Stochastic first-passage modeling of single-event burnout in SiC power MOSFETs

This paper introduces a stochastic first-passage modeling framework that explains how electrothermal fluctuations in SiC power MOSFETs transform the deterministic single-event burnout threshold into a probabilistic transition band, revealing noise-induced subthreshold failures and providing a statistical-physics interpretation of threshold dispersion.

The Gist

The Big Picture: A Game of "Too Much Heat"

Imagine a silicon carbide (SiC) power MOSFET (a type of high-tech electronic switch) as a tiny, high-pressure kitchen. Inside this kitchen, there is a stove (the electric field) and a pot of water (the electrical current).

Usually, this kitchen runs perfectly. But sometimes, a "heavy ion" (like a tiny, fast-moving speck of cosmic dust) flies through the kitchen and knocks over a stack of dishes. This creates a mess of extra energy and heat.

The paper asks a simple question: Will the kitchen recover and clean up the mess, or will it explode (burn out)?

The Old Way: A Sharp Line

Previously, scientists thought of this like a light switch.

• If the mess is small, the kitchen cleans up (Recovery).
• If the mess is big, the kitchen explodes (Burnout).
• There was a single, sharp line in the middle: "If the mess is bigger than X, it explodes."

The New Way: A Foggy Zone

This paper argues that reality isn't a sharp switch; it's more like a foggy zone or a slippery slope.

Because the universe is full of tiny, random fluctuations (like a chef sneezing, a door creaking, or a random gust of wind), the outcome isn't always the same, even if the mess looks identical.

• The Analogy: Imagine trying to balance a broom on your hand. If you push it slightly too far, it falls. But if you are standing on a slightly shaky boat (representing random noise), the broom might fall even if you didn't push it quite as hard. Or, it might stay up even if you pushed it a little too far, just because the boat rocked in your favor.

The paper introduces a "First-Passage" model. Think of this as a cliff edge.

• The "kitchen" is a hiker walking on a path.
• The "burnout" is falling off the cliff.
• In the old view, there was a specific spot where the ground just ended.
• In this new view, the ground is a bit wobbly. Sometimes the hiker takes a lucky step and stays safe. Sometimes they take an unlucky step and fall, even if they were standing in a place that should have been safe.

How the Model Works

The researchers built a simplified mathematical "toy model" to simulate this. They didn't try to map every single atom in the chip (which is too complex). Instead, they looked at two main things:

1. The Crowd (Carriers): How many extra electrons are running around causing trouble?
2. The Fever (Temperature): How hot is the kitchen getting?

They added random noise to the model to represent the unpredictable nature of real life.

• The Result: They found that the "Burnout Line" isn't a line at all. It's a band of probability.
  • In the middle of this band, a chip might survive 50% of the time and burn out 50% of the time, even if the conditions look exactly the same.
  • Sub-threshold Runaway: This is the most surprising finding. Even if the "mess" is small enough that the chip should be safe (according to old rules), the random noise can sometimes push it over the edge. It's like a quiet room suddenly getting loud enough to break a glass just because of a random vibration.

The "Phase Diagram" (The Map of Safety)

The paper creates a map (a phase diagram) that helps engineers understand the situation.

• The X-axis: How strong is the "feedback"? (Is the heat making more electricity, which makes more heat? A runaway loop.)
• The Y-axis: How good is the "cooling"? (Can the kitchen vent the heat fast enough?)

This map divides the world into three zones:

1. Safe Zone: The cooling wins. The kitchen cleans up the mess.
2. Danger Zone: The feedback wins. The kitchen explodes immediately.
3. The "Maybe" Zone (Probabilistic): This is the new discovery. Here, the cooling and the feedback are fighting a tie. Whether the kitchen explodes depends entirely on a roll of the dice (random noise).

Why This Matters

The paper doesn't claim to fix the chips or predict exactly when a specific chip will fail. Instead, it offers a new way of thinking:

• Old Thinking: "If the voltage is below 500V, it's safe."
• New Thinking: "If the voltage is near 500V, there is a chance it will fail, and that chance gets bigger as you get closer to the limit. We need to talk about probabilities, not just hard limits."

Summary

This paper uses math to show that randomness matters. In the high-stakes world of power electronics, you can't just look for a single "safe" number. You have to accept that near the limit, the outcome is a gamble. The "burnout" isn't a sudden switch; it's a slippery slope where luck (or bad luck) plays a huge role in whether the device survives or burns out.

Technical Summary

Technical Summary: Stochastic First-Passage Modeling of Single-Event Burnout in SiC Power MOSFETs

Problem Statement
Single-event burnout (SEB) in silicon carbide (SiC) power metal-oxide-semiconductor field-effect transistors (MOSFETs) is traditionally characterized by deterministic threshold quantities, such as a critical linear energy transfer (LET) or bias voltage. However, near the boundary between device recovery and electrothermal runaway, microscopic fluctuations, event-to-event variability, and incomplete knowledge of local initial conditions render a sharp threshold description insufficient. While deterministic simulations (e.g., TCAD) and empirical probability curves exist, a direct theoretical link between the stochastic evolution of individual electrothermal failure paths and probabilistic failure metrics is lacking. This work addresses the need for a compact stochastic dynamical formulation to connect physical burnout mechanisms with boundary-crossing observables.

Methodology
The authors formulate heavy-ion-induced SEB as a stochastic first-passage problem within a reduced nonlinear electrothermal system. The approach involves three main stages:

1. Deterministic Reduced-Order Model: The spatially resolved device response is projected onto a small set of collective variables representing a "sensitive region" ($\Omega_s$). The state is defined by the effective excess carrier population, $N(t)$, and the coarse-grained temperature, $T(t)$. The dynamics are governed by coupled equations for carrier injection, avalanche multiplication, carrier loss, and thermal relaxation. An absorbing boundary is defined at a critical temperature $T_c$, representing irreversible burnout.
2. Stochastic Extension: The deterministic equations are extended to Itô stochastic differential equations (SDEs). Unresolved event-level variability is introduced via multiplicative noise terms for the carrier population ($\sigma_n$) and additive noise for the thermal variable ($\sigma_\Theta$). These terms represent fluctuations in energy deposition, ion-track geometry, and local device conditions.
3. Numerical Implementation: The system is solved using an Euler-Maruyama scheme with ensemble Monte Carlo simulations. Key observables include the burnout probability ($P_{SEB}$), the transition width ($\Delta \ell$), the first-passage time distribution ($\tau_{FPT}$), and the survival probability ($S(s)$). A feedback-relaxation phase diagram is constructed to map recoverable, probabilistic, and rapidly unstable regimes.

Key Contributions

• First-Passage Formulation: The paper reinterprets SEB not as a single deterministic switching event but as the arrival of a stochastic trajectory at an absorbing thermal boundary. This provides a statistical-physics framework for threshold dispersion.
• Stochastic Threshold Broadening: The model demonstrates that finite fluctuations broaden the sharp deterministic burnout threshold ($\Lambda_{th} = 1$) into a probabilistic transition band. The width of this band ($\Delta \ell$) quantifies the coexistence of recoverable and runaway trajectories under nominally identical control parameters.
• Noise-Induced Subthreshold Runaway: The study identifies a regime where nominally recoverable conditions ($\Lambda_{th} < 1$) can still lead to failure. Rare stochastic excursions driven by noise can trigger first-passage events even when the deterministic trajectory would recover.
• Temporal Resolution of Failure: By analyzing first-passage time statistics, the model distinguishes between rapid, feedback-dominated runaway and delayed, noise-assisted failure, offering a more nuanced view of failure timing than binary pass/fail metrics.
• Phase Diagram Organization: A feedback-relaxation phase diagram is proposed, organizing device behavior into recoverable, probabilistic, and rapidly unstable regimes based on avalanche feedback strength and thermal relaxation rates.

Results

• Deterministic Reference: In the zero-noise limit, the model recovers a sharp boundary where the thermal indicator $\Lambda_{th}$ reaches unity. The deposited work factor ($\Lambda_{dep}$) alone is shown to be insufficient for predicting burnout; the temporal distribution of heating and thermal relaxation are critical.
• Transition Broadening: As noise strength ($\sigma$) increases, the transition from recovery to burnout becomes a smooth S-shaped curve in the burnout probability ($P_{SEB}$) versus ionization strength ($\ell$) plot. The transition width $\Delta \ell$ increases linearly with noise strength.
• Subthreshold Failure: For parameters where the deterministic model predicts recovery, the stochastic model yields a non-zero $P_{SEB}$. This probability increases with noise amplitude, confirming that rare fluctuations can drive the system to the absorbing boundary.
• First-Passage Statistics: The distribution of first-passage times shifts from broad and delayed (near the lower transition edge) to concentrated and rapid (at higher ionization strengths). The conditional mean first-passage time decreases as ionization strength increases.
• Phase Diagram: The $(F, r)$ plane (feedback strength vs. thermal relaxation) clearly separates regimes. The deterministic boundary ($\Lambda_{th}=1$) aligns closely with the $P_{SEB}=0.5$ contour, while the $P_{SEB}=0.1$ and $P_{SEB}=0.9$ contours define the stochastic transition band.

Significance and Claims
The paper claims to provide a statistical-physics interpretation of threshold dispersion in SiC power MOSFET SEB. It argues that the framework complements existing deterministic TCAD simulations and irradiation testing by linking coarse-grained electrothermal dynamics to probabilistic and time-resolved failure observables.

The authors emphasize that their model is intentionally reduced and does not replace spatially resolved device simulations. Instead, it serves as a statistical layer that explains the observed burnout threshold as an emergent stochastic outcome rather than a fixed deterministic boundary. The work suggests that device-specific predictions require calibration of effective parameters against TCAD or experimental data, but the framework offers a unified language for understanding how microscopic variability translates to macroscopic reliability metrics. The study concludes that SEB in SiC power MOSFETs should be viewed as an electrothermal escape process driven by ionization and field-assisted feedback, where the "margin" against failure is inherently statistical.