SPICED: Syntactical Bug and Trojan Pattern Identification in A/MS Circuits using LLM-Enhanced Detection

TL;DR

This work proposes SPICED, a Large Language Model (LLM)-based framework that operates within the software domain, eliminating the need for hardware modifications for Trojan detection and localization, and is the first work using LLM-aided techniques for detecting and localizing syntactical bugs and analog Trojans in circuit netlists.

Abstract

Analog and mixed-signal (A/MS) integrated circuits (ICs) are crucial in modern electronics, playing key roles in signal processing, amplification, sensing, and power management. Many IC companies outsource manufacturing to third-party foundries, creating security risks such as stealthy analog Trojans. Traditional detection methods, including embedding circuit watermarks or conducting hardware-based monitoring, often impose significant area and power overheads, and may not effectively identify all types of Trojans. To address these shortcomings, we propose SPICED, a Large Language Model (LLM)-based framework that operates within the software domain, eliminating the need for hardware modifications for Trojan detection and localization. This is the first work using LLM-aided techniques for detecting and localizing syntactical bugs and analog Trojans in circuit netlists, requiring no explicit training and incurring zero area overhead. Our framework employs chain-of-thought reasoning and few-shot examples to teach anomaly detection rules to LLMs. With the proposed method, we achieve an average Trojan coverage of 93.32% and an average true positive rate of 93.4% in identifying Trojan-impacted nodes for the evaluated analog benchmark circuits. These experimental results validate the effectiveness of LLMs in detecting and locating both syntactical bugs and Trojans within analog netlists.

Figures (4)

• Figure 1: (a) Enhancing syntactical bug detection accuracy of LLM by instruction-following approach (b) Identifying and mitigating bugs in a SPICE netlist.
• Figure 2: (a) An example prompt highlighting the SPICE syntax rules for bug detection and localization (b) LLM correctly detects the bugs injected in the netlist; however, it incorrectly flags Lines 6 and 7 as bugs (highlighted in red) (c) Explicitly updating the rules in the prompt to reduce false positives.
• Figure 3: (a) Flow of supervised-learning approach using machine learning (ML) models (b) Using the supervised-learning analogy to locate Trojan-impacted nodes using LLM ($V_{out}$: primary output voltage, $Spec$: desired output voltage specifications, $V^{node}_i (I^{node}_i$): node voltage (current) corresponding to $i^{th}$ input voltage sample $V^{in}_i$).
• Figure 4: Voltage deviation analysis for A2 Trojan-inserted netlist of circuit '642' from AMSNet amsnet. Voltage deviation of a node $x$ is given by $V_i^{x}-V_{i-1}^x$, where $i-1$ and $i$ are consecutive input voltage samples. (a) Analyze voltage deviation of each node across the 'Normal Input' and 'Trojan-Activation Input' ranges, (b) analyze voltage deviation across nodes in the 'Trojan-Activation Input' range. Combining (a) and (b), we observe that node 37 is a Trojan-impacted node.