Tech BusinessChips & Semiconductors
Semiconductor Engineering blog review covers PCIe trust and AI rack design
The 18 November 2026 round-up gathers industry posts on securing PCIe links, co-designing AI racks as single systems, and moving from monolithic scaling to heterogeneous chips.

Key takeaways
- Semiconductor Engineering's 18 November 2026 blog review collects posts on PCIe trust, rack-scale co-design and heterogeneous scaling.
- Cadence examines how to establish trust as sensitive data and cryptographic keys cross PCIe links.
- Synopsys argues that silicon, memory, cooling and power must be designed together for efficient AI racks.
- Siemens discusses the move from monolithic scaling to chiplets and 3D-IC, and the complexity that follows.
In this article
What the round-up covers
Semiconductor Engineering has published its Blog Review for 18 November 2026, a weekly collection of posts from engineers and executives across the chip industry. According to the publication, the edition covers PCIe and trust, rack-scale co-design, heterogeneous scaling, an AI data centre primer, on-device balancing and European semiconductor trends.
The review links to individual blog posts rather than reporting original findings. Each entry summarises a topic the publication considers relevant to chip designers and system architects.
Security concerns around PCIe links
Cadence's Rajesh Pamujula examines how trust can be established as sensitive data, proprietary models, cryptographic keys and confidential workloads travel over PCIe links. PCIe, or Peripheral Component Interconnect Express, is the high-speed connection standard used inside computers and servers to link processors with other hardware.
The post reflects a wider concern in data centre design: as more valuable data moves between components, the connections themselves become a security boundary. The review does not describe specific mitigations, so readers should consult the original Cadence post or official guidance for details.
Designing AI racks as one system
Synopsys' Neeraj Paliwal argues that efficient racks for demanding AI workloads require extreme co-design. In this approach, silicon, interface IP, memory, cooling and power are planned together as a single system rather than as separate parts.
The argument fits a broader shift in data centre hardware, where power and thermal limits increasingly shape what chips can do. Treating the rack as the design unit is presented as a way to keep performance rising without exceeding those limits.
From monolithic scaling to chiplets
Siemens' Kevin Rinebold and Andras Vass-Varnai look at how semiconductor design is changing as the industry moves away from monolithic scaling. Instead of making one large chip, designers increasingly combine smaller chiplets and stack layers in three dimensions.
The post considers how teams can manage the added complexity that comes with AI-driven 3D-IC and chiplet technologies. The review also notes related work from Arm on 8-bit floating-point formats and from Rambus on how longer contexts and multimodal models shift memory pressure.
Why the topics matter
The collection shows where chip industry attention is concentrated: security on internal links, system-level design for AI, and the packaging and memory choices that support larger models. It also includes a SEMI interview with imec's Patrick Vandenameele on trends reshaping the industry and the role Europe can play in global competitiveness.
For readers outside chip design, the themes point to the physical constraints behind AI computing. Power, cooling, memory bandwidth and data integrity are becoming as important as raw processor speed.
Sources
- Blog Review: Nov. 18Semiconductor Engineering
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



