Implementation of SVM-Based Low Power EEG Signal Classification Chip

Abstract

EEG-based classifiers are being used in a growing number of applications, including assisting cognitive function and medical diagnosis. In this brief, a low power EEG signal detection chip is presented based on supporting vector machine (SVM) for classification, with the application example of epilepsy detection. This brief elaborates the detection algorithm and circuit implementation of the neural detection system. The feature extraction module is based on a 256-point FFT for band energies calculation. An exponential operation circuit with scaling unit is proposed to implement the kernel function in classifier. The exponential function operation circuit combines the coordinate rotation digital calculation method (CORDIC) and approximate circuits to obtain power-delay benefit. Fabricated in 130nm CMOS, the proposed chip achieves wide supply range, low power and high rate of correctness. The seizures detection is with an average accuracy of over 80% and sensitivity of 94.4%. The energy consumption of chip is 1.28uJ/class and the area is only 1.8* <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1.8mm^{2}$ </tex-math></inline-formula> .