SeqCube-HT: state-cube-guided generation of sequential hardware Trojan benchmarks

Abstract Hardware Trojan (HT) detector evaluation depends on benchmarks whose triggers are executable from a defined reset state and whose payload effects reach an observation boundary. Establishing both properties is difficult for sequential Trojans because rare events are coupled through state evolution across cycles. We present SeqCube-HT, a state-cube graph framework that incorporates temporal execution evidence into benchmark construction. Reset-consistent traces are compressed into pre- and post-state cubes, and directed compatibility organizes candidate event sequences. Selected paths are then completed into concrete inputs and paired with sensitized payload paths. An instance is retained as witness-validated only after active and inactive replay and a bounded audit; the evaluated implementation materializes replay vectors, labels, logs, and audit records for that instance. Across four ISCAS’89 circuits and five Extended IP designs, SeqCube-HT produced 3240 witness-validated instances from 4500 attempts (72.0%). Under a matched protocol, the best-performing comparator implementation achieved 50.6% witness-validation yield and 39.5 s per witness-validated instance, compared with 72.0% and 20.9 s for SeqCube-HT. Covariate-matched detector experiments retained insertion-source differences under the evaluated graph probes. These results show that trace-backed temporal construction can efficiently produce witness-validated sequential HT benchmarks under the evaluated trace budget and recorded-execution validation protocol.