Runtime BIST for ReRAM-based CAM using Frequency-Domain Monitoring

Measuring similarity between binary patterns is a fundamental operation in many data-intensive applications, which can benefit significantly from Computation-in-Memory (CiM) architectures using Content-Addressable Memories (CAMs). ReRAM-based CAM (ReCAM) enables energy-efficient and scalable analog in-memory similarity computation. However, its reliability is challenged by manufacturing defects as well as process and runtime variations. This paper proposes a lightweight runtime Built-In Self-Test (BIST) framework to ensure reliable ReCAM operation during lifetime operation. The approach employs a voltage control-based ring-oscillator (VCRO)-based sensing mechanism that converts analog outputs into frequency signatures, enabling compact anomaly detection without high-resolution Analog to Digital Converters (ADCs). Simulation results demonstrate that the proposed framework achieves test coverage of up to 98% while introducing negligible hardware overhead, amounting to at most 0.03% and 1.5% of the ReCAM area and power, respectively.