
Seven New Chip Papers: HBM4 Control, 2D CFETs, Anti-Smuggling 5G
Seven new papers added Nov. 17 span HBM4 base-die controllers, 2D CFETs with single-crystal high-κ dielectrics, sub-volt ferroelectric capacitors and hardware-based anti-smuggling checks.
- By
- Nathan Brooks
- Filed
- Channel
- Semiconductors
- Read
- 3 min read
Semiconductor Engineering added seven technical papers to its research library on November 17, and the list reads like a compressed map of where the industry's problems now sit: memory controllers, ferroelectric capacitors, transistor contacts, and hardware security.
The institutions behind the papers range from UC Riverside to Seoul National University, with stops at Texas A&M, RWTH Aachen, the University of Florida, and a three-country collaboration on 2D transistors. None of these are vendor roadmaps. They are peer-facing research, which makes them a reasonable proxy for what will trouble process engineers roughly five years from now.
What does the HBM4 paper propose?
UC Riverside contributes work on HBM4 controller partitioning across host and base dies. The paper's premise is stated in its title: control functions need not remain solely on the host side of the interface.
Moving controller logic closer to the memory itself is not a new instinct, but HBM4 gives it new urgency. As stacked memory grows denser and the host-side memory controller grows correspondingly complex, the partitioning question becomes commercial, not merely academic. Base-die intelligence costs die area; host-side control costs signal margin. The Riverside work argues the trade-off explicitly.
Where does device physics stand?
Two papers attack the physical layer. A team from the National University of Singapore, Nexstrom, UNSW and colleagues reports low-power 2D CFETs built with single-crystal high-κ dielectrics. Complementary FETs stacked vertically remain one of the credible routes beyond conventional scaling, and the dielectric quality has long been the gating item.
RWTH Aachen University takes on a humbler but relentless problem: degradation of metal–semiconductor contacts, addressed through optimized electron beam evaporation. Contact resistance is the tax every advanced node pays. Any deposition tweak that reduces degradation compounds across billions of contacts per wafer.
The largest consortium — Seoul National University, Hanyang University, POSTECH and colleagues — works on ferroelectric Hf0.5Zr0.5O2 capacitors for DRAM-class sub-volt operation. The paper claims progress on two well-known obstacles: the Kay–Dunn limit and the speed–endurance dilemma. Ferroelectric hafnium-zirconium oxide has spent a decade promising DRAM-relevant capacitance at low voltage; endurance under fast switching is where that promise usually goes to retire.
Can hardware stop chip smuggling and model tampering?
Security accounts for three of the seven papers, which says something about the moment. Texas A&M University presents 5G CATS: 5G-based cryptographic authorization against trafficking of semiconductor devices. The idea is hardware-based location checking — a chip that can verify, cryptographically and over a network, whether it sits where export rules allow it to sit.
The timing requires no elaboration. Export enforcement on advanced AI silicon has become a practical problem for distributors and a reputational one for manufacturers, and a location-aware part would shift some of that burden into silicon itself.
The University of Florida offers SANAI, a scheme embedding secure ECC distributed keyed parity directly in neural network weights to protect against bit-flip attacks. Rather than guarding the memory around a model, the protection lives inside the weights. Model integrity is increasingly a product-liability question, not a research curiosity.
Ruhr University Bochum, the Max Planck Institute for Security and Privacy and Georgia Tech close the list with an audit of randomized caches in hardware, examining challenges and trade-offs. Cache randomization defends against side-channel inference, and the paper's contribution is an honest accounting of what the technique costs.
Why track library additions at all?
Because the composition of the list is itself a signal. Memory architecture, ferroelectric materials, contact physics and hardware-anchored security — seven papers, four problem families, all pointing at the same conclusion. The hard parts of the next device generation are moving below the transistor and above the software stack at the same time.
Semiconductor Engineering maintains its full technical paper library online; the November 17 batch suggests the 2026 conference cycle will argue fiercely over where control logic belongs and how much security silicon can carry.
Source: Semiconductor Engineering
More from Nathan Brooks
Show full bio
Senior reporter covering people, launches and day-to-day practice in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,016 articles
Related articles
uc-riverside-team-quantifies-hbm4-controller-split-between-host-and-base-die
UC Riverside Team Quantifies HBM4 Controller Split Between Host and Base Die
samsung-builds-custom-hbm4e-base-dies-for-meta-on-2nm-process
Samsung Builds Custom HBM4E Base Dies for Meta on 2nm Process
csis-warns-ai-and-hbm-demand-is-reshaping-the-memory-cycle-7aa78bff
CSIS Warns AI and HBM Demand Is Reshaping the Memory Cycle
hanmi-semiconductor-brings-hbm4-and-2-5d-tools-to-semicon-west-d46cfb92
Hanmi Semiconductor Brings HBM4 and 2.5D Tools to Semicon West



