# FluxMateria White Paper # Solar Absorber Discovery and CdTe Replacement Challenge Date: 2026-04-09 ## Executive Summary FluxMateria ran a full solar absorber-and-contact workflow locally in `28.003` seconds. It then ran a stricter incumbent-replacement challenge in `51.78` seconds. The key result was a winner shift: - `InP` won the absorber-only screen - `CdTe` won the full device-stack screen - `Si` emerged as the strongest conservative build package - `CuS` and `CuSe` cleared the defined novel-replacement gate against the `CdTe + ZnO + Mo` baseline That is the point of the study. Solar engineering teams often treat these as separate questions: - what is the best absorber? - what is the best contact stack? - what is the best near-term build candidate? FluxMateria kept those questions inside one workflow and showed that they do not lead to the same answer. The most important validation point is that the workflow converged quickly on real photovoltaic neighborhoods: - mainstream incumbents such as `Si`, `CdTe`, `InP`, and `GaAs` - chalcopyrite edge cases such as `CuInSe2` - earth-abundant copper and tin chalcogenide replacement lanes such as `CuS`, `CuSe`, and `Sn-S-Se` That is not proof of certified device performance. It is evidence that FluxMateria can compress the computational decision layer down to minutes while still landing near chemistries that the solar field has spent years exploring. ## Why This Study Matters Absorber-only ranking is useful, but it is incomplete. A solar material can look excellent in bulk and still stop being the best practical device choice once front-contact and back-contact behavior enter the picture. This study was designed to test exactly that. The working thesis was simple: > the best solar absorber is not necessarily the best solar device stack. The final result supported that thesis cleanly. ## What FluxMateria Did The workflow combined five layers: 1. Curated absorber framing across known photovoltaic families. 2. Bulk absorber screening. 3. Batched front-contact and back-contact evaluation. 4. Solar-native device scoring. 5. Benchmark-calibrated uncertainty and prototype handoff. This matters because the workflow did not stop at band gap or bulk absorber quality. It ranked full device stacks. ## How FluxMateria Got Here In `28.003` Seconds The public pipeline is: 1. Curate the absorber pool. 2. Rank the pool on absorber merit. 3. Sweep realistic front and back contacts. 4. Re-rank the resulting stacks with a solar-native device layer. 5. Attach benchmark support, uncertainty, and build guidance. That is why the final result was not a single simplistic "best material." It was a structured answer: - `InP` for absorber-only strength - `CdTe + ZnO + Mo` for the strongest device stack - `Si` for the strongest conservative build handoff - `CuS` and `CuSe` for the strongest novel CdTe-replacement hypotheses ## Benchmark Credibility The solar-native layer was benchmarked before the final rerun. Calibrated benchmark summary: - `reference count`: `6` - `holdout reference count`: `3` - `family accuracy`: `1.0` - `pce proxy MAE`: `1.887` - `stability MAE`: `0.035` - `cost MAE`: `0.1193` - `manufacturing MAE`: `0.0667` - `scenario alignment`: `4/4` Interpretation: - the module kept the photovoltaic-family distinctions correct on the holdout set - it aligned with all four public-facing benchmark scenarios - the calibration tightened the practical engineering layers without collapsing the ranking behavior Important nuance: - calibration did not improve every metric equally - that is normal - the right public claim is that the module behaved coherently across references and scenarios, not that calibration improved every number ## The Winner Shift ### 1. Absorber-Only Winner: `InP` - direct bulk winner inside the curated pool - strong absorber-only fit - remained highly credible even after full-stack re-ranking Interpretation: - the absorber-only logic still likes III-V quality - FluxMateria did not "break" the bulk answer - it just showed that the bulk answer is not the whole engineering answer ### 2. Full Device-Stack Winner: `CdTe + ZnO + Mo` - absorber: `CdTe` - front contact: `ZnO` - back contact: `Mo` - `pce_proxy_pct`: `21.1` - `solar_device_readiness_score`: `72.0` This was the clearest device-level correction in the study. `CdTe` moved from bulk rank `4` to device rank `1`. Interpretation: - once real contact choices were allowed to matter, the winner changed - this is the strongest evidence that stack co-design matters ### 3. Conservative Build Winner: `Si` - `prototype_handoff_priority_score`: `79.0` - recommended first stack: `ZnO front / NiO back` This is a different kind of winner. `Si` is not the top device-stack answer in this workflow. It is the strongest conservative build package because it combines: - high manufacturing readiness - strong cost competitiveness - strong stability - lower practical build risk Interpretation: - FluxMateria is now separating `best full-stack candidate` from `best immediate build candidate` - that is exactly what a serious engineering workflow should do ## Second-Pass Replacement Challenge After the stack workflow selected `CdTe + ZnO + Mo`, FluxMateria asked a stricter question: > can a novel absorber lane beat the CdTe baseline under a broader replacement objective? The replacement objective combined: - device readiness - PCE proxy - stability - manufacturing readiness - cost competitiveness - safety and toxicity burden - supply-chain resilience - interface recombination burden - passivation burden - novelty - architecture fit The local replacement challenge completed in `51.78 s`. Baseline: - `CdTe + ZnO front / Mo back` Top replacement candidates: ### `CuS` - Best stack: `ZnSe front / Al back` - Incumbent replacement score: `79.0` - Relative replacement gain vs CdTe baseline: `+20.65%` - Solar device readiness score: `78.4` - PCE proxy: `18.09%` - Result: cleared the replacement gate Interpretation: - `CuS` is the highest-upside replacement hypothesis in this pass - it won because of absorber-window fit, defect-tolerant chemistry, low passivation burden, low-cost chemistry, and acceptable architecture fit - it remains experimentally unproven as a CdTe replacement stack ### `CuSe` - Best stack: `ZnO front / Mo back` - Incumbent replacement score: `75.4` - Relative replacement gain vs CdTe baseline: `+15.09%` - Solar device readiness score: `78.4` - PCE proxy: `18.4%` - Result: cleared the replacement gate Interpretation: - `CuSe` is the more literature-adjacent replacement lane - it keeps the `ZnO / Mo` stack logic that worked for CdTe - it improves the modeled replacement objective through lower hazard and supply-chain pressure ### Near-miss frontier lanes The tin chalcogenide candidates remained important: - `Sn2S3Se` - `Sn2SeS3` - `SnSeS` They scored strongly on novelty and cost, but did not clear the strict replacement gate because readiness, stability, or manufacturing terms fell too far below the CdTe baseline. That is useful, not disappointing. The module distinguished: - literature-backed frontier chemistry - credible replacement chemistry - candidates that are interesting but not ready to displace an incumbent ## Public Shortlist ### `CdTe` - Role: top full-stack candidate - Best stack: `ZnO front / Mo back` - Why it matters: the clearest proof that interface-aware scoring changed the answer ### `InP` - Role: top bulk-origin candidate - Best stack in this study: `ZnO front / Al back` - Why it matters: the bulk winner stayed strong, but not unbeatable ### `Si` - Role: top immediate build candidate - Best stack in this study: `ZnO front / NiO back` - Why it matters: the most conservative prototype path is not always the same as the highest-upside device path ### `GaAs` - Role: high-efficiency comparator - Why it matters: a serious III-V comparator that remained credible but not dominant ### `WS2` - Role: exploratory thin-film comparator - Why it matters: a useful edge-of-shortlist case that keeps the workflow from collapsing into only incumbent materials ## Literature Context This is one of the strongest validation points in the study. FluxMateria did not converge on arbitrary chemistry. It converged on material families the field already treats as serious photovoltaic lineages. ### `Si`, `CdTe`, and III-V families NREL's best research-cell efficiency tracking keeps crystalline silicon, CdTe, and III-V families on the same chart of real photovoltaic lineages. Interpretation: - the shortlist stayed inside real solar design space - `InP`, `GaAs`, `CdTe`, and `Si` are not decorative outputs - they are historically serious absorber families ### `CdTe` contact engineering CdTe front-contact alignment has been studied directly, including ZnO-based contact layers. Interpretation: - the `CdTe + ZnO` result is physically credible - FluxMateria did not invent an arbitrary front-contact pair - it converged on a real device-level design variable ### `Si` contact engineering The silicon literature is explicit that modern device performance depends heavily on contact design and passivating-contact strategies. Interpretation: - `Si` emerging as the strongest conservative build package is a serious result - it reflects the fact that manufacturable silicon is still highly relevant when engineering practicality matters ### `CuInSe2` and chalcopyrite back contacts CuInSe2 and related chalcopyrite absorbers remain strongly associated with Mo and MoSe2 back-contact behavior in the literature. Interpretation: - CuInSe2 is still a meaningful refinement target even when it does not survive the strict public shortlist - FluxMateria's treatment of CuInSe2 as an interesting edge case is scientifically defensible ### Copper chalcogenide replacement lanes The `CuS` and `CuSe` result should be interpreted carefully. The literature does not prove that the exact FluxMateria stacks are commercial CdTe replacements. It does show that copper chalcogenides are real photovoltaic research neighborhoods: - `CuS` has appeared in thin-film photovoltaic device work - `CuSe` has appeared in photovoltaic device work and in CdTe absorber-layer modification contexts - copper chalcogenide chemistry is close to broader earth-abundant thin-film PV research efforts Interpretation: - FluxMateria's top replacement lanes are not random - the exact stacks remain build hypotheses - the computational novelty is the rapid integration of absorber quality, contact fit, cost, hazard, passivation, and incumbent-replacement scoring ### Tin chalcogenide near-miss lanes The `Sn-S-Se` candidates are strongly literature-aligned. SnS, SnSe, SnSx, and related mixed chalcogenides are studied because they are earth-abundant and potentially low-cost absorber systems. Interpretation: - FluxMateria correctly kept this family alive - the stricter replacement gate prevented overclaiming - this is a good sign for the pipeline because it can separate "interesting frontier material" from "credible incumbent replacement" ## Why The Runtime Matters The main runtime claim is not that one predictor was fast. The important claim is that FluxMateria completed two normally fragmented decision workflows locally in under a minute each: - first stack co-design pass: `28.003 s` - stricter replacement pass: `51.78 s` That is the right level to compare against conventional fragmented workflows. The relevant comparison is not just a single model call. The comparison is against the conventional sequence of: - selecting absorber candidates - separately screening contact layers - separately considering defect tolerance and passivation - separately considering cost, hazard, and manufacturability - only later deciding what to fabricate FluxMateria compressed that decision layer into a single local loop. ## What This Study Shows - the best absorber is not automatically the best device stack - contact-stack modeling can change the engineering winner - the winner shift survives a dedicated solar-native module - the shortlist is literature-grounded - the pipeline can distinguish `highest-upside stack` from `lowest-friction build` - the pipeline can also identify novel replacement lanes and reject attractive near-misses when the full replacement objective does not support them ## What This Study Does Not Show - certified device efficiency - long-duration reliability proof - full fab-line optimization - the best solar device ever built - that `CuS` or `CuSe` is already a certified CdTe replacement Those remain experimental questions. That is exactly the point: FluxMateria is not replacing fabrication and measurement. It is compressing the decision layer so that fabrication and measurement can start from a much sharper shortlist. ## Selected References - NREL, *Best Research-Cell Efficiencies* https://www.nrel.gov/pv/assets/pdfs/best-research-cell-efficiencies.20210926.pdf - *Band alignment of front contact layers for high-efficiency CdTe solar cells* https://www.osti.gov/servlets/purl/1534330 - *Silicon heterojunction solar cells with passivating contacts: Classification and advanced fabrication strategies* https://www.sciencedirect.com/science/article/abs/pii/S1369702124002088 - *CuInSe2 thin films: preparation, structure, properties, and solar cells* https://www.osti.gov/biblio/5168746 - *Mo/Cu(In,Ga)Se2 back interface chemical and optical properties for ultrathin CIGSe solar cells* https://www.sciencedirect.com/science/article/abs/pii/S0169433211017880 - *CuS thin-film photovoltaic device context* https://www.mdpi.com/1996-1073/13/3/688 - *CdSe/CuSe flexible photovoltaic device context* https://pubs.rsc.org/en/content/articlelanding/2013/cp/c3cp50435b - *New Earth-Abundant Thin Film Solar Cells Based on Chalcogenides* https://www.frontiersin.org/journals/chemistry/articles/10.3389/fchem.2019.00297/full - *SnSe/SnS bilayer absorber context* https://www.sciencedirect.com/science/article/pii/S0038092X23008083 ## Research Use Note This is an in-silico engineering-screening study. Real device fabrication, efficiency certification, durability testing, and manufacturing validation still require laboratory work and production-scale engineering.