Life-Cycle Assessment of Innovative Industrial Processes for Photovoltaic Production: Process-Level LCIs, Scale-Up Dynamics, and Recycling Implications

Featured Application

Compiled process-level LCIs and scale-up insights can support sustainable manufacturing design, recycling strategies, and policy development for next-generation photovoltaic technologies.

Abstract

The rapid commercialization of next-generation photovoltaic (PV) technologies, particularly perovskite, thin-film roll-to-roll (R2R) architectures, and tandem devices, requires robust assessment of environmental performance at the level of industrial manufacturing processes. Environmental impacts can no longer be evaluated solely at the device or module level. Although many life-cycle assessment (LCA) studies compare silicon, cadmium telluride (CdTe), copper indium gallium selenide (CIGS), and perovskite technologies, most rely on aggregated indicators and database-level inventories. Few studies systematically compile and harmonize process-level life-cycle inventories (LCIs) for the manufacturing steps that differentiate emerging industrial routes, such as solution coating, R2R processing, atomic layer deposition, low-temperature annealing, and advanced encapsulation–metallization strategies. In addition, inconsistencies in functional units, system boundaries, electricity-mix assumptions, and scale-up modeling continue to limit meaningful cross-study comparison. To address these gaps, this review (i) compiles and critically analyzes process-resolved LCIs for innovative PV manufacturing routes across laboratory, pilot, and industrial scales; (ii) quantifies sensitivity to scale-up, yield, throughput, and electricity carbon intensity; and (iii) proposes standardized methodological rules and open-access LCI templates to improve reproducibility, comparability, and integration with techno-economic and prospective LCA models. The review also synthesizes current evidence on recycling, circularity, and critical-material management. It highlights that end-of-life (EoL) benefits for emerging PV technologies are highly conditional and remain less mature than for crystalline-silicon systems. By shifting the analytical focus from technology class to manufacturing process and life-cycle configuration, this work provides a harmonized evidence base to support scalable, circular, and low-carbon industrial pathways for next-generation PV technologies.

1. Introduction

The rapid expansion of global PV manufacturing marks a decisive phase in the clean-energy transition. By 2024, global installed PV capacity exceeded 1.6 TW, with annual production surpassing 400 GWp and a compound growth rate near 25% yr⁻¹ [1,2,3]. This surge has been driven by efficiency improvements, cost declines following a 20% learning rate per cumulative doubling of production [4,5,6], and increasing policy alignment with net-zero roadmaps [7,8]. As industrial production scales to the multi-gigawatt and terawatt levels, understanding the environmental performance of PV technologies is no longer limited to device physics or module-level metrics; it must encompass the manufacturing processes themselves that enable large-scale deployment.

Although PV technologies differ in materials, device structures, and manufacturing routes, most modern modules follow a broadly similar production sequence. Manufacturing typically begins with substrate or wafer preparation, such as cutting, texturing, or cleaning glass and silicon surfaces, followed by thin-film or wafer processing, which may involve coating, printing, chemical solution deposition, vapor deposition, or high-vacuum sputtering [3]. These layers are then consolidated through annealing or other thermal treatments to form the functional semiconductor stack. Metallization establishes the electrical contacts needed for charge extraction, while encapsulation and lamination protect the device from moisture, oxygen, and mechanical stress [8]. Finally, module assembly integrates the laminated cell stack with the frame, junction box, and wiring to produce the finished PV module. Each of these steps is associated with specific energy and material flows, which together determine the overall manufacturing footprint of PV technologies [3,8].

Previous comparative LCA studies of PV technologies have mainly evaluated different device classes, such as crystalline silicon, CdTe, CIGS, and perovskite, using high-level indicators. These indicators often include cradle-to-gate impacts, which cover environmental burdens from raw-material extraction (‘cradle’) to the point the product leaves the factory (‘gate’), and cradle-to-grave impacts, which additionally include use and end-of-life (EoL) stages. Another commonly used indicator is energy payback time (EPBT), which measures how long a PV system must operate to generate the amount of energy that was required to produce it, and global-warming potential (GWP), which expresses climate impacts in terms of CO₂-equivalent emissions over a specified time horizon [9,10,11,12].

Several studies, including Muteri et al. [9], Milousi et al. [10], and the ReCiPe-based assessment of Rashedi and Khanam [11], consistently report that thin-film technologies generally exhibit lower embodied energy and greenhouse-gas (GHG) emissions than conventional wafer-based silicon modules, largely due to lower material consumption and lower-temperature processing. ReCiPe is a widely used life-cycle impact assessment (LCIA) method that converts emissions and resource flows into midpoint indicators, such as climate change, toxicity, and endpoint indicators like human health and ecosystem quality, allowing for consistent comparison of environmental burdens [11]. However, these comparative assessments typically rely on aggregated LCI datasets (e.g., Ecoinvent [10]) and therefore do not fully resolve the underlying unit-process data that describe coating, deposition, annealing, encapsulation, and metallization steps that dominate manufacturing impacts [12,13,14].

As next-generation PV technologies, particularly perovskite, tandem, and roll-to-roll (R2R) flexible architectures, move closer to commercialization, the need for process-specific environmental data becomes more urgent. Recent studies of pilot and industrial production lines show that steps such as drying/annealing, substrate preparation, and encapsulation can each contribute more than 60% of total cradle-to-gate GWP [12,13,14]. Detailed inventories covering R2R slot-die coating [15,16], atomic layer deposition (ALD) [17], low-temperature solution processing [18], and vacuum sputtering [19] further demonstrate strong sensitivity to operational conditions, including throughput, yield losses, and the carbon intensity of the electricity mix [20,21]. Yet few evaluations harmonize these datasets or examine how environmental indicators evolve as manufacturing scales from laboratory to pilot to industrial production levels [22,23].

Alongside these developments, new prospective and dynamic LCA frameworks have emerged to integrate technological learning rates and electricity-grid decarbonization with manufacturing LCIs [4,24,25]. These methods show that environmental learning curves can reduce the projected cradle-to-gate footprints of monocrystalline-Si and perovskite PV by 20–80% toward 2050 [24,26]. This aligns with techno-economic scaling models that link production capacity (MWp yr⁻¹) to both cost and embodied-energy reductions [2,13]. Together, these insights indicate that scale itself is an environmental variable, not only an economic one.

EoL management has also gained prominence as global PV deployment expands. European industrial initiatives have demonstrated high-efficiency recovery of glass, aluminum, silver, and silicon from retired modules [27,28], while newer industrial trials report Ag and Si recovery rates above 90% [29,30]. Process-integrated LCAs show that closed-loop recycling can reduce cradle-to-grave GHG emissions by 40–80% [31,32], and policy analyses estimate more than 60 million tonnes of PV waste by 2045 in the EU alone [33]. Integrating these EoL pathways into manufacturing LCAs is therefore essential for circular-economy planning and assessing material criticality [6,34].

Despite this progress, no existing review systematically compiles and harmonizes process-level LCIs across the diverse manufacturing routes now emerging, from R2R coating and solution deposition to ALD, sputtering, and low-temperature encapsulation. Moreover, guidance on scaling rules, uncertainty treatment, and functional-unit normalization for such process inventories remains fragmented. Addressing these gaps is essential for consistent techno-environmental modeling and for designing the next generation of sustainable PV production lines.

This review therefore aims to: (i) compile and critically analyze process-level LCIs reported for innovative PV manufacturing routes, including R2R, vacuum, and solution-processed lines; (ii) quantify scale-up dynamics (lab->pilot->industrial) and sensitivity to yield, throughput, and electricity carbon intensity; (iii) propose standardized methodological rules and open-access LCI templates to enhance comparability, reproducibility, and integration with techno-economic assessments. By shifting the analytical focus from technology class to manufacturing process, this work contributes a harmonized evidence base for evaluating environmental trade-offs at an industrial scale and supports the transition toward circular, low-carbon PV production.

To clarify its contribution relative to prior PV LCA reviews, the novelty of this work lies not in producing a single new LCA result but in providing a harmonized, process-resolved synthesis framework. Specifically, this review (i) organizes manufacturing LCIs by dominant process families; (ii) explicitly addresses scale and technology readiness through data-quality tiers and TRL-aware aggregation checks; (iii) quantifies uncertainty using consistent distributions and Monte Carlo propagation; and (iv) offers an open-access reporting template and worked example to support reproducible comparison and forward-looking modeling. By shifting the analytical focus from technology class to manufacturing process, this work provides a coherent evidence base for evaluating environmental trade-offs at an industrial scale and supports the transition toward circular, low-carbon PV manufacturing pathways.

2. Methodology

2.1. Literature Search Strategy

A structured literature search was carried out in Web of Science, ScienceDirect, and Google Scholar to identify peer-reviewed LCA studies relevant to PV manufacturing. The search covered publications from January 2015 to November 2025. The following exact search terms were used, both individually and in paired Boolean combinations:

• ‘life cycle assessment’,
• ‘LCA’,
• ‘photovoltaic’,
• ‘solar cell’,
• ‘perovskite’,
• ‘thin-film’,
• ‘roll-to-roll’,
• ‘tandem’,
• ‘manufacturing process’,
• ‘recycling’.

The Boolean structure followed the format: (life cycle assessment OR LCA) AND (photovoltaic OR solar cell) AND (manufacturing process OR recycling OR perovskite OR thin-film OR roll-to-roll OR tandem).

This predefined search string ensured reproducibility and minimized ambiguity in record retrieval. The initial search yielded 1824 records (889 from ScienceDirect, 543 from Web of Science, and 392 from Google Scholar).

The subsequent screening and selection procedure is documented in Figure 1, which presents the complete literature-selection pathway in a PRISMA-style flow diagram, from the initial identification to the final inclusion of 102 studies. Records were excluded based on predefined criteria, primarily due to topic mismatch (e.g., non-PV waste streams), focus on single-scenario studies without comparative analysis, or insufficient methodological or inventory transparency. This structured screening approach ensures transparency and reproducibility of the review process in line with PRISMA principles.

2.2. Selection Criteria

To achieve consistency with the filtering steps shown in Figure 1, the inclusion and exclusion criteria were translated directly from the figure’s structure.

Inclusion filters (Figure 1, Step 2):

Articles were retained only if they met the following explicit criteria:

• Publication type: review and research article;
• Research area: environmental science or directly related subfields;
• Language: English.

Exclusion filters (Figure 1, Step 2):

Articles were excluded if they focused on:

• E-waste;
• Water waste;
• Medical waste;
• Construction waste;
• Hazardous waste.

Or if they presented only one treatment scenario, making them unsuitable for comparative environmental analysis.

Applying these transparent filters reduced the dataset from 1824 records to 102 articles, which were then included in the review.

2.3. Review Framework

Following the flow presented in Figure 1, the final set of 102 articles was examined to provide a general overview of the reviewed studies (Figure 1, Step 3). The analysis focused on identifying recurring themes in PV-related LCA research, including the types of manufacturing processes assessed, the system boundaries employed (cradle-to-gate or cradle-to-grave), and the methodological approaches used within environmental science research.

The selected studies were systematically analyzed based on:

• Technology focus: crystalline silicon, CdTe, CIGS, perovskite, tandem, or hybrid thin film;
• Manufacturing process coverage: deposition/coating, encapsulation, substrate preparation, metallization, and recycling;
• Scale and system boundary: laboratory, pilot, or industrial scale;
• Functional unit (FU) and impact categories (e.g., GWP, cumulative energy demand (CED), acidification potential (AP), eutrophication potential (EP));
• The FU is the quantified reference that defines what exactly is being compared in an LCA. It provides a consistent basis for calculating and normalizing environmental impacts across different technologies, processes, or scales;
• Data source type: experimental, modeled, or hybrid;
• LCA methodology: attributional vs. consequential, LCI data transparency, and software tools used;
• Treatment of EoL and circularity: recycling scenarios, recovery of critical materials, and reuse routes.

3. Technology and Process Landscape

Industrial PV manufacturing has diversified beyond conventional crystalline-silicon (c-Si) wafer routes toward perovskite, tandem, and thin-film architectures, bringing new materials, unit operations, and production configurations that reshape both environmental and techno-economic performance.

3.1. Perovskite, Tandem, and Thin-Film Routes: Technical Primer

Perovskite PVs have advanced from laboratory cells to pre-industrial modules exceeding 800 cm² and certified efficiency above 25% [1,14,18,35]. Two dominant device stacks, mesoporous n-i-p and inverted p-i-n, impose distinct sequences of deposition, annealing, and encapsulation with measurable differences in embodied impacts at the mini-module scale [14]. Tandem configurations that pair perovskite top cells with silicon or CIGS bottom cells provide a practical route beyond single-junction limits and are now a focal point of sustainability reviews and prospective LCAs [21,22]. Across technology families (Si, CdTe, CIGS, perovskite), comparative reviews consistently show that absorber chemistry governs toxicity/criticality profiles while substrates and encapsulants dominate energy use, particularly for flexible or building-integrated photovoltaic (BIPV) applications [19,36]. Low-temperature, carbon-electrode perovskites reduce both thermal budgets and metal intensity relative to noble-metal stacks, with recent cradle-to-gate LCAs confirming meaningful GWP reductions at the process scale [18,35]. Taken together, these findings highlight perovskite and tandem platforms as strong candidates for high-efficiency, lower-temperature manufacturing with clear pathways toward flexible and integrated PV products [1,14,21,22,36].

3.2. Unit Operations and Equipment Classes

Process-level environmental burdens in next-generation lines are concentrated in a few unit operations: substrate preparation/clean, solution or vacuum deposition, drying/annealing, contact formation, and encapsulation/lamination [8,12,37]. For solution-processed routes, slot-die and related scalable coaters enable continuous wet deposition with solvent choice and recovery as first-order levers; comparative studies show that switching to greener solvents and continuous coating cuts electricity and solvent burdens versus batch doctor blading [15,16]. ALD and related vacuum steps remain essential for barrier layers and transparent conductive oxides (TCOs); recent assessments quantify their strong non-linear sensitivity to throughput and grid carbon intensity and outline mitigation pathways at the recipe and tool levels [17]. Thermal/photonic annealing frequently dominates step-level energy, reinforcing the need for furnace/oven ecodesign and heat-management strategies in industrial lines [12,38]. For device reliability and circularity, encapsulation choices (glass–glass vs. polymer–metal foils, barrier performance, curing routes) can shift cradle-to-gate GWP by double-digit percentages and determine feasible EoL pathways [12,39,40]. Complementary developments in thermal management and peripheral equipment (e.g., heat sinks, power electronics) round out the equipment landscape and hotspot profile for process-resolved LCIs [40,41]. Harmonizing these unit operations into wet/thermal/vacuum/ambient classes provides a consistent scaffold for cross-technology LCI reporting and later scale-up modeling [8,12,17].

Despite this emerging convergence on unit-operation classes, cross-study comparisons already reveal important contradictions and gaps. Several LCAs of perovskite and thin-film lines reach opposite conclusions on whether solution or vacuum routes have lower embodied impacts because they adopt different functional units (1 m² versus 1 kWp versus kWh-delivered) and treat active/total area and lifetime inconsistently [3,8,12,26]. Likewise, some studies allocate a substantial share of energy use to encapsulation and substrate preparation, whereas others exclude or aggregate these steps into background data, obscuring the relative importance of wet versus vacuum equipment [12,32,40]. A further gap is the limited availability of transparent vacuum-tool LCIs (pump-down, idle power, and chamber-conditioning profiles), which hinders reconciliation of reported hotspot patterns across ALD, sputtering, and evaporation tools [17,37,38]. These issues motivate the more formal harmonization and methodological guidance developed in Section 4 and Section 7.

3.3. Typical Line Configurations

Manufacturing architectures cluster into batch, continuous R2R, and hybrid flows. Batch toolsets (common in wafer-based lines) offer tight control but limited throughput scalability; in contrast, R2R integrates sequential coating, drying, and barrier formation on moving webs to maximize material utilization and line uptime [15,16]. Hybrid lines that combine continuous solution coating of functional stacks with batch vacuum finishing or encapsulation are increasingly used on pilot and pre-industrial perovskite/OPV lines and have demonstrated competitive energy use and yield trajectories [42,43]. System-level integrations (e.g., PV/thermal (PV/T) hybrids) highlight additional layout choices and thermal constraints that influence both product design and line configuration [44]. As Figure 2 illustrates, the choice of batch vs. R2R vs. hybrid determines the frequency and intensity of energy-intensive steps (dry/anneal, vacuum modules) and therefore the dominant contributions to cradle-to-gate GWP and throughput-dependent learning.

3.4. Cross-Study Contradictions and Gaps

Comparative LCAs of crystalline-silicon, thin-film, and perovskite PVs already exhibit systematic contradictions that can be traced to methodological choices rather than intrinsic technology differences. A first source of disagreement is the FU. Studies that report impacts per 1 m² of total module area, without distinguishing between active and inactive regions or normalizing to delivered kWh, often rank low-efficiency or heavily encapsulated designs unfavorably; when the same inventories are expressed per kWp or per kWh-delivered, and lifetime and active-area ratios are applied consistently, the relative ordering between thin films and c-Si can invert by up to an order of magnitude [3,8,12,26]. As a result, some LCAs conclude that thin films or perovskites have systematically lower embodied GWP than mono-Si, whereas others find the opposite once cell efficiency, active-area fraction, and service life are aligned.

A second source of conflict lies in system boundary definitions. Several studies adopt cradle-to-gate boundaries focused on cell or module fabrication and treat balance-of-system (BOS), inverter replacement, and EoL implicitly or via generic database modules [2,24]. Others implement cradle-to-grave or grave-to-cradle models with explicit demanufacturing and recycling steps [30,32]. Inconsistent inclusion or aggregation of encapsulation, glass mass, module frames and BOS components can therefore shift the apparent dominance of manufacturing versus system-level contributions and complicate technology rankings across c-Si, CdTe, CIGS, and perovskite/tandem architectures [2,32].

Third, electricity mix assumptions cause large divergences. Identical processes evaluated under coal-intensive Chinese grids versus hydropower- or nuclear-dominated European or North American mixes show 2–4× swings in cradle-to-gate GWP [2,24]. In some of these regional comparisons, thin films retain an advantage over Si across all grid scenarios, while in others hydropower-linked Si supply chains outperform CdTe or CIGS once upstream ingot/wafer routes are decarbonized [2]. Without explicit reporting of regional electricity, year, and decarbonization assumptions, such context-dependent reversals can be misinterpreted as intrinsic technology properties.

Finally, there are clear data gaps. Transparent, process-resolved LCIs for vacuum tools (especially multi-chamber ALD and high-rate sputtering), solvent management, and encapsulation remain scarce, particularly for emerging perovskite and tandem lines [12,17,37,38]. EoL inventories for these stacks are also fragmented, with heterogeneous allocation rules and incomplete disclosure of foreground versus background flows [30]. These contradictions and gaps motivate the normalization rules (Section 4.1), data-quality tiers (Section 4.3), and reporting checklist (Section 7), which are designed to make cross-study inconsistencies visible and to enable more robust comparative analysis.

4. Process-Level LCI Compilation and Harmonization

Unlike prior PV LCA meta-analyses that primarily compare reported impact indicators, this section synthesizes and harmonizes process-level life-cycle inventories across emerging photovoltaic manufacturing routes. The analysis explicitly links reported inventories to manufacturing configuration, scale, and data quality, enabling comparison across solution-based, vacuum-based, and hybrid process families while preserving transparency regarding uncertainty and technology readiness.

4.1. Normalization Rules (FU, Boundaries, Electricity)

Harmonized LCIs for next-generation PV processes require normalization of FUs, system boundaries, and energy inputs to ensure comparability across studies and technology readiness levels. Most cradle-to-gate studies adopt 1 m² of module area or 1 kWp of rated capacity as the FU, yet results can diverge by more than an order of magnitude if yield, lifetime, or active-area ratios are not consistently applied [3,8,12,26,45]. A transparent reference FU anchored to 1 m² of active layer and expressed in MJ primary/kWp or g CO₂-eq/Wp allows consistent scaling between lab-scale device data and industrial throughput scenarios [46,47,48].

Boundary harmonization similarly requires clear distinction between foreground process modeling (substrate, deposition, annealing, encapsulation) and background datasets (upstream electricity, gases, solvents). International Energy Agency (IEA) guidance [3] and recent prospective frameworks [23,24,26] recommend aligning cut-offs and material inclusions with ISO 14044 [49] while explicitly reporting excluded utilities (vacuum pumps, clean-room HVAC). Grid electricity normalization remains a dominant factor; its contribution to total GWP varies from <20% in low-carbon grids to >60% in coal-intensive systems [2,24,50]. Recent process-level LCAs show that electricity carbon intensity, tool idle times, and heating method (furnace vs. photonic) can explain 70–80% of cross-study variability in impact results [17,51].

When isolating electricity as a single parameter, holding all other process and material assumptions constant, the share of total cradle-to-gate GWP attributable to electricity varies substantially across technologies and grid contexts. Across the LCAs synthesized here, electricity accounts for approximately 35–65% of total cradle-to-gate GWP for c-Si modules, 30–55% for CIGS and CdTe thin films, and 45–75% for perovskite (single-junction and tandem) and organic solar cell (OSC) routes, depending on manufacturing configuration and electricity mix. Varying only the grid carbon intensity between a coal-dominated system (≈800 g CO₂ kWh⁻¹) and a low-carbon system (≈50–100 g CO₂ kWh⁻¹) results in ~50–70% variation in total GWP for c-Si, ~40–60% for CIGS/CdTe, and ~60–80% for perovskite and OSC technologies, confirming electricity as the dominant single driver of cross-study divergence.

To facilitate meta-analysis, all process inventories in this work are expressed using a unified reference of EU-27 grid (2023) [1] for baseline scenarios and a normalized FU of 1 m² active absorber equivalent, unless otherwise specified.

Across the studies harmonized in this review, inconsistent application of these three elements (FU specification, system boundaries, and electricity-mix assumptions) emerges as the dominant source of disagreement between LCAs. For example, perovskite and CdTe modules that appear clearly favorable to mono-Si under a coal-intensive grid and a 1 m² FU can lose this apparent advantage when evaluated per kWh-delivered under low-carbon electricity and cradle-to-grave boundaries that include BOS and EoL [2,3,8,24,26,48]. The normalization rules adopted here are therefore explicitly designed to make such methodological drivers of divergence transparent.

To avoid bias when combining process-resolved LCIs with aggregated background databases (e.g., Ecoinvent), this review applies a strict foreground/background separation. Foreground inventories are harmonized only for unit operations explicitly modeled at the process level (substrate preparation, deposition/coating, drying/annealing, metallization, encapsulation), while background flows (electricity supply, industrial gases, commodity chemicals, generic materials, and transport) are consistently mapped to named datasets through the

Table A4. BACKGROUND_LINKS—Dataset mapping.

Dataset_Name	Database	Region	Year	Dataset_ID	Transport	Comment
Electricity, medium voltage, EU-27, 2023	Ecoinvent	EU-27	2023	elec_mv_eu27_2023	0 km	Used for all electricity entries
Argon, at plant	Ecoinvent	GLO	2020	argon_at_plant	500 km truck	Sputtering purge gas
Al2O3 ALD precursor, at plant	Ecoinvent	GLO	2020	ald_alumina_precursor	300 km truck	Placeholder; adjust to actual
Glass, soda-lime, 3.2 mm, at plant	Ecoinvent	RER	2020	glass_32mm	200 km truck	Front/back glass
Encapsulant, POE/EVA, at plant	Ecoinvent	GLO	2020	encapsulant_poe_eva	300 km truck	POE/EVA film
Silver paste, at plant	Ecoinvent	GLO	2020	silver_paste_lowAg	1000 km truck	Low-Ag paste
Detergent, industrial, at plant	Ecoinvent	GLO	2020	detergent_industrial	200 km truck	Substrate cleaning
DI water, at plant	Ecoinvent	RER	2020	di_water	Onsite	Closed-loop recirculation

(dataset identity, database, region, year, and version).

A concrete example of potential bias arises when generic aggregated glass datasets are used in place of PV-specific glass supply. Using an aggregated dataset such as glass, flat, generic construction can misrepresent both process energy and logistics relative to solar-grade low-iron glass. In particular, generic glass datasets may embed average furnace efficiencies, cullet shares, and transport assumptions that differ from dedicated PV-glass production and logistics. When applied to glass–glass module configurations, this substitution can bias the apparent GWP contribution of glass upward by approximately 10–20%, potentially masking real gains from improved furnace design and higher recycled cullet share observed in PV-oriented supply chains.

4.2. LCIs by Process Family

Process LCIs were grouped into solution-based, vacuum-based, and hybrid thin-film families according to dominant energy and material drivers [8,12,17,32,42,52]. For each, per-step data (solvent mass, energy use, emission factors) were harmonized following the functional-unit and boundary normalization rules given in Section 4.1. Solution processes (slot-die, blade, inkjet): Step LCIs are dominated by solvent use, drying/annealing, and substrate handling. Data from [12,15,16,42,48] indicate embodied energy from 200–400 MJ m⁻², with solvent recovery efficiency as the main uncertainty lever. Vacuum processes (ALD, sputtering, evaporation): Process LCIs from references [17,37,38,53] show strong throughput dependence. When normalized per m², chamber heating and pumping contribute ≈60% of process energy. Advanced pulsed-ALD and low-temperature sputter recipes can halve GWP if throughput is doubled [17,52,54].

Hybrid tandem and encapsulation flows: Integrated lines combining wet and dry steps show intermediate energy intensities (250–350 MJ m⁻²) and greater sensitivity to encapsulation and substrate mass [12,32,40,42,55]. For all process families, common equipment utilities, air handling, vacuum pumps, chillers, and dryers, were separately inventoried using scaling factors derived from [27,37,39,53]. Harmonized process tables (Figure 3 and Figure 4) present the mean, standard deviation, and range of energy and mass flows per step. Within each process family and data-quality tier, reported values represent simple arithmetic means of the underlying literature sources rather than a single pooled global average. Importantly, no weighting by production volume or TRL is applied at this stage; instead, Tier 1–3 stratification is used to preserve transparency between industrial/pilot measurements, scaled laboratory inventories, and modeled literature estimates. This tier-separated reporting avoids obscuring systematic efficiency differences between laboratory and industrial data while allowing direct comparison of ranges across process families.

Table 1. Data-quality classification and uncertainty ranges for process-level photovoltaic LCIs. Tiers 1–3 correspond respectively to measured industrial/pilot data, scaled laboratory inventories, and modeled or averaged literature data. Reported values represent tier-separated arithmetic means and ranges, not TRL-weighted aggregates. Energy intensity values (MJ m⁻²) are normalized following Section 4.1. Coefficients of variation (CoVs) reflect propagated Monte Carlo uncertainty following [23,24,26,48,56,57].

Process Family	Data-Quality Tier 1 (Industrial/Pilot)	Tier 2 (Lab → Pilot Scaled)	Tier 3 (Literature/Model)	Dominant Uncertainty Drivers	Typical CoV (%)	Representative Data Sources
Solution-processed (slot-die, blade, inkjet)	Measured energy ≈ 200–250 MJ m−2	250–400 MJ m−2	300–500 MJ m−2	Solvent recovery η, annealing temperature, substrate mass	30–60%	[12,15,16,42,45,52]
Vacuum-processed (ALD, sputter, evaporation)	150–200 MJ m−2	200–350 MJ m−2	250–400 MJ m−2	Chamber heating, pump idle power, throughput	20–50%	[17,37,38,53,54]
Hybrid thin-film/tandem	220–280 MJ m−2	300–380 MJ m−2	350–450 MJ m−2	Encapsulation mass, yield, interface recombination losses	25–45%	[12,32,40,42,55]
Encapsulation/lamination	40–70 MJ m−2	60–100 MJ m−2	80–120 MJ m−2	Curing route, barrier foil, curing temperature	15–30%	[12,38,39,44]
Ancillary/utilities (HVAC, vacuum, power electronics)	10–20% of total energy	15–25%	20–30%	Grid mix, duty cycle, maintenance schedule	10–25%	[26,37,39,54,56,57]

Table 1. Data-quality classification and uncertainty ranges for process-level photovoltaic LCIs. Tiers 1–3 correspond respectively to measured industrial/pilot data, scaled laboratory inventories, and modeled or averaged literature data. Reported values represent tier-separated arithmetic means and ranges, not TRL-weighted aggregates. Energy intensity values (MJ m⁻²) are normalized following Section 4.1. Coefficients of variation (CoVs) reflect propagated Monte Carlo uncertainty following [23,24,26,48,56,57].

Process Family	Data-Quality Tier 1 (Industrial/Pilot)	Tier 2 (Lab → Pilot Scaled)	Tier 3 (Literature/Model)	Dominant Uncertainty Drivers	Typical CoV (%)	Representative Data Sources
Solution-processed (slot-die, blade, inkjet)	Measured energy ≈ 200–250 MJ m−2	250–400 MJ m−2	300–500 MJ m−2	Solvent recovery η, annealing temperature, substrate mass	30–60%	[12,15,16,42,45,52]
Vacuum-processed (ALD, sputter, evaporation)	150–200 MJ m−2	200–350 MJ m−2	250–400 MJ m−2	Chamber heating, pump idle power, throughput	20–50%	[17,37,38,53,54]
Hybrid thin-film/tandem	220–280 MJ m−2	300–380 MJ m−2	350–450 MJ m−2	Encapsulation mass, yield, interface recombination losses	25–45%	[12,32,40,42,55]
Encapsulation/lamination	40–70 MJ m−2	60–100 MJ m−2	80–120 MJ m−2	Curing route, barrier foil, curing temperature	15–30%	[12,38,39,44]
Ancillary/utilities (HVAC, vacuum, power electronics)	10–20% of total energy	15–25%	20–30%	Grid mix, duty cycle, maintenance schedule	10–25%	[26,37,39,54,56,57]

Within each process family, the observed spread in reported LCIs reflects differences in operational assumptions and manufacturing configurations rather than intrinsic technological distinctions. For solution-based routes, intra-family variability is primarily driven by solvent selection, solvent-recovery efficiency, coating speed, and drying or annealing duty cycles. In vacuum-based processes, differences in chamber utilization, idle power, pump-down requirements, heating profiles, and throughput assumptions account for much of the variation between reported inventories. Hybrid and tandem routes exhibit additional sensitivity to encapsulation mass, substrate thickness, and interface-processing steps, which can shift cradle-to-gate impacts even when nominally similar process families are compared. Consequently, variability within a process family should be interpreted as scale- and configuration-dependent rather than as a fixed property of a given PV technology.

These datasets span perovskite, silicon (Si), OSC, and BOS inventories across regional and multi-technology contexts [52,53,54,55,56,57,58,59].

For transparency, the evidence base supporting each process-family synthesis is reported. Across the final set of 102 included studies, the harmonized process-family comparisons in Figure 3 and Figure 4 are underpinned by n = 35 solution-based LCIs (TRL 2–5), n = 28 vacuum-based LCIs (TRL 4–8), and n = 22 hybrid/tandem LCIs (TRL 3–6); encapsulation-specific inventories contribute n = 17 studies (TRL 5–8). These counts refer to studies providing explicit process-level inventory data after applying the normalization rules in Section 4.1 and the Tier 1–3 classification in Section 4.3.

4.3. Data-Quality Tiers and Uncertainty Characterization

To capture data representativeness across technology readiness levels, we introduce a three-tier data-quality scheme:

Tier 1: Primary industrial data or verified pilot-line measurements.

Tier 2: Laboratory or pre-industrial LCIs adjusted via throughput and yield scaling.

Tier 3: Literature averages or modeled estimates without direct measurement.

The data-quality tier framework clarifies that much of the apparent variability within individual process families arises from differences in technology readiness level, scale, and data representativeness rather than from fundamental process inefficiencies. Laboratory-scale inventories (Tier 2–3) typically exhibit higher energy intensity due to low throughput, unstable uptime, conservative operating assumptions, and limited heat or solvent recovery. In contrast, industrial and pilot-line data (Tier 1) capture stabilized duty cycles, improved tool loading, and optimized thermal and solvent-management systems. Distinguishing these scale- and TRL-related effects is essential to avoid misinterpreting intra-family variability as intrinsic technological divergence.

Following [23,24,26,48,56,60,61], parameter uncertainty for each process was represented by lognormal distributions fitted to reported min–max values. Monte Carlo propagation (10⁴ samples) quantified uncertainty in aggregated cradle-to-gate GWP, showing coefficients of variation of 20–60% for lab-scale data and <15% for industrial references.

To assess whether aggregation choices influence the central tendencies reported in Section 4.2, a TRL-weighted sensitivity analysis was conducted in parallel to the baseline, tier-separated aggregation. In this test, Tier 1 (industrial/pilot) data were assigned double weight relative to Tier 2 and triple weight relative to Tier 3 when aggregating energy intensities within each process family (solution, vacuum, hybrid). This weighting reflects the higher representativeness and efficiency of mature production lines compared with laboratory-scale studies. The resulting TRL-weighted means were 10–20% lower than the unweighted values across all process families but did not alter the relative ranking of solution-, hybrid-, and vacuum-dominated routes nor the identification of dominant hotspots (annealing/drying, vacuum pumping, and encapsulation). This indicates that the main conclusions of the review are robust to reasonable TRL-based weighting assumptions.

A cross-cutting finding from [17,24,26,48,56,57] is that energy intensity and solvent recovery efficiency dominate overall variance, while material yield and uptime become secondary at the industrial scale. To enable transparent meta-analysis, all LCIs in this work are annotated with data-quality flags (Tier 1–3) and uncertainty ranges (95% CI) following the pedigree-matrix structure in Table 1.

5. Scale-Up Dynamics and Sensitivity Analysis

5.1. Scaling from Laboratory to Pilot and Industrial Production

Transitions from laboratory fabrication to pilot and industrial production introduce scale-dependent changes in energy intensity, solvent consumption, throughput, and yield, all of which strongly influence cradle-to-gate impacts. Systematic scale-up reviews highlight that deposition uniformity, tool idle power, and stabilization of continuous-line operation determine whether environmental improvements follow technological learning curves or remain constrained by equipment limitations [23,26]. Pre-industrial perovskite module lines show that drying/annealing, substrate preparation, and encapsulation dominate laboratory footprints but decline markedly with increasing line speed, coating width, and stabilized uptime at the pilot scale [12]. Perovskite fabrication studies consistently confirm that burdens shift from annealing and solvent losses toward utilities, substrate handling, and equipment stabilization as scale increases [15,53].

Empirical learning-curve studies for silicon, thin-film, polysilicon feedstock, and metallization demonstrate persistent 15–25% reductions in energy and material intensity per cumulative doubling of production capacity, providing a quantitative basis for linking laboratory inventories to pilot and industrial scenarios [4,6,25]. Prospective LCA frameworks formalize this relationship by embedding learning rates, technology readiness level (TRL) progression, and throughput normalization into scenario-based modeling rather than fitting a single universal curve [23,24,26]. This approach reflects the fact that learning effects are process-specific and depend on equipment configuration, duty cycle, and yield stabilization rather than cumulative capacity alone.

Within the LCIs synthesized in this review, laboratory-to-pilot transitions typically correspond to effective throughput increases of one to two orders of magnitude, while pilot-to-industrial transitions introduce further reductions through higher tool loading, reduced idle time, and stabilized solvent and thermal management. When these changes are mapped to reported learning-rate ranges, they imply order-of-magnitude reductions in per-unit energy and GWP intensity consistent with learning-curve expectations, without requiring explicit extrapolation beyond observed operating regimes.

Pilot-to-industrial transitions also introduce new system-level interactions: thermal-management components such as heat sinks and phase-change materials, inverter and power-electronics integration, and utilities for heating, ventilation, and air conditioning (HVAC) and cooling begin to account for 10–25% of total process energy [41]. Scale-up demonstrations of titanium dioxide (TiO₂)-nanorod architectures show co-evolution between design, deposition method, and yield, with multi-fold improvements reported when transitioning from small-area lab devices to stabilized pilot lines [13]. Likewise, hybrid PV/T integrations reveal how thermal layout, substrate design, and throughput influence uptime and material efficiency during scale-up [44].

Process-resolved prospective LCAs confirm that industrial routes can reduce cradle-to-gate GWP by 30–70% relative to laboratory baselines once throughput normalization, tool loading, equipment duty cycles, and downtime optimization are incorporated, underscoring that environmental impacts evolve as functions of both scale and process architecture rather than device efficiency alone. Importantly, this reduction range is not universal across all PV technologies or manufacturing steps. The upper end of the range (≈50–70%) is primarily observed for perovskite, OSC, and hybrid solution–vacuum manufacturing lines, as well as for selected thin-film and polysilicon processing steps where strong learning effects and throughput gains have been documented. In contrast, c-Si modules as integrated systems typically lie toward the lower end of this range (≈30–40%), because a substantial share of their environmental footprint arises from upstream material-intensive processes (polysilicon production, ingot growth, wafering) that are already relatively optimized at the industrial scale.

The least scale-up gains are consistently observed for material-dominated steps, such as glass, frames, encapsulant mass, where reductions in operational energy have limited effect, and for certain batch vacuum operations that already operate near optimal loading, where chamber volume and pumping requirements constrain further efficiency improvements. Conversely, annealing/drying and solution-coating steps show the largest relative reductions upon scale-up due to increased line speed, stabilized uptime, and improved duty-cycle utilization.

5.2. Parameterized Scenarios

Scenario-based modeling demonstrates that electricity carbon intensity is the dominant driver of cradle-to-gate GWP for emerging PV manufacturing routes. Carbon-aware optimization frameworks show strong GWP sensitivity to grid mix, duty cycles, and equipment-loading assumptions [62,63], while regional LCAs for perovskite, Si, and OPV systems consistently report 2–4× variation in impacts when identical processes are evaluated under different grid carbon intensities [35,60,64]. These findings are reinforced by studies in building-integrated and storage-integrated PV systems, which highlight the additional influence of BOS configurations and thermal-management pathways [43,65].

When scenario assumptions are harmonized and only the electricity supply is varied, the reviewed studies show that electricity alone typically drives ~40–80% of the total variation in cradle-to-gate GWP, depending on technology and manufacturing route. The highest sensitivity is observed for electricity-dominated, low-temperature processes such as perovskite and OSC manufacturing, whereas wafer-based silicon and thin-film technologies exhibit slightly lower, but still dominant, electricity-driven variability due to the additional contribution of upstream material processing and thermal steps.

Beyond electricity, scenario levers specific to manufacturing architecture remain first-order determinants of environmental intensity. For solution-processed lines, solvent-recovery efficiency, coating method, annealing route, and line speed strongly influence GWP outcomes [53]. Material-substitution scenarios, including carbon-electrode perovskites and reduced-Ag metallization, demonstrate significant GWP reductions relative to Au-based or high-Ag baselines, especially under low-carbon electricity [18,35,55]. Similar sensitivities appear in national-scale LCAs for Chinese PV production, where electricity mix, yield, and material efficiency jointly shape environmental outcomes [57].

Forward projections for south and north Italy and EU-wide electricity systems indicate substantial manufacturing-footprint reductions under 2030–2050 decarbonization trajectories [66,67,68]. Prospective LCA frameworks incorporating learning curves and TRL-based upscaling show that declining grid carbon intensity interacts strongly with yield and throughput improvements, producing non-linear environmental gains [24,69]. Consequential scenario modeling for distributed PV reaches similar conclusions: operational context and regional electricity mix dominate future-footprint estimates [46].

Dynamic LCA-LCC models further reinforce the central role of electricity carbon intensity while capturing cost–impact interactions relevant for industrial decision making and deployment planning [10,56,61]. Prospective assessments of tandem modules extend these insights to multi-junction architectures, revealing combined sensitivities to absorber selection, metallization strategy, and electricity supply [70]. Importantly, these studies demonstrate that learning effects are most robustly represented through scenario-based parameterization of throughput, yield, and grid evolution, rather than through extrapolation of single fitted learning curves.

Collectively, the scenario analyses synthesized here show that grid decarbonization, solvent recovery, yield improvement, and material substitution, particularly silver reduction, remain the most effective and near-term levers for lowering cradle-to-gate impacts across next-generation PV manufacturing routes. This parameterized scenario approach ensures consistency with observed industrial transitions while avoiding overconfident extrapolation beyond available empirical data.

5.3. Comparative Cradle-to-Gate Results Across Routes

Comparative LCAs across crystalline-Si, thin-film, perovskite, tandem, and perovskite-Si hybrid modules consistently show that electricity carbon intensity, manufacturing geography, and module design dominate cross-technology variance. Regional assessments report multi-fold differences (typically 2–4×) in cradle-to-gate GWP for identical module designs due solely to grid mix and upstream supply-chain characteristics, demonstrated in case studies for Burkina Faso, China, and Italy [2,64,67]. Similar patterns appear in national and sub-national assessments, where hydropower- or nuclear-dominated regions achieve substantially lower embodied GHG footprints than coal-based grids [58,67].

To consolidate the numerical ranges discussed across individual studies and to support direct comparison of environmental trade-offs,

Table 2. Comparative environmental indicators across PV technology classes.

PV Technology	Cradle-to-Gate GWP (kg CO2-eq/kWp) 1	EPBT (Years)	Recyclability Maturity	Dominant Critical Materials	Substitution/Recycling Status	Key Sources
c-Si	~400–800	~1.0–2.5	Industrial	Ag, Si, Al	High-yield Ag/Si recovery demonstrated; Ag intensity reduction is an established lever	[3,9,48,56,58]
CdTe	~300–600	~0.5–1.5	Industrial	Cd, Te	Established recycling routes reported in PV LCA/recycling reviews; substitution limited	[3,9]
CIGS	~350–700	~0.7–2.0	Pilot–industrial	In, Ga, Ag	Criticality driven by In/Ga; recycling discussed but less mature than c-Si/CdTe	[3,9]
Perovskite (single-junction)	~150–400	~0.2–0.8	Laboratory–pilot	Pb, Ag, Au	Impacts depend strongly on metallization/contacts; EoL inventories and recovery routes still emerging	[12,19,21]
Perovskite/Si tandem	~250–500	~0.4–1.2	Pilot	Ag, Pb, Si	Prospective LCAs show strong sensitivity to grid mix and material strategies; recycling modeling emerging	[21,22,70]
OSC	~100–300	~0.2–0.6	Laboratory	Ag, specialty polymers	Polymer and encapsulation choices dominate; recyclability largely at early stage	[49]

¹ Values are indicative ranges compiled from PV LCA syntheses and representative studies and remain sensitive to FU definition, system boundaries, electricity mix, efficiency, and lifetime assumptions [3,9,11,48]. EPBT values are particularly location- and irradiation-dependent and should be interpreted as typical ranges reported for representative climates rather than harmonized point estimates [3,9,10,45,48]. Recyclability maturity reflects the degree of industrial demonstration and LCA documentation across PV recycling reviews and representative case studies.

provides a summary comparison of key environmental indicators across major PV technology classes. The table synthesizes indicative ranges of cradle-to-gate GWP, typical EPBT, qualitative recyclability maturity, and dominant critical materials for c-Si, CdTe, CIGS, single-junction perovskite, perovskite/Si tandem, and OSC technologies.

Supply-chain and geopolitical analyses indicate that upstream manufacturing pathways, especially ingot/wafer production for Si and vacuum deposition for thin films, drive large regional carbon-intensity disparities; relocating stages or shifting regional electricity mixes can therefore change technology rankings [7,48]. Multi-regional comparisons also show that encapsulation choice, module architecture, and substrate mass modulate these geographic sensitivities [32,67]. Prospective tandem and perovskite-Si hybrid studies suggest that future grid decarbonization and absorber-stack evolution can shift comparative rankings, with tandems surpassing mono- and multi-Si in several 2030–2050 scenarios [70]. Findings align with updated US inventories: lower-temperature thin films and optimized tandems generally exhibit lower embodied GWP than mono-Si under low-carbon electricity [48]. Complementary national-scale LCAs confirm contexts where hydropower-linked Si chains can outperform thin films, depending on metallization and TCO burdens [2,58].

EoL comparisons show that closed-loop recovery and high-yield recycling can reduce cradle-to-gate GWP by ~40–60% for perovskite/Si tandems and by similar margins for crystalline-Si modules when Ag and Si recovery exceed industrial benchmarks [32,58]. Overall, regional electricity mix, module architecture, and recovery efficiency remain the primary determinants of comparative environmental performance across current and next-generation PV technologies [2,64,67].

Taken together, these comparative results also clarify where LCAs most strongly disagree. When technology assessments are aligned to a common FU (1 kWp or kWh-delivered), harmonized boundaries, and explicitly reported electricity mixes, the relative ordering between mono-Si, thin films, and perovskite/tandem modules becomes much more stable, with most of the residual spread attributable to differences in encapsulation mass, substrate design, and metallization strategy [3,8,12,24,26,48]. In contrast, studies that mix 1 m² and 1 kWp FUs, omit BOS or EoL, or rely on generic aggregated database modules for some technologies but process-level inventories for others often report conflicting technology rankings that cannot be reconciled on physical grounds alone. Highlighting these methodological conflicts is central to interpreting the literature and underpins the reporting rules proposed in Section 7.

6. Recycling, Circularity, and Material Criticality

6.1. EoL Pathways, Recycling, and Logistics

The environmental profile of PV manufacturing extends beyond the production phase into EoL systems. Studies analyzing panel recovery show that demanufacturing, material separation, and logistics significantly influence cradle-to-grave impacts. For instance, recovery processes for c-Si modules show major GHG reductions when silver and silicon are recycled at high rates [34]. Research by Ansanelli et al. [28] evaluates detailed mechanical and chemical recovery pathways for polymer, glass, and metal fractions and shows that treatment-stage energy and chemical use dominate impacts when waste volumes are low, with transport contributing only marginally [28]. Celik et al. [71] demonstrate through logistics modeling that EoL transport distance, consolidation strategy, and fragmentation steps influence energy use and emissions across recovery systems.

Closed-loop recycling scenarios for crystalline-Si modules, evaluated through full grave-to-cradle modeling, show that loop closure is feasible when high recovery efficiencies for Si, glass, Ag, and Al are achieved, with performance strongly dependent on collection volumes and quality of modules entering the recycling stream [72]. Laboratory-scale studies of polymer pyrolysis and delamination processes show how thermal treatment conditions and reactor design influence energy demand during polymer-layer removal, providing insight into how advanced separation steps may integrate into future EoL pathways [73]. Prospective assessments using advanced physical separation technologies indicate that future recycling systems could obtain higher recovery efficiencies and lower impacts as throughput increases and processes become more integrated [30]. Overall, effective EoL system design, including logistics, high-efficiency separation technologies, and high-yield recovery pathways, is pivotal for achieving circular PV supply chains.

From an economic perspective, industrial-scale recycling of c-Si modules is increasingly supported by high-yield recovery of silver. Recent LCA-LCC studies and industrial trials indicate that Ag recovery rates ≥ 90% are economically favorable at current silver market values, provided that recycling is performed in high-throughput, centralized facilities with sufficient waste volumes [29,34,54,72]. In contrast, silicon recovery is often constrained less by market value than by the energy intensity of purification steps; while technical feasibility has been demonstrated, economic viability remains strongly dependent on regional electricity prices and labor costs [31,34,54].

Across the reviewed EoL pathways, reported cradle-to-grave GHG reductions of approximately 40–60% are consistently achieved only under high-recovery recycling scenarios. Studies that model closed-loop or high-value recycling indicate that such reductions require silver recovery rates typically ≥90%, silicon recovery ≥ 80–90%, and glass recovery ≥ 85%, combined with sufficient collection volumes to avoid disproportionate energy and chemical burdens during treatment. When recovery efficiencies fall below these levels, recycling benefits diminish rapidly and may be partially offset by demanufacturing energy demand and material losses. These findings indicate that the climate benefits of PV circularity are threshold-dependent, motivating the quantitative recovery-rate ranges summarized in Section 6.2 and Table 3.

6.2. Allocation, Credits, and EoL Accounting

Allocation of recycled flows, crediting of recovered materials, and boundary settings strongly shape the environmental outcomes of PV EoL modeling. Ali et al. [74] show that adopting avoided-production credits for recovered materials (e.g., Si, Al) can substantially improve the performance of distributed PV systems when displacement factors are realistically defined. Allocation choices within mechanical–chemical recovery of c-Si modules, particularly for glass and metal fractions, can alter GWP results because different output streams inherit different portions of the energy and chemical burdens. Closed-loop modeling confirms that material-reuse credits, particularly for Si and Al, can significantly reduce cradle-to-module impacts when loop-closure efficiency and sorting quality are sufficiently high [72]. Nevala et al. [75] show that allocation in physical-separation systems (e.g., liberation, optical sorting) determines how burdens are distributed between recyclable fractions, especially when comparing mechanical versus more chemically intensive processes. For perovskite modules, hydrometallurgical recycling studies show that credits for recovered Pb, Au, or Ag depend strongly on assumed substitution ratios and on whether the recovered material meets virgin-grade quality specifications [76].

In this context, the magnitude of reported cradle-to-grave GHG reductions is strongly conditional on achieving sufficiently high material-recovery efficiencies and on applying avoided-production credits under system-expansion assumptions. Across the EoL LCAs reviewed, reductions on the order of 40–60% relative to no-recycling baselines are only observed when recovered materials, most notably Si, Ag, glass, and Al, reach high-yield thresholds and are assumed to substitute virgin materials of comparable quality [31,32,54]. At lower recovery efficiencies, or when downcycling dominates, the environmental benefits of recycling diminish rapidly because treatment-stage energy use, chemical consumption, and logistics burdens outweigh avoided-production credits. This threshold behavior motivates the indicative recovery-rate bands summarized in Table 3 and explains why reported circularity benefits vary widely across PV technologies and EoL studies [54,72].

Collectively, these studies underline that transparent allocation rules (mass, economic, system-expansion), consistent substitution assumptions, and clearly defined credits are essential for interpreting PV EoL LCAs across technologies. Despite this progress, several unresolved issues continue to limit the comparability of PV EoL LCAs:

• Heterogeneous allocation choices (mass, economic, energy content) are often applied without sensitivity analysis, leading to sizeable shifts in reported benefits for glass, metals, and polymer fractions [72,74,75].
• Mixing aggregated and process-level inventories (e.g., generic glass/polymer treatment alongside detailed metallurgical steps) can mask hotspots and bias technology comparisons [30,72].
• Quality and substitution potential of recovered materials (Si, Ag, Pb, specialty polymers) are not always reported, making avoided-production credits difficult to validate [72,75,76].

For perovskite and tandem modules, published EoL inventories remain sparse and largely laboratory-scale, with incomplete reporting of chemical use, energy demand, and recovery efficiencies for Pb, Ag, and TCO elements [30,76].

These findings underline that economic performance of PV recycling is highly sensitive to scale, process integration, and energy costs, reinforcing the need to assess recycling pathways jointly through environmental and economic lenses rather than on recovery efficiency alone [30,54,72].

Table 3. Indicative recovery-rate thresholds required to achieve ~40–60% cradle-to-grave GHG reduction through PV module recycling.

Module Type	Material	Recovery Efficiency Threshold (%) 1	Recycling Pathway Assumption	Indicative GHG Reduction Achieved	Key Sources
Crystalline Si (c-Si)	Silicon (Si)	≥80–90	Closed-loop to solar-grade or high-value Si applications	~45–60%	[27,31,32,34,72]
	Silver (Ag)	≥90–95	Hydrometallurgical recovery with avoided primary Ag production		[29,34,54,72]
	Glass	≥85–90	Closed-loop or high-grade flat-glass recycling		[27,28,30]
	Aluminum frame	≥90	Conventional secondary Al recycling		[27,28]
Perovskite/Si tandem	Silicon (Si)	≥80	Closed-loop or high-value reuse	~40–55%	[32,70]
	Silver (Ag)	≥90	Hydrometallurgical recovery		[76]
	Lead (Pb)	≥85–90	Controlled hydrometallurgical recovery		[76]
	Glass	≥85–90	High-grade glass recycling		[30,32]

¹ Recovery thresholds represent indicative bands aggregated from multiple end-of-life LCAs and assume system-expansion or avoided-production credits. Achieved GHG reductions depend on recycling scale, electricity mix, material quality of recovered outputs, and allocation choices.

Table 3. Indicative recovery-rate thresholds required to achieve ~40–60% cradle-to-grave GHG reduction through PV module recycling.

Module Type	Material	Recovery Efficiency Threshold (%) 1	Recycling Pathway Assumption	Indicative GHG Reduction Achieved	Key Sources
Crystalline Si (c-Si)	Silicon (Si)	≥80–90	Closed-loop to solar-grade or high-value Si applications	~45–60%	[27,31,32,34,72]
	Silver (Ag)	≥90–95	Hydrometallurgical recovery with avoided primary Ag production		[29,34,54,72]
	Glass	≥85–90	Closed-loop or high-grade flat-glass recycling		[27,28,30]
	Aluminum frame	≥90	Conventional secondary Al recycling		[27,28]
Perovskite/Si tandem	Silicon (Si)	≥80	Closed-loop or high-value reuse	~40–55%	[32,70]
	Silver (Ag)	≥90	Hydrometallurgical recovery		[76]
	Lead (Pb)	≥85–90	Controlled hydrometallurgical recovery		[76]
	Glass	≥85–90	High-grade glass recycling		[30,32]

¹ Recovery thresholds represent indicative bands aggregated from multiple end-of-life LCAs and assume system-expansion or avoided-production credits. Achieved GHG reductions depend on recycling scale, electricity mix, material quality of recovered outputs, and allocation choices.

6.3. Critical Materials, Scarcity, and Substitution

Material criticality remains a central concern for long-term PV sustainability, especially for elements such as Ag, In, Sn, Pb, and Au. High-efficiency recovery of Ag and Si from c-Si modules has been shown to substantially reduce resource-scarcity indicators, underscoring the importance of metal-recovery yields in circular strategies [35]. Global, technology-spanning assessments of critical-material risks identify distinct vulnerability profiles for Si, thin-film, and tandem architectures, emphasizing that recycling and substitution strategies must be technology-specific [77]. Analyses of silver-use trajectories under net-zero scenarios show that reductions in Ag intensity, combined with robust recycling loops, are necessary to avoid future supply constraints [5]. Comparisons of upcycling and downcycling end-of-life pathways indicate that upcycling, where recovered materials displace high-value inputs, significantly improves material-criticality metrics relative to downcycling into low-grade applications [78].

Substitution pathways provide an additional lever for mitigating critical-material dependence. Evaluations of biobased and fluorine-free encapsulants demonstrate reduced toxicity, lower reliance on fluoropolymers, and improved circularity potential for next-generation PV modules [79,80]. Prospective assessments of tandem supply chains further indicate that critical-material dependence can intensify if substitution and recycling strategies are not integrated early in technology development, highlighting the need for coordinated material-management planning across manufacturing and end-of-life stages. Figure 5 summarizes the critical-material risk landscape for next-generation PV technologies, integrating supply-risk indices, environmental relevance, recycling-efficiency trajectories, and viable substitution pathways for Ag, In/Sn, Pb, and Au.

However, substitution of silver with alternative conductors such as copper or aluminum introduces important performance, reliability, and manufacturability considerations that must be evaluated alongside LCA results. While Ag → Cu/Al substitution offers clear benefits in terms of material criticality and environmental impact, multiple studies highlight potential trade-offs [5,19,32,70,76]:

Electrical performance risks: Cu and Al generally exhibit higher contact resistivity than Ag, particularly at fine finger widths, which can increase series resistance and reduce fill factor if grid geometry and metallization design are not carefully optimized. These effects are most pronounced in high-efficiency devices employing narrow-line metallization.

Reliability and degradation: Cu is susceptible to corrosion and electromigration under damp-heat and bias-stress conditions, necessitating robust diffusion barriers and controlled plating schemes. Al-based pastes can present adhesion and solderability challenges on certain TCOs, with implications for long-term contact stability and module reliability.

Industrial feasibility: Partial Ag-reduction strategies, including narrower fingers, multi-busbar layouts, hybrid Cu/Ag grids, and low-Ag paste formulations, are already implemented in industrial c-Si and emerging thin-film production lines. These approaches typically reduce Ag intensity from approximately 80–100 mg/W to 20–40 mg/W with negligible performance penalties and without major modifications to existing manufacturing infrastructure.

Limits of full substitution: Full replacement of Ag by Cu or Al is technologically promising but not yet a universal drop-in for existing industrial lines. It often requires additional process steps (e.g., plating or diffusion-barrier deposition), modified firing or curing profiles, and extended reliability qualification (e.g., damp-heat and thermal-cycling testing), which can partially offset manufacturing and environmental gains.

For perovskite and tandem modules, economic feasibility of metal recovery is more conditional. Hydrometallurgical recovery of Pb and Ag can be economically viable only when high recovery efficiencies (>85–90%) are achieved and when recycling processes are coupled to sufficiently large and stable waste streams [76]. At lower collection volumes or reduced recovery efficiencies, chemical consumption, energy demand, and treatment costs dominate, resulting in unfavorable cost–revenue balances despite technical feasibility [76]. On the basis of reported cost–revenue analyses, high-yield Pb/Ag recovery appears technically mature and economically viable for industrial-scale recycling streams but is unlikely to be profitable at very low collection volumes or if recovery efficiencies fall below approximately 80–85% [76].

Overall, mitigating critical-material risks in next-generation PV manufacturing requires a combined strategy of high-yield recycling, credible and performance-aware substitution pathways, and tight material-loop closure. Reduced-Ag metallization emerges as a near-term, low-risk option for lowering both environmental impacts and material criticality, whereas full Ag substitution should be assessed jointly with efficiency, reliability, and industrial-line compatibility within prospective LCA frameworks.

7. Methodological Guidance for Transparent Process-Level LCAs

To enable transparent, comparable, and reproducible process-resolved LCAs of PV manufacturing, a minimum set of core methodological and inventory elements are reported. These factors directly influence cradle-to-gate impacts and determine the robustness of scale-up assessments across solution-processed, vacuum-processed, hybrid thin-film, tandem, and crystalline-silicon technologies [3,8,12,24,26,48]. Recent supply-chain assessments further show that manufacturing geography and upstream routing can materially shift embodied emissions, making explicit disclosure of electricity sources and sourcing locations essential [81,82]. Agile LCA approaches in the built environment underline that inconsistent system boundaries and missing energy-flow metadata limit comparability and hinder integration with ecodesign and building-scale modeling [83], while façade-integrated PV decision tools demonstrate the need to co-report functional-unit assumptions, area-to-power conversions, and BOS dependencies when coupling LCA with cost/performance metrics [84].

For emerging PVs (e.g., OSC and perovskites), framework studies emphasize that inconsistent boundaries and incomplete process metadata undermine cross-technology comparisons, reinforcing strict reporting rules [49]. Multi-objective PV ecodesign also shows that optimization outcomes are highly sensitive to transparent background LCIs, energy flows, and declared sensitivity parameters [85]; regional EPBT/LCA case studies further highlight that electricity-mix metadata must be disclosed explicitly [86], and hydropower-linked silicon chains illustrate how supply-chain electricity choices affect foreground emissions and must be documented with spatial/temporal precision [87].

Because current PV LCA studies differ widely in their treatment of FUs, system boundaries, equipment-level energy accounting, solvent and gas flows, and uncertainty characterization, a structured reporting framework is essential.

Table 4. Required reporting requirements for transparent, reproducible, and comparable process-level PV LCAs. A worked example demonstrating the numerical consequences of altering FU and system boundary definitions using these reporting rules is provided in Table 5 and Appendix A.

Category	Required Reporting Items	Why It Matters	Applies to	Sources
Functional unit (FU)	1 m2 active area and 1 kWp; module efficiency, active/total area ratio; yield to FU	Misaligned FUs can shift results by >×10; active-area normalization enables lab → pilot translation	All	[3,8,12,48]
System boundary	Foreground steps (substrate prep, deposition/coating, drying/anneal, metallization, encapsulation) vs. background (electricity mix, gases, solvents); explicit exclusions (HVAC, vacuum idle)	Boundary gaps drive hidden variance; explicit foreground/background splits enable harmonization	All	[3,24,26]
Electricity	Grid mix (region, year), carbon intensity, on-site generation; tool idle vs. loaded power; duty cycles	Top driver of GWP; 2–4× swings across regions; idle time is a frequent hotspot	All	[2,24,48,66,67,81,82,86,87]
Background datasets (identity and version)	Database name; dataset identifier; region; year/vintage; version; transport assumptions; justification for the choice of generic or PV-specific proxy datasets	Prevents hidden cut-off differences and double counting when combining background and process-level inventories	All	[3,48]
Process LCIs (per step)	Energy (MJ/m2), gases/solvents (kg/m2), consumables, rejects/yield, tool setpoints (T, t, P)	Enables step-wise attribution and meta-analysis	Solution, vacuum, hybrid lines	[12,17,42,85]
Solvent management	Solvent identity, recovery efficiency (%), losses, abatement	Major driver in wet lines; recovery is a first-order lever	Solution lines	[12,15,16,42]
Vacuum equipment	Pumping scheme, chamber volume, base pressure, cycle times, idle energy	Throughput-sensitive; chamber/pump energy dominates per m2	Vacuum lines	[17,56]
Annealing/drying	Heating method (furnace vs. photonic), setpoints, line speed, thermal losses	Often the single largest energy step; ecodesign is essential	All	[12,38]
Encapsulation and substrates	Glass/foil types, barrier performance, curing route (UV/thermal), mass per m2	Shifts GWP by double digits; constrains EoL options	All	[12,32,40,42]
Data tier and pedigree	Tier (1–3), measurement vs. model, TRL, vintage; uncertainty ranges	Enables uncertainty propagation and fair weighting	All	[24,26,60]
Scenario levers	Grid carbon pathways, solvent recovery, Ag reduction, yield/throughput	Transparent sensitivity; supports prospective results	All	[24,48,69,70,81,82,86,87]
BOS and power electronics	Inverter rating, lifetime, replacement, efficiency	BOS can be non-trivial in system LCAs	Systems	[37,48,84]
EoL modeling	Collection, transport, process flows, recovery yields (Si, Ag, glass, Al), credits/allocation	Determines circularity benefits; can reduce GWP by ~40–60%	All	[30,32,71,72,73,74,75,88,89,90,91,92]

summarizes the recommended reporting requirements for transparent and reproducible process-level inventories, including functional-unit specification, boundary definition, foreground/background separation, per-step energy and material flows, data-quality tier assignment, and uncertainty documentation. To illustrate the practical implications of the reporting rules summarized in Table 4, a short worked example is provided using the pre-industrial perovskite (p-i-n) module case documented in Appendix A. Using identical foreground LCIs and background dataset links, impacts are recalculated under three alternative reporting choices: (i) functional unit (FU) of 1 m² active area under cradle-to-gate boundaries; (ii) FU of 1 kWp under cradle-to-gate boundaries; and (iii) FU of 1 kWp under cradle-to-grave boundaries including end-of-life recycling. The numerical results, summarized in

Table 5. Worked example illustrating the sensitivity of reported GWP to FU and system-boundary choices for a pre-industrial perovskite module (Appendix A). Identical foreground LCIs and background datasets are used in all cases.

Case	Primary FU (Standardized Reference)	Secondary FU 1	System Boundary	Normalized GWP (Relative to Case A)	Key Observation
A	1 m2 active area	1 kWp	Cradle-to-gate	1.00	Baseline process-level reporting using area-based reference FU.
B	1 m2 active area	1 kWp	Cradle-to-gate	≈1.5–2.0	Conversion to a power-based FU increases reported impacts depending on efficiency and active-area ratio.
C	1 m2 active area	1 kWp	Cradle-to-grave (incl. EoL recycling)	≈0.5–0.6	Inclusion of closed-loop recycling reduces cradle-to-grave GWP by ≈40–50% relative to the no-recycling baseline.

¹ All cases are anchored to the same standardized reporting reference (1 m² active area), and 1 kWp is reported consistently as a derived FU using the same conversion assumptions across cases: module efficiency = 20% (Standard test conditions (STCs) 1000 W/m²) and active-to-total-area ratio = 0.9 (Appendix A, Table A1). The foreground LCIs and background dataset links are identical in Cases A–C; only the FU expression and system boundary differ.

, demonstrate that differences in FU and boundary choice alone can shift reported GWP by factors of ~1.5–2, even when no underlying process data are changed.

A recurring source of bias in comparative PV LCAs is the uncritical mixing of aggregated composite datasets (e.g., ‘multi-crystalline Si module, Europe’) with process-resolved inventories for emerging technologies (perovskite, tandem, OSC). Aggregated module datasets often embed multiple foreground steps (glass, encapsulation, metallization, wafering, utilities) under dataset-specific cut-offs and allocations. If such composite datasets are combined with process-level LCIs, double counting and inconsistent exclusions can occur without being apparent to the reader. For this reason, the present review uses aggregated databases exclusively for explicitly declared background flows (e.g., electricity, gases, commodity chemicals, and generic materials) and requires transparent reporting of dataset identity, region, year, and version for every background link. This separation makes cut-off inconsistencies visible and supports reproducible harmonization across studies.

Importantly, this framework also clarifies how scale-up and learning effects should be handled in prospective LCAs. Rather than prescribing fixed or universally applicable learning curves, the proposed methodology treats scale-up explicitly through scenario-based parameterization of throughput, yield, uptime, duty cycles, and electricity decarbonization, as implemented in Section 5.1 and Section 5.2. This approach reflects the empirical observation that reported learning rates vary widely across processes, technologies, and maturity levels and that extrapolating fitted learning curves beyond observed data can introduce spurious precision. By making scale-dependent assumptions explicit and testable, the framework supports robust prospective modeling while avoiding overconfidence in long-term extrapolation.

To support practical adoption of these recommendations, Appendix A provides a complete standardized data template, including metadata, process maps, foreground inventories, background dataset links, electricity scenarios, data-quality structures, end-of-life modeling, and validation checks. The appendix tables serve both as reproducible documentation for the worked example and as a reusable template for future PV LCA studies.

While the proposed reporting checklist and LCI template provide a harmonized minimum standard for process-resolved PV LCAs, several limitations must be acknowledged. First, for emerging technologies and pre-industrial lines, complete process data are often unavailable; in such cases, scaling, interpolation, or proxy assumptions remain necessary and should be transparently documented. Second, reported results remain sensitive to the choice and version of background databases (e.g., Ecoinvent, GaBi), particularly for electricity supply, materials production, and transport, which can evolve substantially over time. Third, although the template is designed to be compatible with commonly used LCA software tools (e.g., openLCA version: 2.5.0, Brightway2, SimaPro version: 10.3), implementation details such as foreground–background linking and uncertainty propagation may differ between platforms and require careful user interpretation. Finally, the framework does not eliminate uncertainty inherent to prospective assessments but rather makes its sources explicit; results should therefore be interpreted as conditional on declared assumptions rather than as definitive impact values.

Rather than prescribing a rigid format, the reporting checklist offers a harmonized minimum standard that all future process-resolved PV LCAs should meet. Consistent application of these elements will enable meaningful comparison across different manufacturing routes, geographical contexts, and technology-readiness levels, supporting integration with techno-economic analyses and prospective LCA models.

8. Conclusions

Achieving genuinely low-impact next-generation PV technologies ultimately depends on understanding and optimizing the manufacturing processes that enable their industrial deployment, rather than focusing solely on device-level performance. Across perovskite, tandem, thin-film, and crystalline-silicon routes, the dominant contributors to cradle-to-gate impacts are consistently found in energy-intensive steps such as drying and annealing, substrate preparation, vacuum pumping and heating, and encapsulation. These processes exhibit strong sensitivity to electricity carbon intensity, duty cycles, throughput, and idle power, making manufacturing configuration and operational context decisive environmental drivers.

Evidence from pilot and industrial production lines demonstrates that scale-up itself is a major environmental lever. As manufacturing transitions from laboratory conditions to continuous, high-uptime industrial operation, improvements in tool loading, line speed, yield stabilization, and solvent and heat-management systems typically reduce cradle-to-gate environmental footprints by approximately 30–70%, depending on technology and process configuration. These reductions are not universal: the largest relative gains are observed for electricity- and throughput-dominated steps such as solution coating and annealing in perovskite and thin-film manufacturing, whereas more material-intensive or already optimized processes, such as crystalline-silicon wafering and module assembly, exhibit smaller, but still meaningful, improvements toward the lower end of this range.

Comparative LCAs further show that technology rankings are highly context-dependent. Regional electricity mix, manufacturing geography, encapsulation and substrate choices, and supply-chain configuration often exert a stronger influence on environmental outcomes than absorber chemistry alone. When functional units, system boundaries, and electricity assumptions are harmonized, much of the apparent disagreement between studies can be traced to methodological choices rather than intrinsic technology differences.

Circularity emerges as a critical pillar for future PV sustainability. High-yield recovery of silver, silicon, glass, aluminum, and other critical materials can reduce cradle-to-grave greenhouse-gas emissions by approximately 40–60%, provided that recovery efficiencies exceed industrial threshold values and that allocation rules and material-substitution assumptions are applied consistently. These benefits are conditional: recycling outcomes depend strongly on collection scale, process integration, electricity intensity, and the quality and substitutability of recovered materials.

Despite significant methodological progress, uncertainty remains concentrated in vacuum-tool energy use, solvent balances, heating profiles, encapsulation inventories, and end-of-life modeling for emerging stacks such as perovskite and tandem modules. Addressing these gaps requires improved process metrology, transparent reporting of foreground and background inventories, and consistent treatment of scale-up and recycling assumptions.

By harmonizing process-level LCIs, clarifying functional units and system boundaries, and providing standardized, open reporting templates, this work supports reproducibility, comparability, and integration with prospective and techno-economic models. Together, these contributions enable more reliable environmental assessments and support the design of scalable, circular, and low-carbon industrial pathways for next-generation photovoltaic technologies.