Solar PV Module Manufacturing: Cell-to-Module Losses, Lamination Kinetics and the EL Gate
India entered November 2026 with roughly 228 GW of ALMM-listed module capacity and about 38 GW of listed cell capacity. That 6:1 ratio is the most important number in the domestic solar industry: it describes where value is added, not a manufacturing achievement. Module assembly is the low-capex, low-barrier end of the chain — a turnkey 1 GW line lands at roughly ₹170 crore in India, against ₹700 crore for the equivalent ingot-and-wafer capacity and no domestic polysilicon at all.
Which means the people building modules are, by and large, not the people who made the cells. They buy cells at a specification, and their entire value-add is a set of process steps that can only lose power. Every decision — ribbon geometry, encapsulant chemistry, lamination profile, junction box — is an exercise in losing as little as possible.
This article quantifies that loss, works through the interconnection physics, derives the EVA cure window from Arrhenius kinetics, and explains why a line that skips 100% electroluminescence inspection is not measuring its own yield.
1. The cell you are handed
The 2026 production mix is settled. TOPCon took roughly 75% of crystalline silicon cell output in 2025; PERC holds about 14% and is being retired; HJT is around 5%. ITRPV's 2026 mass-production figures give cell efficiencies of 25.6% for TOPCon, 25.6% for HJT and 26.1% for TOPCon-back-contact (TBC), translating to module efficiencies of 23.9%, 23.9% and 24.3% respectively.
The mechanism behind this generation is a tunnelling contact. TOPCon stacks a ~1.5 nm silicon dioxide layer under a doped polysilicon film on the rear: thin enough for majority carriers to tunnel, thick enough to block minority-carrier recombination — passivation and ohmic contact in the same 1.5 nanometres. HJT achieves the same with intrinsic amorphous silicon and a TCO, buying a higher open-circuit voltage (745–748 mV against ~720 mV) and a superior temperature coefficient, −0.24 to −0.26 %/°C against −0.30 %/°C.
Metallisation is where module manufacturing first touches the cell. Mainstream TOPCon prints fingers about 24 µm wide, keeping shading below 3%, and sinters silver-aluminium paste near 750 °C to form a ~10 nm contact; fill factors of 83–84.5% are routine. The catch is silver. Paste consumption has become one of the largest single cost lines in the cell, and by 2026 cost models put the metal at 15–20% of module BOM and up to 30% of cell cost, against about 3% in 2023.
2. What a module can and cannot fix
A solar cell is well described by the two-diode model:
with J_{01} the diffusion (band-to-band, ideality n_1 \approx 1) saturation current density and J_{02} the depletion-region recombination term (n_2 \approx 2). Module assembly can only touch three of these six parameters, and only two of them meaningfully.
- R_s — increased. Soldered ribbon and contact resistance add series resistance. Always a loss.
- J_{ph} — usually increased slightly. Encapsulation replaces the air/Si interface with EVA/Si, and EVA's refractive index (≈1.48) sits far closer to the silicon nitride ARC (≈2.0) than air does. Light that would have reflected off the bare cell now enters.
- R_{sh} — essentially unchanged, though a poorly cured encapsulant can shunt via ionic contamination.
- J_{01}, J_{02}, n — untouched. These are set by wafer quality, passivation and emitter diffusion. No module process recovers a bad cell.
Fill factor is the parameter that reports on this:
and the series-resistance penalty in power terms is the ratio of added resistance to the cell's characteristic resistance:
For a 580 W module at 89 V and 6.5 A, R_{ch} \approx 13.7\ \Omega. Every milliohm of added interconnect resistance costs 0.0073% of power, and a full string's worth of ribbon and solder joints lands in the 15–25 mΩ range. That is the whole game.
3. The cell-to-module ledger
The cell-to-module (CTM) ratio is module power over the sum of the cell powers that went into it:
In the early 2000s CTM loss ran 2–8%. Encapsulation engineering has since pushed mature designs close to unity, and in a few cases past it — a multiwire-back-contact module with anti-reflective glass has demonstrated a 2.3% CTM current gain in I_{sc}, offset by a 1.4% absolute fill-factor drop.
The individual channels, with magnitudes from published CTM decompositions:
- Front glass reflection and absorption: −1.5 to −2.5%. Low-iron tempered 3.2 mm glass transmits 91–92% at AM1.5; AR coating adds 2–3 points.
- Encapsulant absorption and internal interface reflection: −0.5 to −1.0%. Glass/EVA interface reflection is below 0.1% — negligible, because the indices nearly match.
- Optical coupling gain: +1.0 to +2.5%. Light reflected off fingers and busbars is trapped by total internal reflection at the glass/air interface and re-enters the cell. Larger for textured mono than for older poly.
- Cell gap / dead area: −1.0 to −2.0%, and worse for full cells than split cells, since perimeter-to-area ratio is what matters.
- Ribbon shadowing: −0.5 to −1.5%. Round MBB wire recovers part of this by redirecting light sideways into the cell.
- Ribbon resistive (I^2R): −0.5 to −2.5%, strongly geometry-dependent.
- Contact resistance, rear aluminium spreading, junction box and cable: −0.5 to −1.3% combined.
- Mismatch from cell binning: −0.3 to −1.0%. A string is limited by its worst cell, which is why tighter power bins cost the cell supplier money and save the module maker money.
- Rear reflection gain (white backsheet, or bifacial rear irradiance): +1.0 to +2.5%.
Two structural effects sit outside the product. First, module efficiency loses more than module power, because the frame and the white space around the cells add area that generates nothing — about 1% absolute for polycrystalline designs, up to 3% absolute for mono. Second, cell format matters enormously. Fraunhofer ISE's CTM analysis from M0 (156.75 mm) to M12 (210 mm) found CTM power ratio falling about 5 percentage points absolute with increasing cell size, because interconnect losses rise faster than backsheet reflection gains. M12 full cells lose 7.4% of module power in the interconnection alone; third-cutting an M12 recovers +7.7 percentage points absolute of CTM ratio against the full cell. Which is to say the industry's migration to larger wafers only works because it is paired with cell splitting.
4. The ribbon, derived
A tabbing ribbon is not a wire carrying uniform current. Current enters at one end and is progressively deposited into the busbar along the soldered length, so the current at position x is I\,x/L. The dissipated power is
The factor of 1/3 is not a fudge — it is the integral of a linear ramp, and it appears in every interconnect loss model. Take a 580 W module: 144 M10 half-cells, ten round copper wires of 0.4 mm diameter, 91 mm soldered length, string current 6.85 A.
- Cross-section: \pi (0.2)^2 = 0.1257\ \text{mm}^2
- Resistance per metre at \rho = 0.0175\ \Omega\cdot\text{mm}^2/\text{m}: 139 mΩ/m
- Per wire over 91 mm: 12.67 mΩ
- Current per wire (shared across 10): 0.685 A
- Loss per wire: 0.685^2 \times 0.01267 / 3 = 1.98\ \text{mW}
- Per half-cell: 19.8 mW. Whole module: 2.85 W, or 0.49% of 580 W.
Now run the same arithmetic on an M12 full cell at 18.7 A, 210 mm, 66 cells. Per-wire current is 1.87 A, per-wire resistance 29.2 mΩ, loss per wire 34.1 mW, per cell 341 mW, module total 22.5 W — 3.2% of 700 W. Same ribbon, same busbar count, ten times the loss, entirely because the current never got split.
Which is the whole argument for half-cutting. Compare one full cell against two half-cells of the same total area:
Halving the current quarters the loss; halving the length adds another factor of two per piece, with twice as many pieces. Net: a 4× reduction in interconnect loss — which is why every utility-scale module shipped in 2026 is half-cut or third-cut.
Three refinements matter in production. Multi-busbar geometry: going from three flat busbars to fifteen round wires lets finger length drop from 25 mm to 5 mm and finger width from 50 µm to 17 µm — a silver reduction of up to 89% — while the round profile reflects light back onto the wafer rather than shadowing it. The cost is that solder coating uniformity on a round wire is harder to hold, and non-uniform coating is a crack initiation site under thermal cycling. Terminal position: current does not divide equally among busbars; stringing an M6 half-cell with the terminal at an outer ribbon rather than the centre loses about 0.4 W more, with the inhomogeneity costing roughly 2.1% of module power. Rear aluminium: 2D network simulation puts the rear ribbon's effective resistance at 1.7 mΩ against the 2.5 mΩ analytical prediction, while the rear aluminium sheet — normally ignored — contributes 17% of total interconnect resistive loss. The two errors partially cancel; they do not cancel in a thermal model.
5. The solder joint is a fatigue specimen
Every ribbon is soldered at roughly 175–200 °C and cools to room temperature. Over that 150 K, copper wants to contract 0.255% and silicon 0.039%. The mismatch is
Over a 91 mm half-cell that is 0.197 mm of differential contraction. If the interface were fully constrained, the copper would carry \sigma = E\varepsilon \approx 259 MPa — above the 200–250 MPa tensile strength of annealed ribbon.
This is why interconnect ribbon is specified soft: oxygen-free copper ≥99.99%, resistivity ≤0.0175 Ω·mm²/m, tensile strength 200–250 MPa annealed, elongation ≥25%, with a 10–25 µm Sn60Pb40 or lead-free coating. The low yield strength is a feature — the copper must deform plastically so the silicon does not. Residual stress in the wafer after soldering measures around 0.116 GPa: below fracture strength, but not comfortably so in a wafer that already carries edge microcracks from diamond-wire sawing.
The joint is a Cu / solder / Ag-paste sandwich, and the two interfaces are not equivalent. Silver paste has a CTE of 10.4 × 10⁻⁶/K against copper's 17.0 × 10⁻⁶/K, so the solder/Ag interface sees the larger mismatch and is where cracks initiate. After 800 thermal cycles, 57% of contact pads failed where continuous interconnectors linked a pad row, against 27% for isolated joints — which showed no defects at all up to 400 cycles. Peak solder stress is 21 MPa, concentrated longitudinally toward the joint ends.
Two production consequences. First, the outermost contact pads should be larger than the inner ones, because they take the highest tensile stress; pad-to-edge distance is a genuine trade-off, since moving pads outward reduces tensile stress but places them in the pre-damaged region where cracks start. Second, lead-free is not free. SnAgCu reflows 35–40 K hotter than Sn62Pb36Ag2, raising residual stress on cool-down, and creep modelling against real tropical outdoor profiles found 43–58% more accumulated creep damage for Sn3.8Ag0.7Cu than for Sn60Pb40. A 25-year-old field-returned module analysed in the literature had lost 23% of original efficiency, with the failure traced to exactly this: solder cracking at the solder/Ag-paste interface, reproduced in the lab by thermal cycling.
6. The layup, and what each layer is for
A glass-backsheet module is seven material systems stacked and cured. Front glass is 3.2 mm low-iron tempered, 91–92% transmission, or 2.0 mm for glass-glass bifacial at 10–11 CNY/m² against 16.8–18 CNY/m² for 3.2 mm coated. Encapsulant is 0.45–0.50 mm EVA, POE or EPE front and rear — EVA is the incumbent, POE has higher volume resistivity and no acetic acid evolution, which matters for PID and grid-side durability. Cells are ~140 µm of silicon with screen-printed contacts, joined by tinned copper ribbon at 0.2 mm nominal. Backsheet is a PVF/PET laminate 0.3–0.5 mm, or a second glass sheet. Frame is anodised AlMg3 profile 30–35 mm deep, and the junction box is PPO or PA with three bypass diodes and 4 mm² leads.
By mass the stack is almost entirely glass and aluminium. Fraunhofer's breakdown at 11.6 kg/m²: glass 67%, aluminium frame 13%, EVA 7%, backsheet 6%, silicon 2.7%, copper 1.3%, junction box and silicones ~3%. Silicon — the only layer that generates electricity — is a rounding error by weight.
By cost the order inverts. For a 2026 bifacial TOPCon module: complete cell 60–65% (of which polysilicon 10–15%, wafer and cell process 30–35%, silver paste 15–20% of the whole module), glass 10–13%, aluminium frame 8–11%, encapsulant 4–7%, backsheet 2–4%, junction box plus ribbon 2–4%, and labour, depreciation and overhead 5–10%.
In a fully integrated Chinese plant the gate cost of a 580 W module is roughly \0.085/W: polysilicon \0.010–0.012, wafer ex-poly \0.008–0.010, cell ex-wafer \0.019–0.024, glass/EVA/backsheet/frame \0.020–0.026, cell-to-module assembly and overhead \0.015–0.020, and 4–7% margin. The factory's controllable cost is that \$0.015–0.020 assembly block; the BOM is a pass-through. Which is why module assembly is a scale and logistics business, not a margin business.
7. Lamination is a chemical reaction with a dead band
The laminator does three things in sequence: evacuate the air between layers, heat and press to melt and flow the encapsulant, then hold to crosslink it. A typical cycle on a glass-backsheet stack:
- Rough evacuation, ~4 min, stack on the heating plate.
- Fine vacuum, ~60 Pa, pulling residual gas out of the stack.
- Pressure applied, ~85 kPa via the diaphragm; the stack makes full thermal contact and the cure clock starts.
- Hold at 144–152 °C plate temperature. Ten minutes is the historical industrial standard.
- Vent and cool, which stabilises the crosslink network and limits thermal degradation.
EVA cures by peroxide thermolysis. The initiator — typically a peroxide such as TBEC — homolytically cleaves into radicals, abstracts hydrogens from the vinyl acetate side-chain methyl groups, and the resulting chain radicals recombine into a three-dimensional network. Because it is a thermally activated first-order process:
With E_a \approx 130 kJ/mol — representative of organic peroxide decomposition — the time–temperature equivalence against a 145 °C / 10 min reference runs like this:
- 135 °C: 25.0 min for the same conversion
- 140 °C: 15.7 min
- 145 °C: 10.0 min
- 150 °C: 6.4 min
- 155 °C: 4.2 min
- 160 °C: 2.7 min
- 170 °C: 1.2 min
That is a factor of 8 across 25 K. It explains why lamination recipes are aggressive — cycle time is the throughput bottleneck — and why they are dangerous: a 2 K plate non-uniformity at 165 °C is a 20% conversion error; at 145 °C it is 6%.
Measuring conversion properly means differential scanning calorimetry. Heat a cured sample at 10 K/min and integrate the residual exotherm:
An uncured EVA reference gives about 20 J/g; a sample reading 0.6 J/g is 96.8% crosslinked, one reading 2.4 J/g is 88.0%. The alternative is Soxhlet extraction for gel content, the industry's accepted number but destructive, 16–24 hours, and impossible on a moving module.
Which brings us to the specification, where most factories are quietly wrong. Published guidance is all over the place: manufacturer minimums run from 60% to 90%, the original JPL work set 65%, and the widely-quoted industry pass criterion is 75–80%. The measured optimum is narrower and two-sided. For a fast-cure EVA with 34% vinyl acetate and TBEC initiator:
- Below 70%: low chemical and optical stability, weak peel strength at the EVA/backsheet interface, and EVA that still flows — which can shift cells and propagate cracks.
- 84–90%: the optimum, reached at 135 °C/10 min through 145 °C/10 min.
- Above 92%: greater yellowing tendency, less thermally stable adhesion. Over-cure co-occurs with void formation, because peroxide decomposition releases gas faster than it can diffuse out of a thickening melt.
The most uncomfortable number in this article: a field survey of 254 EVA samples extracted from 120 deployed modules found only two thirds in an acceptable range. The 65% JPL minimum is too permissive for a 25-year warranty, and "more is better" is wrong at the top end too.
Double-side heating shortens the cycle 10–30% by halving the thermal path, and lets the plate run hotter safely — a 5-minute process at 170 °C reaches 85% gel content with 3.5% maximum deviation across a full-size module, and conversion saturates at 94–95% above 170 °C for holds of 6 min or more. Peel strength is the other gate: a mainstream encapsulant supplier specifies 53 N/cm minimum on 180° peel to support a 20-year life.
8. Framing, junction box, and the hot spot you are insuring against
The junction box is not a connector housing. It contains the bypass diodes that decide what happens when a module is partially shaded.
Take a 144 half-cell module with three bypass diodes, so 48 half-cells per group. Group V_{oc} is 34.6 V, which is why the diodes are 45 V Schottky parts. Under uniform illumination each cell runs at +0.62 V. Now shade one cell completely: it stops generating, the other 47 keep pushing current through it, and it is driven into reverse bias until it avalanches or the diode conducts.
189 W into a 166 cm² cell is a flux of 114 kW/m² — 114 suns, in a stack insulated by glass on one face and a backsheet on the other. Cell temperature exceeds 150 °C within minutes. That is why the diode exists, why it must be a low-forward-drop Schottky, and why IEC 61215's hot-spot endurance test (MQT 09) is a pass/fail gate rather than a data point. The diode clamps the group to about −0.4 to −0.7 V, dissipation collapses to a few watts inside the diode package, and the diode becomes the component that must survive.
The failure mode that bypasses all of this is a cracked cell or broken interconnect inside a string. A cell that is electrically isolated but physically present becomes a permanently shaded cell in a group whose diode does not know to fire. That is the mechanism behind the dark, burned spots in fielded modules, and it is why electroluminescence matters more than any electrical test.
9. Flash testing, binning, and the only honest yield metric
A finished module goes onto a flash simulator: 1000 W/m², AM1.5G, 25 °C cell temperature, calibrated against a reference cell traceable to a national lab. A good production flasher is uncertain by ±1.5–2%, uncomfortably close to the spread of the power bins it is sold into — which decides whether a module lands in a +5 W bin or a +0 W bin, and compounds across annual output.
Electrical testing tells you P_{max}, V_{oc}, I_{sc} and FF. It does not tell you whether the module will still be at 85% of that in 25 years, because a module with forty microcracked cells, a partially detached ribbon and three cold solder joints will flash-test perfectly well on the day it is made. Series resistance rises later, as cracks propagate and intermetallics coarsen.
Electroluminescence is the answer. Forward-bias the module and the silicon emits at its bandgap — about 1150 nm for c-Si. Image it with a cooled silicon CCD or InGaAs camera and every recombination-active defect appears dark: cracks, inactive areas, broken fingers, disconnected busbars, shunts, and the black lines of a detached ribbon. Sub-second exposure at 20–40 A bias, non-contact, non-destructive.
Good practice is two EL inspections: one after stringing and before layup, catching soldering defects while the string is still reworkable, and one after lamination, catching cracks from handling and lamination and becoming the record that ships with the serial number. An inline EL station costs a fraction of a stringer and catches the defect class warranty claims are actually made of — which is why a factory quoting capacity without one is advertising an assembly operation, not a module factory.
10. The qualification sequence is a sequence
IEC 61215 (design qualification, second edition 2021) and IEC 61730 (safety, third edition 2023) are tested on shared samples in a defined order, and the order is the point. You do not get to run damp heat before thermal cycling.
The IEC 61215-2 sequence includes MQT 01 visual, MQT 02 maximum power, MQT 09 hot-spot endurance, MQT 10 UV preconditioning, MQT 11 thermal cycling (TC50 or TC200 depending on assessed thermal load), MQT 12 humidity-freeze (HF10), MQT 13 damp heat (DH200 or DH1000), MQT 16 static mechanical load, MQT 17 hail, MQT 18 bypass diode thermal and functional, MQT 19 stabilisation including BO-LID, and MQT 21 potential-induced degradation.
The umbrella pass criterion is blunt: no more than 5% power loss after the sequence, no major visual defect, insulation intact. Damp heat at 85 °C/85% RH for 1000 h finds bad encapsulant, bad lamination and bad edge sealing; thermal cycling from −40 to +85 °C finds bad solder joints; humidity-freeze finds delamination. PID testing — module at system voltage under elevated temperature and humidity, typically 60 °C/85% RH for 96 h with an 85 °C variant — finds an encapsulant with too low a volume resistivity, or a cell too susceptible to sodium migration.
Note what the standard says about itself: these are accelerated tests empirically derived to reproduce known field failures, applied uniformly across module types, and "not construed as a quantitative prediction of module lifetime." A module can pass IEC 61215 and fail in eight years. The sequence is a floor, not a warranty.
11. The factory arithmetic
India's module line capex is roughly ₹170 crore per GW against ₹700 crore for ingot-and-wafer, and even module-level equipment is 70–80% imported — stringer and laminator technology is standardised enough that local integration works, which is not true of a diffusion furnace or a diamond wire saw. Cell lines run about ₹620 crore/GW in China and ₹880 crore/GW elsewhere; module assembly is roughly ₹160 crore/GW in China and ₹250 crore/GW for a new Indian, EU or US build.
The sensitivity that kills module factories is yield, and the two yield gates are stringing and lamination. On a 33.4 ¢/W build cost quoted at 36.5 ¢/W — a 3.1 ¢/W, 9.3% margin — a 2% scrap rate at lamination adds 0.67 ¢/W of cost and erases 22% of the margin. No other process improvement in the plant has that leverage.
Policy is currently doing more work than process engineering. ALMM List-II, in force from 1 June 2026, extends domestic-content requirements from modules to cells at an estimated ₹32,500–37,900 crore in FY2026-27 — more in one year than the entire PLI outlay. DCR TOPCon modules traded at about ₹22.3/W in April 2026 against ₹14.2/W for non-DCR, a 57% premium that exists entirely because domestic cell capacity is 38 GW against 228 GW of module capacity.
And the bill of materials is the next constraint. Domestic capacity in September 2026 stood at roughly 20 GW for solar glass, 28 GW for backsheet, 49.5 GW for encapsulant, 54.25 GW for PV ribbon, 66 GW for aluminium frames and 153 GW for junction boxes. Glass is the binding constraint by a factor of three — and glass is also 10–13% of module cost. A country can assemble 228 GW of modules and still be unable to seal them.
12. What this means if you are buying
Four things to ask a module supplier, all answerable, none on a datasheet:
- The EL image of the actual serial number, not a reference image. Post-lamination EL with a visible frame counter is the only evidence that the specific module you received has intact cells.
- The gel content specification and the lamination recipe. A supplier quoting "≥65%" is quoting the 1980s JPL floor; one quoting 75–90% with a documented DSC or Raman inline check is running a controlled process.
- Which encapsulant, and why. EVA is fine for temperate deployments; POE or EPE is right for high-humidity, high-voltage-string or coastal installations — and the answer should be technical, not a price.
- The CTM ratio: module power divided by the summed cell power of the bill of materials. Below 0.96 means the assembly is giving away more than four points, and at ₹22/W that is real money per megawatt.
The gap between 25.6% cell efficiency and 23.9% module efficiency is 1.7 percentage points, and it is not a law of physics. It is a stack of process decisions — ribbon geometry, encapsulant chemistry, cure profile, solder alloy, inspection regime — every one controllable, most invisible on the datasheet. In a market where the cell is bought in and the BOM is a pass-through, those 1.7 points are the only thing a module factory actually makes.