Direct Parameter Extraction for Dispersive Large-Signal GaN-HEMT Models from Nonlinear Measurement

为了模拟单片集成微波电路，需要精确的晶体管模型。这些模型通常由许多测量确定，例如，顺序地确定外部元件、DC行为和电容。虽然许多这些测量是在与常规晶体管操作非常不同的条件下进行的，但假设晶体管性能是一致的，因为关于晶体管物理的基本假设是正确的。组合不同参数提取步骤的结果需要产生晶体管的总体模型。这种假设不能被视为理所当然。在这方面困难的最有趣的晶体管类型是GaN HEMT。陷阱在这种类型的晶体管中仍然占主导地位，并导致色散行为或记忆效应。GaN HEMT通常表现不同，例如，如果执行输出IV曲线、S参数或负载牵引测量。GaN HEMT及其陷阱是一个正在进行的研究，这与如何表征其性能的问题密切相关。因此，本项目试图通过分析算法直接从大信号电压和电流波形确定晶体管模型参数。问题是如何解释和安排这些测量数据，以便在不依赖全局优化的情况下直接提取参数。这种方法的优点是，一方面，确定参数的测量结果类似于真实的操作条件。另一方面，人们仍然有可能深入了解晶体管物理。该项目分为三个部分。第一步，将实施一个简单的测量设置，允许在较低的MHz范围内测量电压和电流波形。与现有的GHz测量系统相比，可以忽略较低MHz范围内电容的影响。如何确定晶体管的电流源的调查将进行基于这些测量。最后，将解决晶体管电容的提取。作为该项目的成果，将知道从大信号电流和电压波形中直接提取晶体管模型参数的理论和实际要求。在测量和晶体管行为方面的要求，使一个成功的提取将给出，和限制将在理论上描述和验证的测量。

Accurate transistor models are required in order to simulate monolithically integrated microwave circuits. These models are commonly determined from numerous measurements, e.g. to determine sequenceally the extrinsic elements, the DC behavior and the capacitances. Although a lot of these measurements are carried out in conditions that are very different from the regular transistor operation, it is assumed that the transistor performance is consistent in the way that the underlying assumptions concerning the transistor physics are correct. Combining the results of the different parameter extraction steps needs to yield a model for the transistor in total. This assumption can't be taken for granted. The most interesting transistor type that is difficult in this respect is the GaN HEMT. Traps are still dominant in this type of transistor and lead to dispersive behavior, or memory effects. The GaN HEMT usually behaves differently, if e.g. output IV curves, S-parameters, or load-pull measurements are performed. The GaN HEMT and its traps is a matter of ongoing research, which is closely linked to the question of how to characterize its performance. This project therefore attempts to determine the transistor model parameters directly from large-signal voltage and current waveforms though analytical algorithms. The question is how this measurement data can be interpreted and arranged in a way that allows a direct parameter extraction without relying on global optimization. The advantage of such an approach would be on one hand, that the measurement from which the parameters are determined resembles the real operation conditions. On the other hand, one would still keep the possibility to get some insight into the transistor physics. The project is divided in three parts.In a first step, a simple measurement setup will be implemented, that allows for the measurement of voltage and current wave forms in the lower MHz range. In contrast to the GHz measurement system that is already available, it will be possible to neglect the impact of capacitances in the lower MHz range. An investigation of how to determine the transistor's current source will be carried out based on these measurements. Finally, the extraction of the transistor capacitances will be addressed. As the outcome of this project, the theoretical and practical requirements for the direct extraction of transistor model parameters from large-signal current and voltage waveforms will be known. The requirements in terms of measurement and transistor behavior enabling a successful extraction will be given, and limitations will be theoretically described and verified by measurement.