基于人工智能近零功耗无线可穿戴芯片关键技术研究

Health Internet-of-Things (IoT) is envisaged as the next wave of technology for healthcare, especially for ageing society. The health IoT is an enabling technology for preventive healthcare, personalized healthcare, and participatory healthcare. The success of health IoT relies on the development of ultra low power wireless wearable biomedical sensing technologies that are capable of adapting to each individual characteristic without frequent calibrations. We propose a machine learning based sensor architecture in the continuous-time discrete-amplitude (CTDA) domain, which utilizes event-driven analog frontend to reduce the sampling rate and enable analog implementation of machine learning algorithms. It is clockless and differs from traditional Nyquist-based architecture. Implementing artificial intelligence based machine learning algorithms is a new technique and has never been done before. In this project, we will be exploring the theory and circuits for this new architecture. Through our work, we will solidify the foundation of analog based machine learning approach with associated circuit building blocks, including analog front end, analog information converter, and analog machine learning engine. Our goal is to develop core technologies for AI embedded wireless wearable sensors.

在现代社会，尤其是在人口老龄化的情况下，健康物联网被视为满足人们健康医护需求，实现个性治疗、预防治疗及参与治疗的新一代技术。以超低功耗（近零功耗）实现体征信息的采集则是实现健康物联网的关键，为此我们在本项目中提出了在连续时间离散幅度域(CTDA)下实现人工智能机器学习的无线可穿戴传感芯片系统架构。CTDA系统完全有别于传统的奈奎斯特采样系统。它无需时钟，能够大大降低稀疏信号的样本数，有望实现极低功耗。在CTDA架构下实现人工神经网络，是一个全新的课题。据我们所知，还没有学者进行过相关研究。我们将对这一新型架构在理论和电路两个层面进行深入研究，提出适合事件驱动的模拟人工智能算法，研究并开创一系列关键电路，包括电极接口模拟前端、模拟信息转换器、模拟机器学习电路等。通过本项目的研究，我们将充分地展现CTDA架构的特点和优越性，丰富其理论基础，并有力推动其实用化。

现代社会，尤其是在人口老龄化的情况下，健康物联网被视为满足人们健康医护需求，实现个性治疗、预防治疗及参与治疗的新一代技术。超低功耗无线可穿戴智能生理特征信息采集芯片是实现健康物联网的关键。本项目从无线可穿戴传感芯片架构入手，提出以连续时间离散幅度域(CTDA)信号流为基础，借助幅值穿越模数转换器（Level-Crossing ADC, LC-ADC）将模拟信号转换成连续时间离散幅度信号，以事件驱动取代传统的固定采样频率系统，压缩生理信号的采样点数，进而减少无线传输的数据量，实现超低功耗传感芯片。基于这一思路，本项目提出了基于LC-ADC的事件驱动无线智能人体生理传感芯片系统架构，解决了LC-ADC设计中的功耗瓶颈，开创性地提出了高能效事件驱动心电信号处理方法，将无线传感芯片的能效提高了1～2个数量级，同时实现了近零功耗嵌入式心电分类算法。本项目的主要成果包括：基于LC-ADC的事件驱动无线传感芯片系统架构，单偏置电压LC-ADC的实现，世界首款1.3微瓦符合国际标准（AAMI）的事件驱动五类心电分类芯片，以及多种心率检测和心电分类的算法，并采用本项目开发的芯片演示了一款使用时间长达七天的超低功耗柔性心电贴，为本项目各项关键技术的商用奠定了基础。