion to $3 trillion over the next decade.

### Source excerpt

Interconnects key in AI infrastructure; floating nets; mitigating Control Flow attacks; PNT trends; SEMI Market Symposium. The post Blog Review: Nov. 11 appeared first on Semiconductor Engineering.

## Reducing Lithography Learning Cycles To Accelerate FOPLP Ramps

DevFeed: [Reducing Lithography Learning Cycles To Accelerate FOPLP Ramps](<https://devfeed.tech/articles/reducing-lithography-learning-cycles-to-accelerate-foplp-ramps-126572.md>)

Original publisher: [Read original article](<https://semiengineering.com/reducing-lithography-learning-cycles-to-accelerate-foplp-ramps/>)

Author: John Chang

Published: 2026-11-10T08:08:35Z

Content type: article

Language: en

Sources: [Semiconductor Engineering](<https://devfeed.tech/sources/semiconductor-engineering.md>)

Topics: [tsmc](<https://devfeed.tech/topics/tsmc.md>), [Predictive AI](<https://devfeed.tech/topics/predictive-ai.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [blogs-ptes](<https://devfeed.tech/tags/blogs-ptes.md>), [cycles](<https://devfeed.tech/tags/cycles.md>), [fan-out-panel-level-packaging](<https://devfeed.tech/tags/fan-out-panel-level-packaging.md>), [foplp](<https://devfeed.tech/tags/foplp.md>), [intelligent-innovation](<https://devfeed.tech/tags/intelligent-innovation.md>), [learning](<https://devfeed.tech/tags/learning.md>), [lithography](<https://devfeed.tech/tags/lithography.md>), [metrology](<https://devfeed.tech/tags/metrology.md>), [onto-innovation](<https://devfeed.tech/tags/onto-innovation.md>), [overlay](<https://devfeed.tech/tags/overlay.md>), [test-measurement-and-analytics](<https://devfeed.tech/tags/test-measurement-and-analytics.md>), [yield](<https://devfeed.tech/tags/yield.md>), [yield-prediction](<https://devfeed.tech/tags/yield-prediction.md>)

### AI overview

FOPLP yield prediction can shorten lithography learning cycles by estimating yield from panel metrology and process settings before engineers commit to qualification runs. The algorithms evaluate pattern errors and die shifts alongside exposure-field configuration, panel and die layouts, and overlay thresholds; visualizations such as heat maps and vector plots help locate variation and root causes. Correction strategies trade yield against throughput: global correction preserves throughput but can leave local errors, while zone and die-by-die correction can improve yield at a throughput cost; site-by-site correction is presented as a compromise. In an HVM substrate assessment, predicted and measured yield initially differed by 1.2%-2.7%; after standardizing manual inspection, errors fell to 0.2%-0.8%. The article says simulation can screen conditions before qualification substrates, and in-line integration can support earlier defect detection during production. These reported results describe the study and approach, not a guarantee for other processes.

### Source excerpt

How yield prediction technology can reduce trial-and-error in advanced packaging lithography. The post Reducing Lithography Learning Cycles To Accelerate FOPLP Ramps appeared first on Semiconductor Engineering.

[Next page](<https://devfeed.tech/sources/semiconductor-engineering.md?cursor=WyIyMDI2LTExLTEwVDA4OjA4OjM1KzAwOjAwIiwgIjRjZWFjNTdhLWMzZjEtNDJkOC04OGUzLWRlMzBkYTRkNTViZiJd>)