跳到正文
返回论文列表
cs.CR提交于 已译

基于 LLM 智能体的已验证网表到 RTL 恢复

Verified Netlist-to-RTL Recovery with LLM Agents

Ahmet Efe · Recep Bugra Uludag · Ulya R. Karpuzcu

中文摘要

门级网表在 IP 集成商与晶圆厂之间流通,也可从芯片中提取。将其逆向工程为可读的寄存器传输级(RTL)代码有助于攻击者理解和修改设计,但由于字级操作分散在大量门中,这一过程仍然困难。大语言模型(LLM)能编写和优化 RTL,但往往无法从整个网表恢复 RTL。提出 LOCA(LOgic Cone Attack),一种基于 LLM 的攻击方法:将匿名化网表划分为逻辑锥,由并行智能体重写各逻辑锥,再组装成紧凑 RTL。每个逻辑锥包含一组相关寄存器的完整次态逻辑,为智能体提供小而明确的任务。测试的五种划分方法中,仅分组逻辑锥能实现恢复。LOCA 只保留经证明与网表等价的 RTL,并以压缩率衡量其质量。在测试设计上,保功能混淆无法阻止 LOCA,而未知密钥或隐藏单元功能则使其无法证明恢复了原始设计。在 21 个模型-设计组合中,LOCA 恢复 17 个,而整网表提示仅恢复 6 个。跨模型单次运行恢复率从 5% 升至 74%,使用低成本模型时中位成本低于 1 美元。

关键要点

  1. 01问题:门级网表逆向为 RTL 困难,字级操作分散在大量门中,LLM 难以从整网表直接恢复。
  2. 02方法:LOCA 将网表划分为逻辑锥,并行 LLM 智能体重写各锥并组装紧凑 RTL,仅保留经证明等价的 RTL。
  3. 03结果:21 个模型-设计组合中恢复 17 个,整网表提示仅 6 个;单次运行恢复率从 5% 升至 74%。
  4. 04成本:使用低成本模型时中位成本低于 1 美元。
  5. 05局限:未知密钥或隐藏单元功能会阻止 LOCA 证明恢复原始设计。

解读

尚无解读。

原始英文摘要

arXiv:2611.25324v1 Announce Type: new Abstract: Gate-level netlists circulate among IP integrators and foundries and can be extracted from chips. Reverse engineering them into readable register-transfer-level (RTL) code would help an adversary understand and modify a design, but remains difficult because word-level operations are scattered across many gates. Large language models (LLMs) can write and optimize RTL, but often fail to recover it from a whole netlist. We propose LOCA (LOgic Cone Attack), an LLM-based attack that divides an anonymized netlist into logic cones, has parallel agents rewrite them, and assembles compact RTL. Each cone contains the complete next-state logic of a group of related registers, giving an agent a small, meaningful task. Of five partitioning methods tested, only grouped cones lead to recovery. LOCA keeps only RTL proven equivalent to the netlist and measures its quality by compression. Function-preserving obfuscation does not stop LOCA on the tested designs, whereas unknown keys or hidden cell functions prevent it from proving recovery of the original design. Across 21 model-design pairs, LOCA recovers 17, versus 6 with whole-netlist prompting. Per-run recovery across models rises from 5% to 74%, at a median cost below one dollar with a low-cost model.

同方向论文 · cs.CR

查看全部 →

反向发现 · 这篇论文相关的工具与提示词

去交叉发现 →

关联提示词 · 1

编程github.com/f/awesome-chatgpt-prompts·外部收录·

CLAUDE.md Generator for AI Coding Agents

You are a CLAUDE.md architect — an expert at writing concise, high-impact project instruction files for AI coding agents (Claude Code, Cursor, Windsurf, Zed, etc.). Your task: Generate a production-ready CLAUDE.md file based on the project…