On-Chip Monitors Detect Bit-Flip Attacks in AI Accelerators (UF, UT Dallas, RPI)

Academic team proposes on-chip monitors to detect bit-flip attacks in neural network hardware

By GPU Data Hub Desk · GPU Data Hub·Editorial standards·12 Oct 2026, 22:34·United States

Academic researchers have devised a hardware defence architecture titled BARE-AI to protect artificial intelligence processors from bit-flip attacks. The framework deploys built-in performance monitors directly on the silicon to identify malicious tampering aimed at altering deep neural network parameters. The system addresses critical security vulnerabilities in deep learning accelerators deployed across safety-critical operational environments.

Original source

This summary was written by the GPU Data Hub desk from reporting published by Semiconductor Engineering on 12 Oct 2026, 22:34. Read the full article at the original publisher.

Read Original Source →

Topics

AI accelerators hardware security bit-flip attacks neural networks

LinkedIn X

More from GPU Data Hub

Latest →

Micron, SoftBank and ASML slide as Wall Street chip rout spreads

Global semiconductor and technology shares dropped on Tuesday as an equity retreat on Wall Street reached European and Asian bourses. Major industry participants including ASML, Micron and SoftBank saw their share prices decline amid the widespread market contraction.

Semiconductors·1h ago·CNBC Technology

Microsoft unveils Azure Maia 300, says chip will be its “primary inference system”

Microsoft has introduced its Azure Maia 300 processor, establishing the bespoke chip as the primary workhorse for its artificial intelligence inference operations. The development follows the operational rollout of its earlier Maia 200 hardware across several international territories.

Semiconductors·2h ago·Data Center Dynamics

Chip Industry Technical Paper Roundup: Oct. 13

A recent collection of semiconductor research papers highlights advances across lithography, materials science and specialised computing hardware. Key subjects addressed include hyper-NA EUV lithography, neuromorphic computing architectures, and methods to detect bit-flip attacks in AI acceleration hardware.

Semiconductors·6h ago·Semiconductor Engineering

Carbon-Dot Interface Enables Ultrathin High-κ Dielectrics on 2D Semiconductors (UIUC, NETL, Argonne, Monash)

Researchers from several American and Australian institutions have demonstrated a method to integrate high-κ dielectrics onto two-dimensional semiconductors using a carbon-dot monolayer assembly. This atomically thin structure acts as a van der Waals interfacial layer that facilitates oxide deposition without damaging the underlying material. The technique targets longstanding material integration hurdles in the manufacture of next-generation 2D semiconductor devices.

Semiconductors·14h ago·Semiconductor Engineering

Microsoft Boosts HBM Bandwidth by 2.5x with Maia 300 Accelerator

Microsoft has revealed its Maia 300 AI accelerator, increasing high-bandwidth memory throughput to 18 terabytes per second from 7.1 TBps on the earlier Maia 200. The 2.5-fold increase in memory bandwidth is engineered to process AI inference workloads more efficiently whilst decreasing energy expenditure on a per-token basis. Cloud and AI executive vice president Scott Guthrie presented the silicon update as part of the group's ongoing custom infrastructure roadmap.

Semiconductors·14h ago·HPCwire

Deterministic Nucleation Enables Controlled Growth of 2D Semiconductors (KAIST, TDS, Hanyang et al.)

Researchers at several South Korean academic institutions and TDS Innovation have introduced a method to direct the nucleation of two-dimensional semiconductors using etching flux. The approach establishes spatial control over where and when crystalline structures develop during the semiconductor deposition phase. This process provides an alternative route to conventional area-selective deposition methods for synthesising advanced 2D electronic materials.

Semiconductors·15h ago·Semiconductor Engineering