Abstract

In this study, a novel approach is presented for developing ultra-low power analog classifiers capable of effectively handling multiple input features while maintaining high levels of accuracy and minimizing power consumption. The proposed methodology is built upon a Voting model that leverages Gaussian likelihood functions. To evaluate the performance of the proposed methodology, a comparison is conducted against the widely used analog Bell-shaped classifiers. Real-life breast cancer dataset is employed for this comparison. The models are trained and the results are processed using the Python programming language. The hardware design and result processing utilize Cadence IC Suite, implementing the TSMC 90 nm CMOS process technology.