This chapter is on the security assessment of artificial intelligence (AI) and neural network (NN) accelerators in the face of fault injection attacks. More specifically, it discusses the assets on these platforms and compares them with ones known and well-studied in the field of cryptographic systems. This is a crucial step that must be taken in order to define the threat models precisely. With respect to that, fault attacks mounted on NNs and AI accelerators are explored.
Technische Universität Wien (TU Wien)（维也纳工业大学）；New York University Abu Dhabi (NYUAD)（纽约大学阿布扎比分校）
From tiny pacemaker chips to aircraft collision avoidance systems, the state-of-the-art Cyber-Physical Systems (CPS) have increasingly started to rely on Deep Neural Networks (DNNs). However, as concluded in various studies, DNNs are highly susceptible to security threats, including adversarial attacks. In this paper, we first discuss different vulnerabilities that can be exploited for generating security attacks for neural network-based systems. We then provide an overview of existing adversarial and fault-injection-based attacks on DNNs. We also present a brief analysis to highlight differen…
Injection of transient faults as a way to attack cryptographic implementations has been largely studied in the last decade. Several attacks that use electromagnetic fault injection against hardware or software architectures have already been presented. On microcontrollers, electromagnetic fault injection has mostly been seen as a way to skip assembly instructions or subroutine calls. However, to the best of our knowledge, no precise study about the impact of an electromagnetic glitch fault injection on a microcontroller has been proposed yet. The aim of this paper is twofold: providing a more …
As Field-programmable gate arrays (FPGAs) are widely adopted in clouds to accelerate Deep Neural Networks (DNN), such virtualization environments have posed many new security issues. This work investigates the integrity of DNN FPGA accelerators in clouds. It proposes DeepStrike, a remotely-guided attack based on power glitching fault injections targeting DNN execution. We characterize the vulnerabilities of different DNN layers against fault injections on FPGAs and leverage time-to-digital converter (TDC) sensors to precisely control the timing of fault injections. Experimental results show th…
Fault attacks consist in changing the program behavior by injecting faults at run-time in order to break some expected security properties. Applications are hardened against fault attack adding countermeasures. According to the state of the art, applications must now be protected against multi-fault injection. As a consequence developing applications which are robust becomes a very challenging task, in particular because countermeasures can be also the target of attacks. The aim of this paper is to propose an assisted methodology for developers allowing to harden an application against multi-f…
The impeccable design of sensitive and cryptographic circuits (CC) against fault injection attacks is essential for modern data storage, communication, and computation systems that are susceptible to fault injection attacks. The robustness of a CC against voltage glitch attacks increases with an on-chip voltage regulator that considers the impact of topology and component selection on the fault injection robustness. With an emphasis on increasing the number of phases in a multiphase voltage regulator and component selection, an on-chip voltage regulator with multiphase configuration and an opt…
We propose HASHTAG, the first framework that enables high-accuracy detection of fault-injection attacks on Deep Neural Networks (DNNs) with provable bounds on detection performance. Recent literature in fault-injection attacks shows the severe DNN accuracy degradation caused by bit flips. In this scenario, the attacker changes a few weight bits during DNN execution by tampering with the program's DRAM memory. To detect runtime bit flips, HASHTAG extracts a unique signature from the benign DNN prior to deployment. The signature is later used to validate the integrity of the DNN and verify the i…
Fault injections are increasingly used to attack/test secure applications. In this paper, we define formal models of runtime monitors that can detect fault injections that result in test inversion attacks and arbitrary jumps in the control flow. Runtime verification monitors offer several advantages. The code implementing a monitor is small compared to the entire application code. Monitors have a formal semantics; and we prove that they effectively detect attacks. Each monitor is a module dedicated to detecting an attack and can be deployed as needed to secure the application. A monitor can ru…
Hardware and software of secured embedded systems are prone to physical attacks. In particular, fault injection attacks revealed vulnerabilities on the data and the control flow allowing an attacker to break cryptographic or secured algorithms implementations. While many research studies concentrated on successful attacks on the data flow, only a few targets the instruction flow. In this paper, we focus on electromagnetic fault injection (EMFI) on the control flow, especially on the instruction cache. We target the very widespread (smartphones, tablets, settop-boxes, health-industry monitors a…
In this article, we describe a methodology that aims at either breaking or proving the security of CRT-RSA implementations against fault injection attacks. In the specific case-study of the BellCoRe attack, our work bridges a gap between formal proofs and implementation-level attacks. We apply our results to three implementations of CRT-RSA, namely the unprotected one, that of Shamir, and that of Aumüller et al. Our findings are that many attacks are possible on both the unprotected and the Shamir implementations, while the implementation of Aumüller et al. is resistant to all single-fault att…
The IoT consists of a lot of devices such as embedded systems, wireless sensor nodes (WSNs), control systems, etc. It is essential for some of these devices to protect information that they process and transmit. The issue is that an adversary may steal these devices to gain a physical access to the device. There is a variety of ways that allows to reveal cryptographic keys. One of them are optical Fault Injection attacks. We performed successful optical Fault Injections into different type of gates, in particular INV, NAND, NOR, FF. In our work we concentrate on the selection of the parameters…
Silicon Austria Labs（奥地利硅实验室）；Graz University of Technology（格拉茨技术大学）；Nanyang Technological University（南洋理工大学）；Slovak University of Technology（斯洛伐克技术大学）
Neural networks have been shown to be vulnerable against fault injection attacks. These attacks change the physical behavior of the device during the computation, resulting in a change of value that is currently being computed. They can be realized by various fault injection techniques, ranging from clock/voltage glitching to application of lasers to rowhammer. In this paper we explore the possibility to reverse engineer neural networks with the usage of fault attacks. SNIFF stands for sign bit flip fault, which enables the reverse engineering by changing the sign of intermediate values. We de…
Resistive Random Access Memory (RRAM) is a type of Non-Volatile Memory (NVM). In this paper we investigate the sensitivity of the TiN/Ti/Al:HfO2/TiN-based 1T-1R RRAM cells implemented in a 250 nm CMOS IHP technology to the laser irradiation in detail. Experimental results show the feasibility to influence the state of the cells under laser irradiation, i.e. successful optical Fault Injection. We focus on the selection of the parameters of the laser station and their influence on the success of optical Fault Injections.
Voltage Fault Injection (VFI), also known as power glitching, has proven to be a severe threat to real-world systems. In VFI attacks, the adversary disturbs the power-supply of the target-device forcing the device to illegitimate behavior. Various countermeasures have been proposed to address different types of fault injection attacks at different abstraction layers, either requiring to modify the underlying hardware or software/firmware at the machine instruction level. Moreover, only recently, individual chip manufacturers have started to respond to this threat by integrating countermeasures…
Today's mobile devices contain densely packaged system-on-chips (SoCs) with multi-core, high-frequency CPUs and complex pipelines. In parallel, sophisticated SoC-assisted security mechanisms have become commonplace for protecting device data, such as trusted execution environments, full-disk and file-based encryption. Both advancements have dramatically complicated the use of conventional physical attacks, requiring the development of specialised attacks. In this survey, we consolidate recent developments in physical fault injections and side-channel attacks on modern mobile devices. In total,…
Electromagnetic fault injection (EMFI) is a well known technique used to disturb the behaviour of a chip for weakening its security. These attacks are mostly done on simple microcontrollers. On these targets, the fault effects are relatively simple and understood. Exploiting EMFI on modern system-on-chips (SoCs), the fast and complex chips ubiquitous today, requires to understand the impact of such faults. In this paper, we propose an experimental setup and a forensic process to create exploitable faults and assess their impact on the SoC micro-architecture. On our targeted SoC (a BCM2837), th…
The outsourcing of the design and manufacturing of integrated circuits has raised severe concerns about the piracy of Intellectual Properties and illegal overproduction. Logic locking has emerged as an obfuscation technique to protect outsourced chip designs, where the circuit netlist is locked and can only be functional once a secure key is programmed. However, Boolean Satisfiability-based attacks have shown to break logic locking, simultaneously motivating researchers to develop more secure countermeasures. In this paper, we present a novel fault injection attack to break any locking techniq…
Fault injection attacks can cause errors in software for malicious purposes. Oftentimes, vulnerable points of a program are detected after its development. It is therefore critical for the user of the program to be able to apply last-minute security assurance to the executable file without having access to the source code. In this work, we explore two methodologies based on binary rewriting that aid in injecting countermeasures in the binary file. The first approach injects countermeasures by reassembling the disassembly whereas the second approach leverages a full translation to a high-level …
Physically accessible devices such as sensor nodes in Wireless Sensor Networks or "smart" devices in the Internet of Things have to be resistant to a broad spectrum of physical attacks, for example to Side Channel Analysis and to Fault Injection attacks. In this work we concentrate on the vulnerability of ASICs to precise optical Fault Injection attacks. Here we propose to use metal fillers as potential low-cost countermeasure that may be effective against a broad spectrum of physical attacks. In our future work we plan to evaluate different methods of metal fillers placement, to select an eff…
If devices are physically accessible optical fault injection attacks pose a great threat since the data processed as well as the operation flow can be manipulated. Successful physical attacks may lead not only to leakage of secret information such as cryptographic private keys, but can also cause economic damage especially if as a result of such a manipulation a critical infrastructure is successfully attacked. Laser based attacks exploit the sensitivity of CMOS technologies to electromagnetic radiation in the visible or the infrared spectrum. It can be expected that radiation-hard designs, sp…
Nanyang Technological University（南洋理工大学）；Harbin Institute of Technology（哈尔滨工业大学）
As deep learning systems are widely adopted in safety- and security-critical applications, such as autonomous vehicles, banking systems, etc., malicious faults and attacks become a tremendous concern, which potentially could lead to catastrophic consequences. In this paper, we initiate the first study of leveraging physical fault injection attacks on Deep Neural Networks (DNNs), by using laser injection technique on embedded systems. In particular, our exploratory study targets four widely used activation functions in DNNs development, that are the general main building block of DNNs that crea…
Side-channel and fault injection attacks reveal secret information by monitoring or manipulating the physical effects of computations involving secret variables. Circuit-level countermeasures help to deter these attacks, and traditionally such countermeasures have been developed for each attack vector separately. We demonstrate a multipurpose ring oscillator design - Programmable Ring Oscillator (PRO) to address both fault attacks and side-channel attacks in a generic, application-independent manner. PRO, as an integrated primitive, can provide on-chip side-channel resistance, power monitoring…
In this paper we study the existing CRT-RSA countermeasures against fault-injection at-tacks. In an attempt to classify them we get to achieve deep understanding of how they work. We show that the many countermeasures that we study (and their variations) actually share a number of common features, but optimize them in different ways. We also show that there is no conceptual distinction between test-based and infective countermeasures and how either one can be transformed into the other. Furthermore, we show that faults on the code (skipping instructions) can be captured by considering only fau…
Machine Learning applications are acknowledged at the foundation of autonomous driving, because they are the enabling technology for most driving tasks. However, the inclusion of trained agents in automotive systems exposes the vehicle to novel attacks and faults, that can result in safety threats to the driv-ing tasks. In this paper we report our experimental campaign on the injection of adversarial attacks and software faults in a self-driving agent running in a driving simulator. We show that adversarial attacks and faults injected in the trained agent can lead to erroneous decisions and se…
In this paper, we investigate the advanced circuit features such as wordline- (WL) underdrive (prevents retention failure) and overdrive (assists write) employed in the peripherals of Dynamic RAM (DRAM) memories from a security perspective. In an ideal environment, these features ensure fast and reliable read and write operations. However, an adversary can re-purpose them by inserting Trojans to deliver malicious payloads such as fault injections, Denial-of-Service (DoS), and information leakage attacks when activated by the adversary. Simulation results indicate that wordline voltage can be i…
Logic Locking is a well-accepted protection technique to enable trust in the outsourced design and fabrication processes of integrated circuits (ICs) where the original design is modified by incorporating additional key gates in the netlist, resulting in a key-dependent functional circuit. The original functionality of the chip is recovered once it is programmed with the secret key, otherwise, it produces incorrect results for some input patterns. Over the past decade, different attacks have been proposed to break logic locking, simultaneously motivating researchers to develop more secure coun…
Characterizing the Fault Response of the Intel Neural Compute Stick 2 Under Single-Pulse Electromagnetic Fault Injection
Faculty of Informatics and Information Technologies, Slovak University of Technology（信息与信息技术学院，斯洛伐克技术大学）；State Key Laboratory of Blockchain and Data Security（区块链与数据安全国家重点实验室）；TTControl GmbH
Injection of transient faults can be used as a way to attack embedded systems. On embedded processors such as microcontrollers, several studies showed that such a transient fault injection with glitches or electromagnetic pulses could corrupt either the data loads from the memory or the assembly instructions executed by the circuit. Some countermeasure schemes which rely on temporal redundancy have been proposed to handle this issue. Among them, several schemes add this redundancy at assembly instruction level. In this paper, we perform a practical evaluation for two of those countermeasure sc…
Fault attacks against embedded circuits enabled to define many new attack paths against secure circuits. Every attack path relies on a specific fault model which defines the type of faults that the attacker can perform. On embedded processors, a fault model consisting in an assembly instruction skip can be very useful for an attacker and has been obtained by using several fault injection means. To avoid this threat, some countermeasure schemes which rely on temporal redundancy have been proposed. Nevertheless, double fault injection in a long enough time interval is practical and can bypass th…
Gate camouflaging is a known security enhancement technique that tries to thwart reverse engineering by hiding the functions of gates or the connections between them. A number of works on SAT-based attacks have shown that it is often possible to reverse engineer a circuit function by combining a camouflaged circuit model and the ability to have oracle access to the obfuscated combinational circuit. Especially in small circuits it is easy to reverse engineer the circuit function in this way, but SAT-based reverse engineering techniques provide no guarantees of recovering a circuit that is gate-…
Bit-slicing is a software implementation technique that treats an N-bit processor datapath as N parallel single-bit datapaths. The natural spatial redundancy of bit-sliced software can be used to build countermeasures against implementation attacks. While the merits of bit-slicing for side-channel countermeasures have been studied before, their application for protection of post-quantum algorithms against fault injection is still unexplored. We present an end-to-end analysis of the efficacy of bit-slicing to detect and thwart electromagnetic fault injection (EMFI) attacks on post-quantum crypt…
Analysis of LLM Vulnerability to GPU Soft Errors: An Instruction-Level Fault Injection Study
School of Computer Science (National Pilot Software Engineering School), Beijing University of Posts and Telecommunications（计算机学院（国家试点软件工程学院），北京邮电大学）；State Key Lab of Processors, Institute of Computing Technology, Chinese Academy of Sciences（处理器国家重点实验室，中国科学院计算技术研究所）
AMD Secure Encrypted Virtualization (SEV) offers protection mechanisms for virtual machines in untrusted environments through memory and register encryption. To separate security-sensitive operations from software executing on the main x86 cores, SEV leverages the AMD Secure Processor (AMD-SP). This paper introduces a new approach to attack SEV-protected virtual machines (VMs) by targeting the AMD-SP. We present a voltage glitching attack that allows an attacker to execute custom payloads on the AMD-SPs of all microarchitectures that support SEV currently on the market (Zen 1, Zen 2, and Zen 3…
GPU clouds have become a popular computing platform because of the cost of owning and maintaining high-performance computing clusters. Many cloud architectures have also been proposed to ensure a secure execution environment for guest applications by enforcing strong security policies to isolate the untrusted hypervisor from the guest virtual machines (VMs). In this paper, we study the impact of GPU chip's hardware faults on the security of cloud "trusted" execution environment using Deep Neural Network (DNN) as the underlying application. We show that transient hardware faults of GPUs can be …
