Software-Hardware Codesign for Efficient In-Memory Regular Pattern Matching
summary
Provides a conservative static analysis that labels counted regex regions as easy (single-register) or ambiguous (needs per-state vector), compiles regexes into non-deterministic counter automata (NCA), and extends a COMMA-like in-memory accelerator with 17-bit counter modules and shiftable bit vector modules (200–2000 bits). Analysis runs under a second on real datasets; NCA execution cuts memory (10 bits vs 1000 states) and the prototype lowers energy up to 76% and area up to 58% on benchmarks.
AI-written summary · about
chapters (4)
related talks
19mHybridSA: GPU Acceleration of Multi-Pattern Regex Matching using Bit Parallelism
17mA Difference World: High-performance, NVM-invariant, Software-only Intermittent...
32mSecSi Product Development
20mACES: Automatic Compartments for Embedded Systems
45mHackathon Presentations
23mRunning an Open Source LLM on Open Source Hardware
48mCapturing 0Day Exploits With Perfectly Placed Hardware Traps
10mAI custom Software/Hardware Interface improving performance 5-10x