Publikation
In-Memory SAT Evaluation on RRAM Crossbars: From Exact Verification to Sampling-Based Falsification
Fatemeh Shirinzadeh; Abhoy Kole; Kamalika Datta; Saeideh Shirinzadeh; Rolf Drechsler
In: Cross-disciplinary Conference on Memory-Centric Computing (CCMCC 2026). IEEE Cross-disciplinary Conference on Memory-Centric Computing (CCMCC), October 28-30, Lyon, France, 2026.
Zusammenfassung
In-memory computing with RRAM crossbar architectures provides a massively parallel substrate for accelerating
Boolean reasoning. In this work, we introduce an in-memory
SAT-based verification framework that maps Boolean formulas
directly onto crossbar arrays. While deterministic CNF-based
evaluation enables exact SAT/UNSAT decisions, exhaustive exploration of all input assignments leads to prohibitive latency, even
when executed in parallel in-memory. To address this limitation,
we propose a sampling-based in-memory falsification method
that explores the input space through randomized evaluation
directly on the crossbar. The proposed approach focuses on
detecting functional discrepancies between designs by identifying
counterexamples rather than proving equivalence exhaustively.
As soon as a mismatch is observed, non-equivalence is established,
enabling rapid bug detection with significantly reduced computational effort. Experimental results on standard benchmark
suites demonstrate that the proposed method achieves substantial reductions in execution latency and crossbar utilization
compared to exhaustive evaluation, while effectively detecting
non-equivalent designs. These results highlight the potential
of sampling-based in-memory falsification as a scalable preverification strategy for RRAM-based computing systems
