New Gas Identification Method Using Gas Sensor-Amplifier Merged Array and In-Memory Computing-Based Preprocessing

We present a novel gas identification method using a sensor-amplifier merged array and in-memory computing-based preprocessing. An amplifier array that effectively integrates pFET-type and resistive sensors senses gases and selectively amplifies the sensing signals. The amplified array signals are preprocessed in the AND-type nonvolatile memory (NVM) array and converted into refined data having a unique format for each gas type. Owing to well-refined data for gas identification, even a small neural network (3-128-64-5) can identify 25 gas atmospheres (5 gas types × 5 concentrations) with high accuracy (mean squared error = 0.3).