An Optimized Hardware Inference of SABiNN: Shift-Accumulate Binarized Neural Network for Sleep Apnea Detection

This article presents the design of an optimized hardware-based neural network (NN) called a shift-accumulate binarized NN (SABiNN). SABiNN is used in detecting respiratory-related diseases such as sleep apnea (SA) among adults. Initially, a three-hidden-layer-based NN model was trained, validated, and tested with open-source apnea polysomnography (PSG) datasets collected from the PhysioNET databank. Single-lead electrocardiography (ECG) and pulse oximeter data were collected, preprocessed, and digitized for network training. The NN was later transformed into SABiNN, demonstrating model accuracy of 81.5% (CI: ±3.5) with an energy efficiency of 5 mJ on reprogrammable hardware. The precision rate of the model was further increased by redesigning the XNOR gate of the multiply–accumulate (MAC) operation with the NAND gate-based XNOR. This redesign process significantly improved the overall model’s classification and precision. Further expansion of SABiNN was carried out to achieve a higher accuracy rate (over 88%) which was designed on the CMOS platform using a 130-nm open-source process design kit (PDK) developed by Google and Skywater. The proposed model on the CMOS platform used a chip area of 0.16 mm ² and showcased an energy efficiency of 1 pJ. A total of ~11k CMOS cells with two 16-bit input and one 1-bit output pins were used to realize the SABiNN on CMOS. The success of this design technique can be leveraged in developing a fully system-on-a-chip (SoC) integrated wearable system for SA detection.