可程式化生醫訊號量測之低能耗數位訊號處理核心

Abstract

有效率的生醫訊號記錄與訊號處理是發展遠距醫療健康照護、非侵入式義肢設備、以 及其他將生醫訊號轉換成電訊號處理系統的必要條件。生醫訊號記錄系統需要處理低 振幅、低頻率的生醫訊號，而目標是超低能耗以延長電子裝置使用時間。在此計畫中， 我們提出一個低能耗的生醫訊號記錄系統以處理腦波圖(EEG, Electro-encephalogram)、 心電圖(ECG, Electro-cardiogram)、與肌電圖(EMG, Electro-myogram)訊號。針對常用的 生醫訊號處理技術，包括雜訊去除、事件偵測、特徵比對、訊號分離或聚集提供一低 能耗、低複雜度硬體可程式化架構與晶片實現。在微架構設計上，針對生醫訊號處理 低工作頻率的特性，將發展位元階層運算邏輯與遞代運算來處理非線性運算並使電路 操作在能量消耗最低的工作點。由於電路需要操作在極低電壓區域，將針對所需運算 單位研發低電壓邏輯與記憶電路。本計畫第一年評估所需的演算法效能與支援的參數 範圍，透過製程參數模擬以決定最佳電路架構與電路實現。第二年將整合演算法與硬 體設計進行晶片實作與量測。

Efficient biosignal recording and processing is a pre-requisite for the development of remote health-monitoring systems, non-invasive prosthetic equipment, and several other applications that require interfacing biological signals to electronic circuitry. Biosignal recording systems need to deal with low-amplitude signals at very low frequencies, and are subject to extremely stringent power constraints. In this work, we will develop a power-efficient biosignal recording system for handling EEG (Electro-encephalogram), ECG (Electro-cardiogram), and EMG (Electro-myogram) signals. A high-performance reconfigurable architecture will be developed to support commonly used biomedical signal processing techniques such as noise reduction, event detection, feature matching, signal separation and clustering. At the micro-architecture level, bit-level and iterative operators will be developed to deal with non-linear operations at the minimal energy point. To operate circuits at an ultra low-voltage regime, we will develop low-voltage logic and memory circuit techniques. In the first year, we will evaluate the algorithmic performance and design parameters. The optimal architecture and circuit design will be decided by the process simulation results. In the second year, the chip will be implemented and tested for system integration.