An investigation into the reliability of the silicon dioxide/silicon carbide material system.

摘要

The goal of this thesis is to determine the reliability of thermally grown oxide films on SiC. The necessity of performing reliability measurements is to be able to design and fabricate power MOSFETS. Reliability testing occurs under accelerated temperature and field conditions. This allows extrapolation of the data to at-use conditions. This is the first time that a set of comprehensive reliability measurements is taken on the SiC semiconductor. Using the constant voltage stress test technique, time dependent dielectric breakdown (TDDB) measurements are made on both n-type and p-type 6H SiC capacitors. Preliminary measurements are also made on 4H SiC capacitors. The purpose of taking TDDB scans on both n-type and p-type substrates is to understand how the polarity of charge injection affects the reliability of the oxide. TDDB measurements are made at three temperatures(145 C, 240 C, and 305 C) and three or four field values. All measurements are made in accumulation mode so that the applied voltage drops across the oxide only. The results show failure for thermally grown oxides on SiC is multi-modal, consisting of a two part extrinsic failure mode and an intrinsic failure mode. Extrapolation of the intrinsic n-type data taken at 145 C shows that the t{dollar}sb{lcub}50%{rcub}{dollar} at 3 MV/cm is approximately 2,000,000 years. The extrapolation of the extrinsic data shows that t{dollar}sb{lcub}50%{rcub}{dollar} is 10 years. At higher temperatures t{dollar}sb{lcub}50%{rcub}{dollar} decreases for both extrinsic and intrinsic failures. As expected, the activation energy for the extrinsic failures is less than that of the intrinsic failures. This confirms that the mechanism of failure between the two modes differs. In the p-type configuration, charge is injected from the gate into the oxide. This proves to be less damaging than injecting from the semiconductor. The p-type data also varies from the n-type data in the fact that the p-type devices fail slower. As a result, the extrapolation to at-use conditions for p-type data is longer than the n-type extrapolations.