# Chiplets and LLMs Expand Hardware Attack Surface

_Chiplets and LLMs revolutionize chip design but dramatically expand the hardware attack surface, necessitating new security paradigms for both systems and EDA flows._

**Updated:** 2026-08-22
**Published:** 2026-08-06
**Source:** https://www.startuphub.ai/ai-news/ai-research/2026/chiplets-and-llms-expand-hardware-attack-surface

---

The semiconductor industry faces a dual revolution: the rise of 2.5D chiplet systems and the integration of Large Language Models (LLMs) into Electronic Design Automation (EDA) flows.

Chiplets & LLMs EmergeCore

revolutionizing chip design with modularity and AI-driven automationFrom the article 2 mentionsThe semiconductor industry faces a dual revolution: the rise of 2.5D chiplet systems and the integration of Large Language Models (LLMs) into Electronic Design Automation (EDA) flows.leads toExpand Attack SurfaceDriverFrom the articleWhile chiplets promise benefits in yield and modularity, and LLMs enhance design productivity, they fundamentally expand the hardware attack surface across architectural, logical, and physical levels.viaHeterogeneous SystemsContextchiplets introduce complex interconnections and diverse component interactionsFrom the article 2 mentionsLooking forward, LLM systems themselves can be instrumental in advancing hardware security for modern systems, including complex chiplet designs.LLM EDA PipelinesContextAI integration into design flows creates novel vectors for exploitationFrom the articleThe integration of LLMs into EDA pipelines presents unique security challenges.requiresIsolated Roots of TrustCoresecuring chiplet architectures with physically separated security anchorsFrom the articleSecuring chiplet architectures necessitates a robust defense strategy, particularly focusing on physically isolated Root of Trust (RoT) implementations.Defend EDA PipelinesCoreaddressing unique threats introduced by LLMs in design automationFrom the articleThe integration of LLMs into EDA pipelines presents unique security challenges.Split ManufacturingContextleveraging 2.5D active interposers for robust physical isolationFrom the articleThe researchers highlight a powerful approach leveraging 2.5D split manufacturing and active interposers to achieve this necessary isolation.Enhanced Hardware SecurityOutcomea unified security analysis for the dual revolution in chip designFrom the article 2 mentionsThis symbiotic relationship between AI and hardware security is poised to shape the future of secure semiconductor development.

## The Proliferation of Vulnerabilities in Heterogeneous Systems

While chiplets promise benefits in yield and modularity, and LLMs enhance design productivity, they fundamentally expand the hardware attack surface across architectural, logical, and physical levels. This convergence introduces new vectors for exploitation that demand a unified security analysis.

## Fortifying Chiplet Systems with Isolated Roots of Trust

Securing chiplet architectures necessitates a robust defense strategy, particularly focusing on physically isolated Root of Trust (RoT) implementations. The researchers highlight a powerful approach leveraging 2.5D split manufacturing and active interposers to achieve this necessary isolation.

## Defending LLM-Driven EDA Pipelines Against Novel Threats

The integration of LLMs into EDA pipelines presents unique security challenges. This paper identifies native threats inherent to LLM-driven EDA security and reviews existing state-of-the-art defense techniques designed to mitigate these vulnerabilities. The analysis, found on [arXiv](https://arxiv.org/abs/2608.05063v1), underscores the critical need for proactive security measures in these advanced design flows.

## Advancing Hardware Security Through AI Collaboration

Looking forward, LLM systems themselves can be instrumental in advancing hardware security for modern systems, including complex chiplet designs. This symbiotic relationship between AI and hardware security is poised to shape the future of secure semiconductor development.

---

Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory.