Skip to content

Advanced Electro-Thermal Modeling of Edge Termination in Diamond and Ultra-Wide Bandgap Power Devices

FieldValue
JournalApollo (University of Cambridge)
Published2026-03-19
AuthorsMartin Kah, Nazareno Donato, Nicolas Rouger, Rebecca J. Watkins, Calum S. Henderson
DOI10.17863/cam.128411

In the paper’s transient field comparison at 10⁶ V/s, the hybrid reduced the maximum electric field by 30%. At that condition, the reduction at the diamond surface was as large as 90%. It had the highest simulated likelihood of withstanding a 1200 V surge. These results are simulation predictions, not measurements from fabricated devices.

Conventional edge-termination design began with a quasi-static reverse-bias ramp of 1 V/s in mixed-mode TCAD. Breakdown was extracted from the abrupt reverse-current increase caused by avalanche, then compared with a one-dimensional parallel-plane limit. The impact-ionization map used coefficients reported by Hiraiwa and colleagues. The target was approximately 1200 V for an 800 V automotive DC architecture with transient overvoltage. The mapped design used NA = 5 × 10¹⁶ cm⁻³ and a 2 µm epitaxial layer. It lies near the transition between punch-through and non-punch-through depletion.

Most experimental diamond devices in the literature achieved less than 60% of the theoretical one-dimensional limit. The authors state that reverse-bias failure in these devices remained dominated by material and processing limits. Lack of efficient n-type homo-epitaxial doping hinders reliable single- or multi-zone JTE and ring terminations in diamond. Field plates therefore remain practical, but they shift stress into oxides with lower critical fields.

Mixed-mode TCAD solved the external circuit and internal device equations together at every time step. The first stage applied 1 V/s to estimate blocking capability and rank candidate terminations. The second stage used isothermal voltage ramps or square waveforms with adjustable slopes, spanning 1 V/s to 10⁶ V/s. This range reproduced key features of dynamic reverse-bias qualification while covering quasi-static and fast-transient behavior. The final stage solved the Poisson, carrier transport, and heat-flow equations self-consistently. Heat left through a 300 µm type-Ib substrate with a 0.3 K/W backside thermal resistance; all other boundaries were adiabatic. Outputs included reverse current, recovery kinetics, electric-field profiles, peak junction temperature, and the impact-ionization integral. Breakdown was assigned when reverse current rose sharply and the impact-ionization integral exceeded unity.

Donor ionization explains the fast-ramp failure

Section titled “Donor ionization explains the fast-ramp failure”

Phosphorus has an ionization energy of 0.57 eV. Previous results cited by the paper put its room-temperature ionization below 1% when ND exceeded 10¹⁵ cm⁻³. The ionization time constant ranged from 1 ms to 10 s, depending on the phosphorus capture cross-section. During 10 to 100 V/s ramps, complete and incomplete ionization produced comparable current. The current was dominated by displacement until the 1100 V bias stabilized. At 10⁴ to 10⁶ V/s, incomplete ionization produced a sharp current rise consistent with premature avalanche. The current then relaxed toward the dark-current level over the donor ionization time constant. Only a small ionized fraction initially supplied space charge, so the electric field crowded at the Schottky contact edge. At the 10⁶ V/s sweep, complete ionization kept the peak field at the pn-junction interface after 1 ms. The incomplete-ionization field reached a localized maximum higher than diamond’s critical field under the most optimistic conditions. Released electrons entered that high-field region and increased impact ionization before the donor population equilibrated. This transient charge imbalance reverses the termination ranking under fast reverse-bias stress.

The hybrid termination changes the controlling mechanism

Section titled “The hybrid termination changes the controlling mechanism”

The compared devices were unterminated, field-plate-only, JTE-only, and hybrid JTE plus Al₂O₃ field plate. Although all devices were designed for 800 V under DC conditions, the hybrid showed the highest likelihood of withstanding the short transient spike. The other three configurations could not sustain that spike in the transient comparison. The JTE used a phosphorus-doped n-layer with ND = 10¹⁸ cm⁻³ and a thickness of 200 nm. The p-type drift layer retained NA = 5 × 10¹⁶ cm⁻³. The authors note that fabricating such a thin, highly doped n-diamond layer remains difficult. The simulated hybrid uses the field plate during the early fast-ramp stage. After sufficient recovery, the JTE contribution resumes control as donor ionization completes. Earlier JTE optimization for slow ramps can therefore produce a worse result under dynamic stress than deliberately detuned parameters. Field-plate geometry still requires optimization because the governing field parameters differ between slow and fast ramps.

Self-heating alters recovery in the JTE-only diode

Section titled “Self-heating alters recovery in the JTE-only diode”

The electro-thermal comparison applied a 10⁶ V/s reverse-bias ramp that saturated at 1100 V. Initial lattice temperatures were 278 K, 298 K, and 398 K. The hybrid showed no significant self-heating because its terminated region handled low power density at low leakage current. The JTE-only diode rose by approximately 20 K from room temperature. Field crowding and avalanche-induced heating concentrated the rise near the Schottky contact edge. Heating accelerated reverse-current recovery because the JTE dopant ionization ratio increased and the ionization time constant shortened. Starting at 398 K removed the self-heating rise because diamond’s critical electric field increased with temperature. Moderate heating shortened the current-increase interval during turn-off. Yet the JTE-only design suffered a significant increase in turn-off loss, which could limit power handling and switching frequency. The short avalanche duration made the modeled response insensitive to changes in the thermal impedance network.

The transient analysis included avalanche breakdown but did not include premature oxide breakdown. The termination ranking therefore depends on an oxide thick enough to withstand the elevated surface fields at avalanche. The hybrid lowers oxide stress during the interval when an FP-only design would retain a high field. Lower average and peak oxide fields should reduce leakage through trap-assisted tunnelling, WKB tunnelling, and Fowler-Nordheim emission. The authors expect lower field magnitude and shorter stress time to extend oxide and device lifetimes. That lifetime benefit is an expectation from the cited degradation models, not a measured result. Normalized field plots used a reference maximum of about 25 MV/cm across the structures. The paper warns that this value depends on modeling assumptions and is not a strict physical limit. Improved transient blocking may also require a thicker oxide than a JTE-only design. The hybrid provides its transient advantage at a fixed device footprint relative to FP-only architectures. The paper reports no fabricated-device validation for the hybrid and does not model oxide failure.

Termination selection for fast automotive switching cannot be based on a static breakdown voltage alone. The calculation should include ramp rate, incomplete donor ionization, oxide field exposure, and self-heating. For diamond JTEs, the phosphorus ionization time constant sets the timescale over which the field distribution changes. The field plate governs the earliest transient, while the JTE governs the recovered quasi-static state. Real SiC and GaN converters can reach 10⁹ to 10¹⁰ V/s during switching, above the paper’s explicit simulation range. The authors state that diamond transient limitations are fully developed above 10⁶ V/s and further increases do not qualitatively change breakdown. That statement is a modeled extrapolation rather than a measurement at converter slew rates. Compared with prior work, the main result is a reversal of the termination ranking when dynamic behavior is included. Prior experimental diamond devices generally remained below 60% of the theoretical one-dimensional breakdown limit. The workflow targets high-voltage, fast-switching diamond devices for electric-vehicle inverters and onboard DC-DC converters. It also provides a method for evaluating other ultra-wide-bandgap materials when their dopant and dielectric kinetics are known.

The modeled device requires a 300 µm type-Ib diamond substrate, a p-type drift layer, a phosphorus-doped JTE, and a Schottky contact. The diamond materials catalog lists MPCVD-grown single-crystal diamond from 0.1 to 500 µm, which includes the modeled 300 µm substrate thickness. 6CCVD’s engineering support can assess the catalog’s Au, Pt, Pd, Ti, W, or Cu contact choices, type-Ib suitability, and phosphorus-doped JTE process gaps.

View Original Abstract

Traditional design and optimization of Si-based power devices rely on quasi-static edge termination modeling that has proven to be effective in reproducing experimental results. However, this approach fails to accurately describe emerging wide- and ultra-wide bandgap (UWBG) power devices operating in fast-switching environments, such as in automotive applications, where devices are subject to rapid voltage and over-current transients ( $\textit {dV}/\textit {dt} \gt 100$ V/ns). The absence of a robust design methodology that captures the transient electro-thermal response of the semiconductor, oxide, and passivation layers often results in devices that underperform in real operating conditions. This work addresses this longstanding challenge by introducing a comprehensive dynamic simulation framework that reproduces key aspects of dynamic reverse-bias (DRB) testing used for automotive qualification. Using a diamond-based Schottky barrier diode (SBD) as a case study, we demonstrate that conventional quasi-static and time-dependent simulation strategies lead to markedly different design outcomes. The proposed transient approach captures the interplay between field redistribution, charge trapping, and self-heating effects, offering enhanced predictive accuracy and robustness for the next generation of power devices.