TY - JOUR
T1 - A low-power processor with configurable embedded machine-learning accelerators for high-order and adaptive analysis of medical-sensor signals
AU - Lee, Kyong Ho
AU - Verma, Naveen
PY - 2013/4/5
Y1 - 2013/4/5
N2 - 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 ×.
AB - 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 ×.
UR - http://www.scopus.com/inward/record.url?scp=84879931688&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84879931688&partnerID=8YFLogxK
U2 - 10.1109/JSSC.2013.2253226
DO - 10.1109/JSSC.2013.2253226
M3 - Article
AN - SCOPUS:84879931688
SN - 0018-9200
VL - 48
SP - 1625
EP - 1637
JO - IEEE Journal of Solid-State Circuits
JF - IEEE Journal of Solid-State Circuits
IS - 7
M1 - 6493458
ER -