A Low-Power Processor With Con figurable Embedded Machine-Learning Accelerators for High-Order and Adaptive Analysis of Medical-Sensor Signals Kyong Ho Lee , Student Member , IEEE, and Naveen V erma , Member , IEEE

Abstract

Low-power sensing technologies have emerged for acquiring physiologically indicative patient signals. However, to enable devices with high clinical value, a critical requirement is the ability to analyze the signals to extract speci fic medical informa- tion. Y et given the complexities of the underlying processes, signal analysis poses numerous challeng es. Data-driven methods based on machine learning offer distinct solutions, but unfortunately the computations are not well supported by traditional DSP. This paper presents a custom processor that integrates a CPU with configurable accelerators for discriminative machine-learning functions. A support-vector-mac hine accelerator realizes various classification algorithms as well as various kernel functions and kernel formulations, enabling r ange of points within an accu- racy-versus-energy and -memory trade space. An accelerator for embedded active learning enables prospective adaptation of the signal models by utilizing s ensed data for patient-speci fic customization, while minimizing the effort from human experts. The prototype is implemented in 130-nm CMOS and operates from 1.2 V–0.55 V (0.7 V for SRAMs). Medical applications for EEG-based seizure detection and EC G-based cardiac-arrhythmia detection are demonstrated using clinical data, while consuming 273 J and 124 J per detection, respectively; this represents 62.4 and 144.7 energy reduction compared to an implemen- tation based on the CPU. A patient-adaptive cardiac-arrh