Carrier Mobility BENCHMARK
This page reports electron mobility (μe) at 300 K predicted from composition alone for 23 semiconductors spanning III-V, II-VI, IV-VI, and elemental families. 6.2% MAPE using production transport physics.
This page reports electron mobility (μe) at 300 K predicted from composition alone for 23 semiconductors spanning III-V, II-VI, IV-VI, and elemental families. 6.2% MAPE using production transport physics.
Performance by material family, sorted by accuracy
| Family | N | MAPE | Worst Material | Status |
|---|---|---|---|---|
| Elemental sp³ (Si, Ge, C) | 3 | 0.2% | Ge −0.4% | Excellent |
| III-V (direct-Γ + indirect) | 9 | 5.0% | GaP −9.5% | Excellent |
| IV-VI narrow-gap | 3 | 6.4% | PbTe −12.5% | Excellent |
| II-VI semiconductor | 7 | 8.4% | CdTe +13.6% | Excellent |
| IV-IV indirect (SiC) | 1 | 20.4% | SiC +20.4% | Edge case |
| OVERALL | 23 | 6.2% | SiC +20% | Excellent |
All predictions from composition only. 22 of 23 materials sit within ±15% of experiment; SiC at +20% is the only outlier.
Complete benchmark cohort, sorted by experimental mobility
| Material | Family | Exp. μe (cm²/V·s) |
FluxMateria (cm²/V·s) |
Error | Status |
|---|---|---|---|---|---|
| AlP | III-V indirect | 80 | 81 | +1.0% | PASS |
| AlN | III-V wurtzite | 135 | 140 | +3.4% | PASS |
| ZnS | II-VI | 165 | 145 | −12.0% | PASS |
| AlAs | III-V indirect | 200 | 185 | −7.4% | PASS |
| AlSb | III-V indirect | 200 | 192 | −4.1% | PASS |
| GaP | III-V indirect | 200 | 181 | −9.5% | PASS |
| ZnO | II-VI oxide | 210 | 195 | −7.1% | PASS |
| ZnTe | II-VI | 340 | 372 | +9.4% | PASS |
| CdS | II-VI | 350 | 383 | +9.3% | PASS |
| ZnSe | II-VI | 500 | 503 | +0.6% | PASS |
| PbS | IV-VI | 600 | 633 | +5.6% | PASS |
| SiC | IV-IV indirect | 800 | 963 | +20.4% | EDGE |
| CdSe | II-VI | 900 | 840 | −6.7% | PASS |
| GaN | III-V wurtzite | 1,000 | 975 | −2.5% | PASS |
| PbSe | IV-VI | 1,000 | 1,010 | +1.0% | PASS |
| CdTe | II-VI | 1,050 | 1,193 | +13.6% | PASS |
| Si | Elemental sp³ | 1,400 | 1,403 | +0.2% | PASS |
| PbTe | IV-VI | 1,730 | 1,514 | −12.5% | PASS |
| C (diamond) | Elemental sp³ | 2,200 | 2,197 | −0.1% | PASS |
| GaSb | III-V narrow-gap | 3,000 | 2,862 | −4.6% | PASS |
| InN | III-V wurtzite | 3,200 | 3,463 | +8.2% | PASS |
| Ge | Elemental sp³ | 3,900 | 3,883 | −0.4% | PASS |
| GaAs | III-V direct-Γ | 8,500 | 8,292 | −2.5% | PASS |
Complete benchmark cohort. Experimental values from Madelung (2004), Yu & Cardona (2010), and the NSM Database. Mobilities at 300 K, lightly-doped or intrinsic single-crystal samples.
How predictions spread around experimental values
10 of 23 materials match experiment to within ±5%. The signed errors balance across the cohort: roughly half over-predictions and half under-predictions, with no systematic bias toward higher or lower mobility. Only SiC sits outside ±15%; the residual stems from a known reference-value convention (averaged density-of-states vs band-edge mobility) rather than a missing physics term.
Carrier mobility prediction trade-offs: accuracy, speed, and data dependence
| Metric | FluxMateria | DFT (Boltzmann transport) | ML (CGCNN / GNN) |
|---|---|---|---|
| μe error | 6.2% MAPE (23 materials) 10/23 within ±5% |
20–100% typical (per material) | 30–200% MAPE on out-of-domain |
| Required input | Composition only | Full crystal structure + scattering parameters | Composition + featurized graph + training set |
| Runtime | Sub-millisecond | Hours per material | Milliseconds inference, hours training |
| Training data | None | None (calibrated functional) | Thousands of labeled mobility values |
| Fitted parameters | 0 fitted | Pseudopotential + scattering rate cutoffs | Millions |
| Novel compositions | Physics-grounded extrapolation | Recompute required | Often degrades beyond training domain |
Key takeaway: FluxMateria delivers 6.2% MAPE on intrinsic electron mobility from composition alone across 23 semiconductors covering all the major industrial families (Si / Ge / GaAs / GaN / InP / III-V / II-VI / IV-VI / SiC), at sub-millisecond runtime, with no crystal structure input. DFT-based Boltzmann transport remains the ab initio reference but requires full structure relaxation and is orders of magnitude slower. ML approaches need large labeled mobility datasets (rare in the literature) and degrade on novel chemistries.
The single number that decides how fast a semiconductor can switch, sense, or convert energy
Carrier mobility (μ) measures how quickly electrons (or holes) drift through a material when an electric field is applied. It is the proportionality constant between drift velocity and field, reported in cm²/V·s. A higher mobility means current flows more easily, switches turn on and off faster, and less energy is lost to heat for the same delivered current.
Mobility is intrinsic to the material: it is set by composition, bonding, and how strongly carriers scatter off the lattice. It is what separates a 1,400 cm²/V·s silicon wafer from an 8,500 cm²/V·s GaAs wafer — and is the reason GaAs dominates RF and Si dominates logic.
Pick the channel material for a new transistor generation. Mobility decides whether a candidate clears the speed/power target before any wafer is grown.
Compare wide-bandgap candidates (GaN, SiC, Ga₂O₃, diamond) for EV inverters, data-center power, and grid converters where every percent of efficiency is billions in fuel and cooling.
Screen new absorber chemistries (perovskites, chalcogenides, kesterites) for adequate carrier collection before committing months to film growth and device fabrication.
Maximise the power factor σS² (thermoelectrics) or organic-TFT mobility (printed displays, RFID, biosensors), where mobility is the single biggest lever on figure of merit.
Rank thousands of candidate compositions per day for transport quality — something DFT cannot do at that throughput, and ML surrogates cannot do reliably outside their training set.
De-risk supplier qualification and process-node decisions where a wrong call on a channel material costs nine figures and 18 months of lost time-to-market.
Today, getting a trustworthy mobility number means either a multi-week DFT + Boltzmann-transport pipeline or an ML model that only works on materials similar to its training set. Both are bottlenecks the moment you want to screen a new chemistry, an alloyed composition, or a defect-engineered variant.
A composition-only predictor that returns μe in milliseconds, with benchmark evidence used for validation, collapses that bottleneck. It moves mobility from a downstream verification step into an upstream design variable that every materials and device team can iterate against in real time.
How FluxMateria predicts intrinsic electron drift mobility
Electron mobilities are computed from composition by the production electronic-transport route. The engine identifies the bonding family, derives the carrier transport channel, and returns μe at room temperature using the production transport model.
Si, Ge, C (diamond)
GaAs, GaSb, InN…
AlAs, AlP, AlSb, GaP
GaN, AlN, InN
ZnO, ZnS, ZnSe, ZnTe, CdS, CdSe, CdTe
PbS, PbSe, PbTe
SiC
Primary data sources for experimental validation
Electron mobility is computed from Flux transport physics with fixed endpoint scaling. The semiconductor reference set is used to measure accuracy across material families.
Enter any semiconductor formula and get electron mobility predictions alongside band gap, elastic, thermal, and optical properties — all from composition alone.