Skip to content

Semi-empirical model for Impurity-to-Band energy reduction in semiconducting boron-doped diamond

FieldValue
JournalDiamond and Related Materials
Published2026-04-15
AuthorsG. Alba, D. Araujo
DOI10.1016/j.diamond.2026.113645

The semi-empirical model reduced the sum of squared residuals by 29% relative to the adapted Pearson and Bardeen curve with K’ = 0.5 for compiled boron-doped diamond data. Its gain came from treating compensation as impurity band occupancy and electrostatic disorder. A fit that also included measurements with unspecified compensation reduced the residual sum by 20%.

Diamond has a band gap of about 5.45 eV, but its boron acceptor level is deep. Only about 1% of boron acceptors are ionized at room temperature, so incomplete ionization remains important well above room temperature.

The Pearson and Bardeen model relates activation energy to average dopant separation under conditions in which most dopants are ionized. That assumption is weak for diamond, where compensation often controls charge equilibrium.

The authors therefore start from the low-temperature ground state, where strong freeze-out makes compensation and disorder central to charge equilibrium. The model combines wavefunction overlap, which broadens the acceptor impurity band, with compensation-dependent occupancy and electrostatic fluctuations. Both effects depend on static dopant geometry and compensation, so the final expression has no explicit temperature term.

At high boron concentration, neighboring acceptor wavefunctions overlap and discrete boron levels broaden into an impurity band. The authors describe this broadening through the average transfer integral, (\langle J\rangle), whose wavefunction-overlap term falls rapidly with interatomic distance.

The nearest-neighbor distance is set by (R_0 \approx (4\pi/3)N_a^{-1/3}), so increasing (N_a) broadens the band and lowers the activation energy. The model writes (\langle J\rangle) as an exponential function with shape parameter (\gamma) and critical concentration (N_c). The position where (E_a) reaches zero is therefore tied to (N_c).

For an uncompensated, freeze-out sample, the model treats (N_c) as an upper concentration boundary for this zero-activation-energy event. The calibration fixes (N_c) at (3 \times 10^{20}\ \mathrm{cm}^{-3}) and (E_{a0}) at 0.37 eV, following cited literature.

Compensation, occupancy, and electrostatic disorder

Section titled “Compensation, occupancy, and electrostatic disorder”

The compensation ratio (K’ = N_d/N_a) describes the donor-to-acceptor concentration ratio. At 0 K, donor electrons fill the lowest impurity band states. This occupancy shifts the next acceptor states toward higher activation energies, so compensation opposes the reduction caused by higher boron concentration.

For (K’ \gtrsim 1%), the ionized acceptor concentration is dominated by compensation and remains nearly constant. The authors then use (E_a \approx E_F) and connect (E_a) to the inverse cumulative distribution of the impurity band density of states.

The density of states is first represented by a Gaussian centered on (E_{a0}), with standard deviation (\delta E = \langle J\rangle/n). The parameter (n) sets how many standard deviations fit within the overlap energy. At (K’ = 0.5), the model gives (E_a = E_{a0}) regardless of (\langle J\rangle).

Electrostatic disorder from randomly arranged ionized donors and acceptors is combined into one phenomenological reduction term. It depends on compensation and the average dopant separation (N_a^{-1/3}). The valence band edge shift is neglected because freeze-out leaves too few free carriers to sustain that long-range effect. Valence band tailing remains part of the disorder contribution.

To remove a nonphysical divergence at (K’ = 0), the inverse-distribution term is replaced by a power law. The approximation gives (R^2 > 0.997) for (n > 3), while (n’ \approx 1.56n - 1.87) gives (R^2 > 0.999). Thus the final model is analytical and has no explicit temperature dependence.

The authors fitted two parameters, (\gamma) and (n’), with nonlinear least squares in MATLAB. Stage one used published single-crystal data with an explicit (K’); one reference supplied only a mean (K’). Restricting the calibration to this set reduced microstructural uncertainty and constrained the fitted shape parameters.

The authors then held (\gamma \approx 2.5) and (n’ \approx 3) constant. Stage two fitted one effective (K’) across the broader dataset to represent a most probable compensation level. Data above (N_c) that showed metallic behavior were excluded from fitting. Points lacking (K’) were retained for a separate comparison fit.

Compensation ratios from Hall measurements depend on the assumed hole effective density-of-states mass, whose literature values disperse. This shifts (K’) and weakens the parameter estimates.

QuantityFitted or fixed valueCondition
(N_c)(3 \times 10^{20}\ \mathrm{cm}^{-3})Fixed for the low to moderate compensation assumption
(E_{a0})0.37 eVFixed ground-state ionization energy
(n’)(3.0 \pm 1.0)Stage one fit
(\gamma)(2.5 \pm 1.1)Stage one fit
(n)(3.0 \pm 0.6)Derived from (n’)
(K’)0.14Stage two global fit
Feasible (K’)0.06 to 0.3095% parameter-boundary sensitivity

Most points with explicit (K’) fall between model curves for (K’ = 0) and (K’ = 0.5). This bracket is consistent with (K’ < 0.3) for realistic samples. The (K’ = 0.5) curve represents the adapted Pearson and Bardeen term.

The optimized global fit used (K’ = 0.14) and produced the 29% sum of squared residuals reduction. Adding points with unspecified (K’) yielded (K’ = 0.19) and a 20% reduction.

A sensitivity analysis over the 95% confidence limits gives a feasible (K’) range of 0.06 to 0.30. The reported compensation average of about 0.07 at median and low doping lies near the lower boundary.

The central (\gamma) gives a Mott criterion constant (C_M \approx 0.25), consistent with Mott’s prediction and prior boron-doped diamond observations. The corresponding uncertainty range is 0.17 to 0.44. Fontaine’s model included temperature-dependent terms, whereas this model does not. Both treatments identify impurity band broadening as the principal reduction mechanism.

The (\gamma) parameter remains poorly constrained because experimental points are sparse in the high-doping, low-(K’) regime. Activation energies were usually extracted from the high-temperature linear region of Arrhenius plots. At higher (N_a), hopping conduction shifts to higher temperatures and overlaps that region, compressing the usable fitting window. The authors suggest this helps explain why many heavily doped studies omit a reported (K’).

At high doping, points below the (K’ = 0.14) curve may reflect increased point-defect incorporation during CVD growth. That explanation is proposed in the paper and was not measured directly.

Because the literature data came from different reactors and growth conditions, 0.14 is an effective macroscopic baseline rather than a universal boron concentration and compensation relationship. The model can accept a concentration-dependent (K’(N_a)) profile when a reactor’s behavior is known. The constant-(N_c) assumption may also change slightly with compensation.

For diamond power devices, nominal boron concentration alone does not define the activated hole population. Compensation sets impurity band occupancy, while electrostatic disorder sets part of the activation-energy reduction.

Engineers can supply a reactor-specific (K’(N_a)) profile to predict drift-layer conductivity and specific on-resistance, (R_\mathrm{on}). TCAD modelers can use the closed-form expression without an inverse cumulative distribution function.

The authors propose using the two-stage fitting method to extract effective (K’) from new activation-energy data and compare CVD growth conditions. The procedure links Hall measurements, compensation, and device transport in one fitting process. A reactor-specific fit should replace the global baseline before using the model for process qualification.

For single-crystal boron-doped diamond used in semiconducting drift layers, the model requires the acceptor concentration and compensation ratio. 6CCVD’s diamond materials catalog includes boron-doped diamond, and its single-crystal and polycrystalline diamond are grown by MPCVD. The engineering support link can be used to specify the material and measurement requirements needed to feed these model inputs.

View Original Abstract

A semi-empirical analytical model which describes the reduction of impurity-to-band activation energy in boron-doped diamond, induced by doping and compensation levels. The model’s foundation mainly lies in the broadening of the impurity band. Crucially, compensation is integrated into this framework as a key multifaceted variable accounting for electrostatic disorder and impurity band occupancy. A two-stage optimization methodology is used for the comparative analysis with available experimental data. This model serves as a robust predictive framework for the electrical characterization and design of diamond-based semiconductor devices. The model’s predictions align closely with literature data, significantly outperforming previous theoretical frameworks by capturing the influence of compensation across the entire doping range.

  1. 2024 - Diamonds in the current: Navigating challenges for the integration of diamond in power electronics
  2. 2023 - Non-volatile tuning of normally-on and off states of deep depletion ZrO2/O-terminated high voltage diamond MOSFET [Crossref]
  3. 2026 - Boron-doped diamond solution-gate field-effect transistor (BDD-SGFET) biosensor for gene mutation detection [Crossref]
  4. 2025 - Superconducting diamond for quantum applications: CVD diamond doping control and characterization
  5. 1949 - Electrical properties of pure silicon and silicon alloys containing boron and phosphorus [Crossref]
  6. 1998 - Activation energy in low compensated homoepitaxial boron-doped diamond films [Crossref]
  7. 2020 - Modeling current transport in boron-doped diamond at high electric fields including self-heating effect [Crossref]
  8. 2021 - Recent progress in deep-depletion diamond metal-oxide-semiconductor field-effect transistors [Crossref]
  9. 1996 - The relationship between resistivity and boron doping concentration of single and polycrystalline diamond [Crossref]