Pre-Fabrication to Post-Deployment: A Unified ML-Driven Pipeline for IC Logic Locking
This paper presents a unified, four-stage machine learning-driven pipeline that automates the selection of optimal logic locking techniques, integrates a lightweight and cryptographically robust STREAMLOCK mechanism with Dynamic Partial Reconfiguration for runtime updates, and achieves superior hardware efficiency and security across the entire IC protection lifecycle.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the modern world, the chips that power everything from smartphones to medical devices are rarely made by a single company in a single place. Instead, the design, verification, and manufacturing of these integrated circuits are split among different teams and factories across the globe. While this global cooperation has made technology cheaper and faster to produce, it has also created a dangerous vulnerability. Untrusted factories or third-party suppliers could steal the intellectual property, insert malicious hardware, or reverse-engineer the designs to create unauthorized copies. To stop this, engineers use a technique called logic locking. Imagine a circuit as a complex machine; logic locking adds a secret switchboard that keeps the machine from working unless the correct secret code is entered. Without this code, the machine looks intact but produces garbage results. However, for years, this defense has had three major flaws. First, choosing the right type of lock for a specific chip has been a manual, guesswork-heavy process. Second, once the chip is built, the lock is static and unchanging, making it vulnerable to attackers who study it over time. Third, there has been no good way to update the security after the chip is already in the field.
A team of researchers at the University of North Carolina at Charlotte has built a new system to fix all three problems at once. They created a unified pipeline that uses machine learning to automatically design, build, and maintain secure chips from the factory floor to the final user. The process begins with a computer program that acts like an expert architect. It examines the raw blueprint of a circuit and, using a trained artificial intelligence model, predicts which of five different locking strategies will work best for that specific design. This removes the need for human engineers to manually decide on a security method, ensuring the right tool is used for the job every time.
Once the best strategy is chosen, the system builds the lock. For many standard circuits, it uses a method that inserts secret gates at the most critical points to maximize confusion for an attacker. For circuits that need stronger protection, the system employs a new, lightweight security engine called STREAMLOCK. Unlike traditional locks that use a single, unchanging secret code, STREAMLOCK generates a constantly shifting stream of random numbers. This stream is mixed into the circuit's logic, meaning the relationship between the input and the output is never the same twice. This dynamic behavior defeats attackers who try to study the chip's fixed behavior to crack the code. The researchers tested this new engine on a variety of standard benchmark circuits and confirmed that it works perfectly when the correct secret is applied, while completely scrambling the output when the wrong key is used.
The researchers also compared this new approach against older methods and against using a heavy-duty encryption standard known as AES, which is often used for high-security data but is very expensive in terms of chip space and power. Their tests showed that the new STREAMLOCK method offers much stronger security than the older, static locks, but it uses up to nine times less space and twenty times less power than the heavy AES method. This places it in a sweet spot: it is far more secure than the old ways but far more efficient than the heavy-duty alternatives.
The final piece of the puzzle addresses the problem of chips that are already in use. Even a strong lock can eventually be broken if it never changes. To solve this, the team integrated a feature called dynamic partial reconfiguration. This allows the security settings of the chip to be updated while it is running, without shutting the system down. It is similar to changing the combination on a safe while the safe is still locked and in use, rather than having to break the safe open to reset it. The researchers found that adding this ability to update the security on the fly adds almost no extra cost to the chip's size or power consumption. By combining automatic design selection, a powerful new encryption engine, and the ability to update security in real-time, this pipeline offers a complete, automated solution for protecting hardware throughout its entire life, from the moment it is designed to the day it is finally retired.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.