← All benchmarks Back to Solar Device Studio →
FluxMateria Materials · Device benchmark

Solar Device Studio benchmark.

A fixed-gate evaluation of parameterized crystalline-Si Al-BSF device response. Internal qualification and independent published-profile transfer evidence are reported separately; external laboratory qualification remains open.

Crystalline-Si Al-BSF only Jsc · Voc · fill factor · PCE No fitted correction factors External lab qualification pending
4 / 4
Terminal gates passed
On both reported evidence tracks
384
Internal devices
One production lineage; parameterized inputs
1
Independent linked device
Published same-device transfer check
81 / 81
Focused regressions
Executed solar tests passing

Read the evidence in three lanes

The qualification label matters as much as the number.

Lane 01 · passed

Internal parameterized qualification

384 complete device rows from one industrial production lineage. The model receives declared optical, lifetime, doping, junction, contact, and geometry inputs; measured terminal outcomes are retained for scoring.

Lane 02 · positive transfer evidence

Independent published-profile check

One same-device record links the reported physical state, a 91-point published optical profile, and independently validated terminal performance. It passes all four gates, but it is not a sealed external test.

Lane 03 · pending

External laboratory qualification

Still requires instrument-native optical files, direct doping measurements, multiple cells, multiple process lots, and an independent blind evaluation. No production certification is claimed.

Results against frozen gates

Mean absolute error is reported in each metric's native unit.

Internal parameterized qualification · 384 devices4 / 4 PASS
MetricWorst-case MAEFrozen goalResult
Short-circuit current density (Jsc)0.979306 mA/cm²≤ 2.0 mA/cm²PASS
Open-circuit voltage (Voc)0.004815 V≤ 0.025 VPASS
Fill factor0.002363≤ 0.035PASS
Power-conversion efficiency (PCE)0.419561 pp≤ 2.0 ppPASS
Scope: one production lineage. Reported values are the worse retained endpoint where a physical boundary envelope applies. This is strong internal engineering evidence, not cross-factory validation.
Independent published-profile transfer check · n = 14 / 4 PASS
MetricFull-width worst errorFrozen goalResult
Short-circuit current density (Jsc)0.671325 mA/cm²≤ 2.0 mA/cm²PASS
Open-circuit voltage (Voc)0.011782 V≤ 0.025 VPASS
Fill factor0.016185≤ 0.035PASS
Power-conversion efficiency (PCE)0.277840 pp≤ 2.0 ppPASS
Scope: one 18.9% screen-printed Al-BSF reference cell. The full published optical-width uncertainty and both retained rear-boundary endpoints are included. This is positive transfer evidence, not external certification.

How much of each error budget was used?

Every metric is normalized to its own frozen maximum so unlike units can be compared honestly.

Normalized terminal-error use by evidence track

Internal parameterized qualification · n = 384 Independent published-profile transfer · n = 1 100% = frozen gate
JscGoal ≤ 2.0 mA/cm²
Internal
49.0%
Transfer
33.6%
VocGoal ≤ 0.025 V
Internal
19.3%
Transfer
47.1%
Fill factorGoal ≤ 0.035
Internal
6.8%
Transfer
46.2%
PCEGoal ≤ 2.0 percentage points
Internal
21.0%
Transfer
13.9%
Lower is better. A bar at 50% uses half of the permitted error; a bar reaching 100% touches the failure boundary. These values are error-to-gate ratios, not an invented overall accuracy percentage. Both evidence tracks remain below every frozen gate.

Benchmark protocol

A parameterized device test: physical state in, terminal outcomes scored afterward.

Declared before prediction

  • Ordered cell architecture and layer geometry
  • Measured optical response or equivalent spectral input
  • Base doping and bulk lifetime
  • Junction recombination and series resistance
  • Illumination, temperature, active area, and provenance

Withheld until scoring

  • Measured short-circuit current density
  • Measured open-circuit voltage
  • Measured fill factor
  • Measured power-conversion efficiency
  • Any residual chosen to make the prediction agree

Fixed acceptance gates

  • Jsc MAE ≤ 2.0 mA/cm²
  • Voc MAE ≤ 0.025 V
  • Fill-factor MAE ≤ 0.035
  • PCE MAE ≤ 2.0 percentage points
  • 100% focused regression success

Fail-closed rules

  • No silent imputation of missing required inputs
  • No target-selected physical boundary
  • No fitted residual or empirical correction factor
  • Failed and incomplete rows remain visible
  • Qualification labels cannot be promoted by accuracy alone
Optical robustness: the independent check uses all 91 published reflectance markers and propagates the complete reported minimum interval. The former single-anchor reconstruction was discarded because it allowed an arbitrary point inside a flat minimum to own Jsc.

Coverage and limitations

What this benchmark establishes—and what it does not.

Evidence inventory

Industrial source rows reviewed404
Internally scored complete rows384
Independent terminal devices located7
Independent source lineages5
Fully linked published-profile candidates1
Published reflectance markers used91

Not established yet

  • Independent blind performance across multiple factories or process lots
  • Prediction from fabrication recipe alone
  • Manufacturing yield, degradation, reliability, or bankability
  • PERC, TOPCon, heterojunction, tandem, or thin-film qualification
  • Non-silicon absorber qualification
  • Production or warranty-grade certification
Downloadable benchmark evidence

Take the numbers with you.

The public packet contains the machine-readable benchmark CSV, a scoped public receipt, source lineage, limitations, and SHA-256 checksums. The 384-device internal qualification and the one-device independent transfer check remain explicitly separated.

Download benchmark packet (ZIP) Download results (CSV) View public receipt (JSON)

Eight result rows · four frozen metrics per evidence track · no fitted correction factors · external lab qualification pending

← Return to the Solar Device Studio module

Sources and next reading

Primary evidence context and the practical engineering story.

Industrial dataset

400 industrial crystalline-Si cells

The public measurement dataset underlying the internal development panel.
Independent primary source

Georgia Tech Al-BSF reference cell

The dissertation source for the same-device published-profile transfer check.
Case study

A 28.51% SOTA-class HBC design projection

Follow the precedent-anchored 75-µm intervention, explicit optical penalty, 5.13% modeled gain, and laboratory test boundary.
Module

Explore Solar Device Studio

See the current capability, field workflow, evidence boundaries, and expansion path.

A benchmark should make the boundary visible.

The Si Al-BSF result is internally qualified and independently checked against one published same-device profile. The next scientific step is direct-input, multi-lot laboratory validation.

Request the validation packet →