A Simulation-Driven Dataset for Anomaly Detection Targeting RISC-V ML Accelerators RISC-V ML Hizlandiricilarina Yönelik Anomali Tespiti için Simülasyon Tabanli Bir Veri Kümesi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636646
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: Anomaly detection, ML accelerators, RISC-V, Simulation-based dataset, Software-defined vehicle (SDV), Telemetry monitoring
- Marmara Üniversitesi Adresli: Evet
Özet
With the growing adoption of the software-defined vehicle (SDV) approach in automotive systems, RISC-V-based ML accelerators are increasingly used in latency-sensitive and reliability-critical settings. This makes early run-time anomaly detection critical. However, under conditions where access to hardware is limited and large-scale telemetry collection is difficult, obtaining reproducible and labeled anomaly data is challenging. In this work, we present a simulation-based dataset that combines window-level statistics from a gem5 RISC-V simulator with power and energy metrics estimated using McPAT. The dataset covers 6 ML-like workloads, 5 anomaly types, and multiple hardware configurations; for each window, it provides telemetry and anomaly labels. The dataset aims to provide a reproducible basis for training and evaluating anomaly detection models targeting RISC-V-based ML accelerators and to support analysis of how different telemetry feature groups respond to controlled run-time anomalies.