Paramus — Band Gap Semiconductor Screen

Overview

Predict band gaps for candidate materials and rank them against a target window. The value is the screen-and-rank loop with electronic-materials judgment, not a single band-gap number.

Guidance vs execution: band gaps come from tested Paramus ML/DFT tools with provenance. Do not estimate gaps from empirical rules alone.

When to Use

• Ranking candidate compositions/structures for a target band gap (PV, LED, sensor)
• Pre-screening a materials library before expensive DFT/experiment Do not use for: full optoelectronic device modeling (hand off to the engineering workflow).

Workflow

Candidate materials (compositions / structures) + target gap window + application
  ↓ 1. Parse & validate structures  → Paramus materials parser
  ↓ 2. Featurize                    → composition/structure descriptors
  ↓ 3. Predict band gap             → Paramus ML model (fast) or DFT engine (accurate)
  ↓ 4. Score vs target window       → inside/outside + margin
  ↓ 5. Rank & report                → ranked table + gap ± uncertainty + provenance

Procedure

1. Discover tools via search/get_schema ("band gap prediction", "materials featurization").
2. Parse/validate structures or compositions; report failures.
3. Featurize consistently across the set.
4. Predict band gaps (ML for breadth; escalate top hits to DFT for accuracy). Capture uncertainty + method + version.
5. Score against the target window; rank and report with confidence.

Domain Judgment

• ML band gaps are a fast filter; escalate promising candidates to DFT before decisions.
• Note the known DFT band-gap underestimation; state the method and any correction.
• Uncertainty straddling the window boundary → "uncertain-fit", not a confident hit.

Fallbacks

• DFT engine unavailable → deliver ML-only ranking, clearly marked as a screen not a decision.
• Endpoint unreachable → stop and report; no local band-gap estimates.