Regular Session II
summary
Session showcases four concrete results: an RTL evaluation flow that integrates with ProSmart and Garnet 2.0, finds Sinful’s single-seed variability, and introduces Sinful RTL that cuts simulation time by ~40x while matching cycle-accurate results within <1% error (benchmarked up to 400M cycles); an ML detector trained with Garnet/gem5 features (packet hop counts, src/dst, packet types) using XGBoost that flags eavesdropping with high accuracy across mappings; a new attack class BHT that falsely blames routers and a CBHD defense that cuts latency overhead from ~9.8% to ~2.6%; and Noiser, a Karma-based address encryption + remap scheme showing ~92% drop in contention variance with ~7.4% avg performance loss.
AI-written summary · about
chapters (10)
- 0:00Start
- 10:25Average Packet Latency
- 17:42Eavesdropping Attack Detection Using Machine Learning in Network on Chip Architectures
- 18:04Why Eavesdropping Attacks Are Important
- 19:21Why the Security of Network Concept Is Important
- 20:13Eavesdropping Attack
- 21:37Machine Learning Based Attack Detection Methodology
- 41:42Consequences of Bht
- 56:12Noisier Mitigating Knock Side Channel Attacks Using Encrypted Addresses and Remapping
- 1:00:21Time Based Partitioning
related talks
47mRSAC Security Scholar Poster Pitch-Off
01h 05mTheory and Security
45mIntegrated Circuit Offensive Security
56mFIRSTCON21 - June 8 | Stream 3 Part 1 (12:00-13:00 UTC)
55mThis Month In Security Episode 10: April, 2023 - Edition
19mDefending against Traffic Analysis Attacks with Flexible In-Network Obfuscation
55mReliable and Secure Embedded Systems
46mRSAC Security Scholar Poster Pitch-Off