No abstract

No abstract

Recently there have been very interesting approaches to mimic the memory operation of the biological system with microelectronics technology in pursuit of ultralow power brain-like computing systems. However, the previous works dealt with the transition between short-and long-term memories and spike-timing-dependent plasticity (STDP) separately, although both are essential features of biological synaptic systems. In this thesis, a novel Si-based device structure, called Si-based Floating-body Synaptic Transistor (SFST), is proposed to implement both of synaptic characteristics with a single silicon device. The structure of the SFST is based on the capacitor-less DRAM for short-term memory and is intended to have long-term memory by locating the floating-gate at the backside. Since SFST is composed of only silicon and silicon oxide, it can be fabricated using exiting silicon-technology and its superb compatibility with the currently dominant CMOS technology is a great advantage. ii The floating-body effect and hot carrier injection are used to mimic short-and long-term memories, respectively. In addition, the transition from short-term memory to long-term memory can be obtained without any change of input bias. The STDP of the SFST is also investigated. Whether the connectivity is strengthened or weakened is determined only by the relative spike timing. The triggering mechanism and STDP characteristics of the SFST are demonstrated using ATLAS TM TCAD device simulator of SILVACO Inc.

The silicon-on-insulator (SOI) lateral doublediffused metal-oxide-semiconductor (LDMOS) with high-k multi-fingers (HKMFs) is proposed and investigated. The fingertips are distributed at specific locations to modulate the electric field distributions and improve the device performances. First, the electric field peaks formed at the fingertips could optimize the electric field distributions, which improves the breakdown voltage (BV) of the LDMOS effectively. Meanwhile, the multi-fingers are embedded into the drift region to increase the optimal drift doping concentration, which facilitates the positive conduction of the device and reduces the specific ON-resistance (R on,sp ). The simulation results show that the proposed HKMF-LDMOS with five multi-fingers increases the BV by 59.2%, reduces R on,sp by 37.8%, and improves the figure of merit (FOM) by 4.07 times when compared to the conventional LDMOS.

No abstract