RAP: Reconfigurable Automata Processor
Ziyuan Wen, Alexis Le Glaunec, Konstantinos Mamouras, Kaiyuan Yang · 2025
Regular pattern matching is essential for applications such as text processing, malware detection, network security, and bioinformatics.Recent in-memory automata processors have significantly advanced the energy and memory efficiency over conventional computing platforms.However, these processors are typically optimized only for one type of automata, limiting their capability to efficiently support regex processing under diverse real-world workloads.This paper presents RAP, the first reconfigurable in-memory automata processor for efficient regular pattern matching across diverse workloads.It supports Nondeterministic Finite Automata (NFA), Nondeterministic Bit Vector Automata (NBVA), and Linear NFA (LNFA) through reconfigurable architecture and circuit designs, and a compiler for translation.RAP is evaluated in 28nm CMOS PDK, achieving 1.2-1.5×higher energy efficiency and 1.3-2.5×higher compute density compared to SotA automata processors for NFA (CA and CAMA) over diverse real-world benchmarks.It also achieves 1.6× higher compute density and similar energy efficiency as BVAP, a SotA optimized for bounded repetitions.Finally, RAP is >100× and >1000× more energy efficient than SotA GPU and CPU solutions.