A Low-Power Processor With Configurable Embedded Machine-Learning Accelerators for High-Order and Adaptive Analysis of Medical-Sensor Signals

• Published in: IEEE journal of solid-state circuits Vol. 48; no. 7; pp. 1625 - 1637
• Main Authors: Kyong Ho Lee, Verma, N.
• Format: Journal Article Conference Proceeding
• Language: English
• Published: New York, NY IEEE 01.07.2013; Institute of Electrical and Electronics Engineers
• Subjects: Active learning (subject-specific adaptation); Adaptation models; Applied sciences; biomedical electronics; Brain models; Computational modeling; Data models; Design. Technologies. Operation analysis. Testing; Electronics; Exact sciences and technology; General equipment and techniques; Instruments, apparatus, components and techniques common to several branches of physics and astronomy; Integrated circuits; Integrated circuits by function (including memories and processors); Kernel; machine learning (artificial intelligence); medical signal processing; Physics; Semiconductor electronics. Microelectronics. Optoelectronics. Solid state devices; Sensors (chemical, optical, electrical, movement, gas, etc.); remote sensing; support vector machine (SVM); Support vector machines; Static random access memory; Prototype; Random access memory; support vector machine (SVM); Information extraction; Support vector machine; Implementation; Learning; Prospective; Complementary MOS technology; machine learning (artificial intelligence); Medical application; Discriminant analysis; Measurement sensor; Active learning (subject-specific adaptation); Algorithm; Kernel method; medical signal processing; Learning (artificial intelligence); Integrated circuit; Signal processing; Low-power electronics; Digital signal processor; Signal analysis; Artificial intelligence; Biomedical electronics
• ISSN: 0018-9200; 1558-173X
• DOI: 10.1109/JSSC.2013.2253226

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 specific medical information. Yet given the complexities of the underlying processes, signal analysis poses numerous challenges. 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-machine accelerator realizes various classification algorithms as well as various kernel functions and kernel formulations, enabling range of points within an accuracy-versus-energy and -memory trade space. An accelerator for embedded active learning enables prospective adaptation of the signal models by utilizing sensed data for patient-specific 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 ECG-based cardiac-arrhythmia detection are demonstrated using clinical data, while consuming 273 μJ and 124 μJ per detection, respectively; this represents 62.4× and 144.7× energy reduction compared to an implementation based on the CPU. A patient-adaptive cardiac-arrhythmia detector is also demonstrated, reducing the analysis-effort required for model customization by 20 ×.