Buffer optimization for wide-bandgap RF and Power devices

Abstract

The core of high performance and reliable operation of wide bandgap devices lie in the design of buffer layers. This thesis is primarily focused on the impact of the buffer on
the operation of AlGaN/GaN HEMTs, devoted to both RF GaN on Silicon Carbide (SiC) and power switching GaN on Silicon devices. The demand for higher power is increasing which requires improvement at the material and device level. Nominally undoped GaN is usually slightly n-type due to Nitrogen vacancies and impurities such as Oxygen acting as donors. A dopant such as Iron (Fe) and Carbon is used to make the buffer more insulating. Fe, an acceptor with the energy of 0.72 eV below the conduction band, sits on the Nitrogen site and has been quite successful in achieving an insulating buffer resulting in good RF performance. Carbon, on the other hand, is a better choice for the power industry which requires high breakdown voltages and has been able to deliver resistivity as high as 1013 Ω.cm. However, Carbon is a deep acceptor when on the Nitrogen site (CN ) with an acceptor trap level 0.9 eV above the valence band making the buffer p-type. A p-type buffer underneath the 2DEG forms a vertical p-n junction which can
make the buffer float and act as a reservoir for time-dependent charge storage. A background level of Carbon will inherently be incorporated during GaN growth due to the presence of organic carrier gases in MOCVD. It is the role of this background Carbon in Fe doped buffer layers and its subsequent consequences, which forms the crux of this thesis.
Wafers with all parameters identical but for the background Carbon level has been subjected to DC, Pulse I-V, transient and breakdown studies yielding valuable information about conduction mechanism in the buffer. Kink effect, which is a hysteresis in the output characteristics of a transistor, is shown to be strongly dependent on background Carbon density. An explanation based on a "leaky dielectric" model of a floating semi-insulating GaN buffer together with conventional Fe and C deep-level defects has been tested and applied successfully. The proposed model is a more realistic approach in comparison to two other models reported in past which are based on unusual deep level defect properties and cannot be explained by conventional defect models. The proposed insight into the mechanism for kink offers a route to its control and suppression. Positive and negative magnitude drain current transient signals with 0.9 eV activation energy have been seen, corresponding to changes in the occupation of Carbon acceptors located in different regions of the GaN buffer. The observation of such signals from a single trap type also raise questions on conventional interpretations of these transients based on the bulk 1-D deep-level transient spectroscopy (DLTS) models for GaN devices with floating regions.
These wafers showed very different Pulse I-V behaviour which can be explained based on their buffer doping; however, under RF IV measurement they appeared identical. This absence of correlation between Pulsed and RF IV brings into question the applicability of Pulsed I-V measurements alone as a tool for extracting nonlinear device models in the case of GaN HEMTs, which is a widespread practice. Drain injected breakdown study on these wafers resulted in the identification of the two-step process, one initiated by a leakage path between the source and drain and other due to electron punch-through current. Simulations and Electroluminescence studies under operation confirmed impact ionization, with its impact being much smaller compared to punch through.
Effect of Carbon doping on buffer behaviour for GaN on Silicon devices aimed for power applications has also been evaluated, using stepped and ramped substrate bias. For the first time, experimental evidence of lateral charge spreading beyond the active area of the device and its impact on neighbouring devices has been explained. The lateral charge spreading beyond the active area can be a concern for wafer-level reliability or system integration.

The last two Chapters introduces two possible next-generation RF transistors. GaN on diamond transistors with ultra-thin GaN and diamond as a heat sink has been shown with excellent thermal and electrical performance. As an alternative, a new wideband gap material Gallium Oxide (β-Ga2O3), which could be a choice for future RF device has been evaluated using Pulse I-V and large-signal RF. These devices show minimal surface and buffer trapping, yielding a record RF performance. However, poor thermal conductivity is the performance limiting factor, which if countered well can lead to a promising material for future devices.

Date of Award	1 Oct 2019
Original language	English
Awarding Institution	University of Bristol
Supervisor	Martin H H Kuball (Supervisor)