Ultra-low Power ECG Processing ASIC Techniques for Always-on Cardiovascular Disease Monitoring

The ECG based always-on cardiovascular disease (CVD) monitoring device, either in the form of implantable device or wearable equipment, has received substantial attention recently. In this talk, the system considerations, resource efficient signal processing algorithm, and the hardware implementation techniques for microwatt/sub-microwatt ECG signal processing ASIC will be presented. Two design examples, based on classical time-domain analysis and artificial neural network (ANN), respectively, will be discussed. In the first design, an adaptive derivative-based time-domain detection algorithm with low computation overhead is proposed for the time-domain R-peak and arrhythmia detection. In order to save as much as possible cardiac information with the limited memory size, a hierarchical data buffer structure is proposed. In the second design, a hardware efficient abnormal CR detection algorithm using hybrid tiny classifiers is proposed. An efficient QRS complex morphology classification algorithm using ternary neural network is proposed, in which the QRS complex is converted into a binary image and then classified. The decision logic to classify 13 different types of cardiac rhythms with patient adaptive approach has been designed. The design principles of energy efficient biomedical signal processing ASICs will be illustrated through these two design examples.