[Paper Review] Automatic Extraction of Secrets from the Transistor Jungle using Laser-Assisted Side-Channel Attacks

Thilo Krachenfels, Tüba Kiyan+2 more|arXiv (Cornell University)|Feb 23, 2021

Physical Unclonable Functions (PUFs) and Hardware Security51 references4 citations

TL;DR

This paper presents a novel, automated method to extract secret keys from hardware root-of-trust (RoT) devices using laser-assisted side-channel attacks combined with deep learning. By training convolutional neural networks (CNNs) on black-box measurements from a known device, the approach enables high-accuracy key extraction from unknown victim devices without prior knowledge of the IC layout or memory cell structure, demonstrating success across three distinct hardware platforms including 20 nm FPGAs and 180 nm microcontrollers.

ABSTRACT

The security of modern electronic devices relies on secret keys stored on secure hardware modules as the root-of-trust (RoT). Extracting those keys would break the security of the entire system. As shown before, sophisticated side-channel analysis (SCA) attacks, using chip failure analysis (FA) techniques, can extract data from on-chip memory cells. However, since the chip's layout is unknown to the adversary in practice, secret key localization and reverse engineering are onerous tasks. Consequently, hardware vendors commonly believe that the ever-growing physical complexity of the integrated circuit (IC) designs can be a natural barrier against potential adversaries. In this work, we present a novel approach that can extract the secret key without any knowledge of the IC's layout, and independent from the employed memory technology as key storage. We automate the -- traditionally very labor-intensive -- reverse engineering and data extraction process. To that end, we demonstrate that black-box measurements captured using laser-assisted SCA techniques from a training device with known key can be used to profile the device for a later key prediction on other victim devices with unknown keys. To showcase the potential of our approach, we target keys on three different hardware platforms, which are utilized as RoT in different products.

Motivation & Objective

• To address the high cost and labor intensity of reverse engineering ICs for side-channel attacks.
• To overcome the limitation of requiring detailed knowledge of chip layout and memory cell design in traditional side-channel attacks.
• To automate the key extraction process using machine learning on laser-based side-channel measurements.
• To demonstrate that secret keys can be extracted from diverse hardware platforms without prior knowledge of their internal structure.
• To enable scalable, generic key extraction applicable to real-world RoT devices used in critical systems.

Proposed method

• Leverage laser-assisted side-channel analysis (SCA) techniques—thermal laser stimulation (TLS) and laser logic state imaging (LLSI)—to capture device activity on a training device with known secret keys.
• Collect black-box measurements from the training device using laser scanning microscopy (LSM), capturing transient responses correlated with internal logic states.
• Train convolutional neural networks (CNNs) on these labeled measurements to learn patterns associated with specific bit values in memory cells.
• Apply the trained CNN models to victim devices with unknown keys, using the same measurement setup without retraining.
• Use the trained models to predict secret key values directly from raw LSM images, bypassing manual reverse engineering of layout or cell structure.
• Validate the approach across three different hardware platforms: 20 nm FPGA, 180 nm microcontroller, and 60 nm FPGA, all used as RoT components.

Experimental results

Research questions

• RQ1 Can deep learning models trained on laser-based side-channel measurements from a known device accurately predict secret keys on unrelated victim devices without layout knowledge?
• RQ2 To what extent can CNNs generalize across different IC process technologies and memory types to extract secrets from unknown devices?
• RQ3 Does the proposed method eliminate the need for manual reverse engineering of IC layouts and memory cell geometries in side-channel attacks?
• RQ4 How robust is the key extraction process in the presence of irrelevant chip activity and noise?
• RQ5 Can this approach be generalized to other FA microscopy techniques beyond optical SCA?

Key findings

• The proposed method successfully extracted secret keys from three distinct hardware platforms—20 nm FPGA, 180 nm microcontroller, and 60 nm FPGA—without any knowledge of the IC layout or memory cell design.
• Trained CNN models achieved high-accuracy key prediction on victim devices, even when the devices exhibited significant background activity and irrelevant circuit operations.
• The approach eliminated the need for manual reverse engineering of memory cell locations and bit interpretation, significantly reducing attack complexity and time.
• The method is generic and scalable, applicable across different process technologies and memory types, including SRAM and embedded non-volatile memory.
• The approach remains effective even when the attacker does not know the exact location of targeted memory cells or how to decode bit values from raw measurements.
• The study demonstrates that deep learning can automate and generalize laser-based side-channel attacks, reducing the adversary’s effort to a minimal setup and training phase.