Machine learning (ML) models can be trade secrets due to their development cost. Hence, they need protection against malicious forms of reverse engineering (e.g., in IP piracy). With a growing shift of ML to the edge devices, in part for performance and in part for privacy benefits, the models have become susceptible to the so-called physical side-channel attacks. ML being a relatively new target compared to cryptography poses the problem of side-channel analysis in a context that lacks published literature. The gap between the burgeoning edge-based ML devices and the research on adequate defenses to provide side-channel security for them thus motivates our study. Our work develops and combines different flavors of side-channel defenses for ML models in the hardware blocks. We propose and optimize the first defense based on Boolean masking . We first implement all the masked hardware blocks. We then present an adder optimization to reduce the area and latency overheads. Finally, we couple it with a shuffle-based defense. We quantify that the area-delay overhead of masking ranges from 5.4× to 4.7× depending on the adder topology used and demonstrate a first-order side-channel security of millions of power traces. Additionally, the shuffle countermeasure impedes a straightforward second-order attack on our first-order masked implementation.

A 1028challenge-response strong-PUF circuit is fabricated in 14nm CMOS, demonstrating modeling attack resistance across 6-million training samples. Two-stage nonlinearly cascaded PUF array organization with adversarial challenge selection increases modeling complexity by 4×, limiting ML attack accuracy to ~50%. A configurable cross-coupled inverter-based entropy source with stability-aware challenge pruning enables 9.8× higher array density, linearly-scaling enrollment data and worst-case bit-error-rate of 0.26% measured across 650-850mV, 0-100°C operation.

Low-area energy-efficient security primitives are key building blocks for enabling end-to-end content protection, user authentication in IoT platforms. This paper describes 3 designs that employ energy-efficient circuit techniques with optimal hardware-friendly arithmetic for seamless integration into area/battery constrained IoT systems: 1) A 2040-gate AES accelerator achieving 289Gbps/W efficiency in 22nm CMOS, 2) Hardened hybrid Physically Unclonable Function (PUF) circuit to generate a 100% stable encryption key. 3) All-digital TRNG to achieve >0.99 min-entropy with 3pJ/bit energy-efficiency.

Low-area energy-efficient security primitives are key building blocks for enabling end-to-end content protection, user authentication in IoT platforms. This paper describes 3 designs that employ energy-efficient circuit techniques with optimal hardware-friendly arithmetic for seamless integration into area/battery constrained IoT systems: 1) A 2040-gate AES accelerator achieving 289Gbps/W efficiency in 22nm CMOS, 2) Hardened hybrid Physically Unclonable Function (PUF) circuit to generate a 100% stable encryption key. 3) All-digital TRNG to achieve >0.99 min-entropy with 3pJ/bit energy-efficiency.

Low-area energy-efficient security primitives are key building blocks for enabling end-to-end content protection, user authentication, and consumer confidentiality in the IoT world that is estimated to surpass 50billion smart and connected devices by 2020. This paper describes design approaches that blend energy-efficient circuit techniques with optimal accelerator microarchitecture datapath, and hardware friendly arithmetic to achieve ultra-low energy consumption in security platforms for seamless adoption in area/battery constrained and self-powered systems. Industry leading energy-efficiency is demonstrated with three designs, fabricated and measured in advanced process technologies: 1) A 2040-gate arithmetically optimized composite-field Sbox based AES accelerator achieves 289Gbps/W peak energy-efficiency while offering 432Mbps throughput in 22nm tri-gate CMOS, 2) Hybrid Physically Unclonable Function (PUF) circuit leverages burn-in induced aging to reduce bit-error, followed by temporal-majority-voting, dark-bit masking, and error-correction conditioning techniques to generate a 100% stable full-entropy key with 190fJ/bit energy consumption in 22nm tri-gate CMOS. 3) A light-weight all digital TRNG uses in-line correlation suppressor and entropy-extractor circuits to achieve >0.99 min-entropy with 3pJ/bit measured energy-efficiency while operating down to 300mV in 14nm tri-gate CMOS.

Static random access memories (SRAM) are a major constituent in high performance microprocessors and systems-on-a-chip. With scaling of technology, manufacturing process variations in SRAMs are of significant concern. SRAM cells that are marginal due to process variations suffer from stability issues where random thermal noise and random telegraph noise (RTN) become determinant in memory state. This introduces uncertainty in testing, where passing a test does not necessarily mean that SRAM is good. A marginal cell may pass testing, but fail during operation. To address this problem, we introduce a new metric for probabilistic fault coverage. We present simulation methods for measuring such coverage and propose N-detect and multilevel wordline (WL) voltage techniques to increase the detection probability of marginal cells. We demonstrate the benefit of these testing approaches on a sample SRAM array, designed in 32 nm predictive technology models. To simulate marginal cells, we oversample the tail of the process variations distribution. Transient noise simulations show that thermal noise can lead to a fault coverage loss of 20% and RTN can lead to a maximum fault coverage loss of up to 75%. N-detect technique achieves close to a 100% fault coverage at n = 100 for thermal noise and n = 70 for RTN. Multilevel WL technique requires 20-30 mV boost in WL voltage during read test to achieve near ideal fault coverage at minimal yield loss. A combination of these techniques is shown to be effective in increasing fault coverage with a tradeoff between test time and yield loss.

From system-on-a-chip to high performance processors, SRAM is a critical component. In highly scaled CMOS devices, process variation is a major concern as it affects SRAM stability which often sets the floor on supply voltage and the ceiling on operating temperature of a semiconductor chip. Consequently, low-voltage and high temperature testing are often part of manufacturing test flow. In this paper, we show that for marginal cells, thermal noise is a major corrupting factor that affects the outcome of testing. A cell with large process variation which should ordinarily fail during memory test may pass due to impact of thermal noise at high temperature. To address this uncertainty during testing, we propose a stochastic metric for test coverage. We also propose application of N-detect and Multi-level Word Line (WL) techniques to improve test coverage based on this stochastic metric. Simulation studies on 32nm PTM models indicate varying probability of faulty bit detection across the spectrum of random thermal noise that lead to erroneous test results. Multiple accesses to each bit cell during test increases the fault coverage from -10% to near ideal 100%. Boosting WL voltage during read test and scaling it below nominal voltage during write test accelerates fault detection. Simulation of a 1KB SRAM array test case shows an improvement in fault coverage from -88% to 100% by increasing the number of detects to 100.

Hardware Trojans have become a growing concern in the design of secure integrated circuits. In this work, we present a set of novel hardware Trojans aimed at evading detection methods, designed as part of the CSAW Embedded System Challenge 2010. We introduced and implemented unique Trojans based on side-channel analysis that leak the secret key in the reference encryption algorithm. These side-channel-based Trojans do not impact the functionality of the design to minimize the possibility of detection. We have demonstrated the statistical analysis approach to attack such Trojans. Besides, we introduced Trojans that modify either the functional behavior or the electrical characteristics of the reference design. Novel techniques such as a Trojan draining the battery of a device do not have an immediate impact and hence avoid detection, but affect the long term reliability of the system.

Advancing technology nodes have increased the impact of lithography variation on design yield and performance. Although lithographic distortion is emerging as the dominant mode of failure, particulate defect still remains an important source of defect. In this work, we present a novel critical area analysis technique that deals with non-deterministic line edges that arise due to statistical lithography variations. Previous critical area analysis techniques are based on fixed printed structures with variable defect sizes. The proposed solution integrates lithographic variations within critical area analysis, using available tools while keeping the computational complexity in check. The key optimizations include weighted stratified sampling of important lithographic parameters and Monte Carlo based Probability of Failure (POF) calculation that is scalable to large chips. Simulation results on ISCAS benchmarks show that inclusion of lithography induced physical variations can increase critical area by as much as 2X, while overall yield due to particulate defects in large designs may decrease by more than 5%; validating both the need for an integrated solution and its feasibility on large designs.