Analytical Models Of LDMOS Based On Numerical Simulation

Lateral high-voltage power device LDMOS (Lateral double-diffused MOSFET) has advantages of high breakdown voltage, large gain, wide dynamic range, low distortion and compatibility with low-voltage circuit process. It is more and more widely used in power integrated circuits and microwave integrated circuits. With the development of semiconductor technology, especially the development of SOI(Silicon on Insulator) technology, SOI LDMOS has been used in high-speed power integrated circuits and RF integrated circuits. Therefore, it is very important to study the characteristics of bulk silicon LDMOS and SOI LDMOS because of the practical applications.Due to the complicated structure of LDMOS and the nonuniformly channel doping concentration, analytical models are very difficult to establish. The models previous divided the on-state region of LDMOS . into two parts, linear region and saturation region. They work out each region' s I-V equation respectively, and combine the two equations. If the structure of LDMOS is very complicated , the equations perhaps cannot be solved. Using these models to simulate circuits can produce a large quantity of calculation that is uneasy to converge. In this paper, the electrical characteristics of LDMOS are simulated by a 2-D device simulator MEDICI . Fitting these curves of resistance changing with drain voltage by MATLAB, we get an I-V equation which is available in the whole on-state region, and derive the macromodel of LDMOS circuits. Whatever the structure of LDMOS is, we can create the macromodel by numerical simulation quickly. Because the model is based on numerical simulation, it has high accuracy and contains fewer parameters easily extracted. According to the equivalent circuit above,we design a high-voltage CMOS inverter. The relationship between the parameters in the macroraodel and the circuit parameters of CMOS inverter (such as the trailing edge time of the NLDMOS) can be derived. Then IC circuit designer can adjust the parameters of LDMOS to get the required circuit parameters. The power dissipation of high-voltage LDMOS is difficult to measure. Using the macromodel we can calculate mean square power dissipation , then the power dissipation of high-voltage LDMOS has been derived. Throughout calculating CMOS inverter power dissipation and power gain, we find power gain is inverse ratio to frequency. Therefore, high-voltage LDMOS is only working under low frequency. When the high-voltage LDMOS is applied to real circuits , the most serious problem is high power dissipation. We can divide an LDMOS into several units and parallel connect them. As the result, the power dissipation could decrease. This paper also analyses the breakdown voltage influenced by field plate. According to the distribution of LDMOS field, we calculate the breakdown voltage when the field plate is infinite and finite. The relationships between breakdown voltage and field plate length can be gotten. From this analysis, a suitable solution has been obtained and the theoretical prediction is well verified by the experimental data.As to the analysis of SOI LDMOS, several typical structures that can hold high drain voltage were introduced. The application of RESURF principle is discussed in detail. We use the quasi-two-dimensional method to get the potential distribution for depleted channel region of fully-depleted SOI LDMOS, and an analytic threshold voltage model has been established. We calculate the threshold voltage when back gate bias is zero and minus, and find that the threshold voltage decreases when the channel length reduces. The doping concentration of drift region can directly influence the breakdown voltage in SOILDMOS. Based on the numerical simulation results , we define the critical doping concentration^ in the drift region at first. Whenthe real concentration is lower than Nc, the breakdown occurs at the drain. When the real concentration is higher than Nc, the breakdownoccurs at the end of channel. We give the model of breakdown voltage, and the results are close to numerical simulation values. So the models we derived are correct.