Abstract
在過去的數十年裡，科學家努力欲將人腦的記憶模式實現在電腦上．有許多類比記憶體被開發出來，並且可實作在矽晶片上，進而模擬生物上的突觸與神經元的功能．本論文提出了一個新的類比記憶體結構，並藉由可擦拭可程式化唯讀記憶體的技術，模擬一個神經元的動態存取動作．在本論文中除了介紹基本的動作機制之外，並提出完整的模擬結果．此外，對於元件結構上的考量與未來的改進方向也加以討論．Over the last several decades, neuromophic scientists have longsought to realize the way human brain works in a computer.Various on-chip analog storage devices were designed to achievethe biological functions of synapses or neurons. In this work wepropose a novel device structure for an electrically trainablesingle-transistor silicon neuron, based on the Flash EEPROMtechnology. This device contains multiple control gates and thecorresponding floating gates to specify the act of an electronicneuron.Signals biased on the control gates are just like theinput excitations of synapses, and the charges on thecorresponding floating gates determine the weighting factors ofthe synapses. We propose the basic structure and operations ofour device, including the analogy to a biological neuron and themethods of updates with simulation results. Conclusions weremade with a simplified equivalent circuit model, and severalstructural suggestions for future development are alsodiscussed.