When I first began supervising undergraduate laboratory work in optoelectronic materials and devices, I noticed a persistent gap between what students could explain on paper and what they could actually accomplish with their hands. They could recite the operating principle of a photovoltaic device, sketch a band diagram, and define the photovoltaic conversion efficiency in an examination, yet almost none of them had ever held a spin coater, loaded a precursor solution into a glovebox, or interpreted a current density–voltage curve that they themselves had produced. This gap motivated me to build a project-based, open-ended experiment around perovskite solar cells, a family of devices that has moved from laboratory curiosity to one of the most intensively studied photovoltaic platforms in the world within little more than a decade. My aim was not merely to add another verification exercise to the curriculum, but to construct a genuinely exploratory setting in which students would formulate a question, design a fabrication route, deposit a functional thin film, complete a full device stack, measure its behaviour under simulated sunlight, and then argue about why the numbers came out the way they did.
I have since run this project repeatedly with mixed cohorts drawn from optoelectronic information science and engineering. The version I describe here centres on a single, sharply defined variable: the way an antisolvent, chlorobenzene, is introduced during the spinning of the light-absorbing layer. This choice is deliberate. It is simple enough that a student team can master the handling within a few sessions, yet subtle enough that the resulting changes in film morphology produce large, unambiguous differences in device performance. In my experience, nothing teaches the concept of structure–property relationships in perovskite solar cells more convincingly than watching a short-circuit current density fall from roughly 13 mA cm⁻² to under 3 mA cm⁻² merely because a dripping step was moved by a few seconds.
Background and Motivation
Photovoltaic technology divides broadly into silicon-based and non-silicon-based families. Crystalline silicon still dominates the market, with record device efficiencies approaching 26 %, but the energy payback associated with high-purity wafer production, together with the capital intensity of the manufacturing line, keeps the cost floor relatively high. Thin-film alternatives based on perovskite solar cells offer a different proposition. The absorber can be deposited from solution at temperatures below 150 °C, the total amount of active material required is minuscule, and the certified efficiencies of the best laboratory devices now rival those of established silicon technology. The catch, and it is a serious one, is operational stability: the perovskite lattice is sensitive to moisture, oxygen, thermal stress, and illumination, and degradation pathways that are negligible on the timescale of a laboratory measurement become decisive over a service life of twenty years.
For teaching purposes, however, that instability is not a drawback. It is a natural entry point into a discussion of encapsulation, interface engineering, ion migration, and the distinction between intrinsic and extrinsic degradation. I encourage students to treat the fragility of perovskite solar cells as an engineering constraint rather than a defect, and to ask what design decisions follow from it.
The core functional layer is the perovskite absorber itself. Its crystal structure is described by the general formula ABX₃, in which the A-site cation occupies the corners of a cubic lattice, the B-site cation sits at the centre of a BX₆ octahedron, and the X-site anion occupies the octahedral vertices. In hybrid halide systems, A is typically an organic cation such as methylammonium or formamidinium, sometimes combined with an inorganic species such as caesium; B is a divalent metal cation, most commonly lead but sometimes tin; and X is a halide, iodide, bromide, chloride, or a mixture of these. Because the A, B, and X sites can be substituted and mixed over a wide compositional range, the optical gap, carrier mobility, and structural tolerance of the material can be tuned continuously.
The geometrical viability of a given composition is conveniently assessed through the Goldschmidt tolerance factor and the octahedral factor:
$$t=\frac{r_A+r_X}{\sqrt{2}\,\left(r_B+r_X\right)}$$
$$\mu=\frac{r_B}{r_X}$$
where rA, rB, and rX are the effective ionic radii. Empirically, a cubic perovskite is favoured when t lies between roughly 0.9 and 1.0 and μ lies between roughly 0.44 and 0.90. I ask students to compute these quantities for the compositions used in the laboratory and to rationalise why the mixed-cation formulation adopted here, combining caesium, formamidinium, and methylammonium on the A site with iodide and bromide on the X site, produces a more robust film than any single-cation analogue. The exercise costs almost nothing in laboratory time and reliably produces a lively discussion about why empirical stability rules, not just calculated tolerance factors, govern real device behaviour.
Device Architecture of the Perovskite Solar Cells Studied
A complete device comprises five functional elements: a transparent conducting oxide electrode, an electron transport layer, the perovskite absorber, a hole transport layer, and a metal back contact. Depending on which transport layer faces the illuminated transparent electrode, the stack is classified as either conventional (n–i–p) or inverted (p–i–n). In the conventional configuration the electron transport layer is deposited on the transparent electrode; in the inverted configuration the hole transport layer is deposited first. The two arrangements differ substantially in their charge-extraction dynamics, their hysteresis behaviour, and their compatibility with tandem architectures.
I selected the inverted configuration for this project for several practical reasons. Inverted perovskite solar cells generally exhibit lower hysteresis, which means that the current density–voltage curve measured in the forward sweep is closer to that measured in the reverse sweep, and students are therefore less likely to be confused by measurement artefacts when they compare samples. The inverted stack is also compatible with low-temperature processing throughout, and it tolerates the organic hole transport material used here without requiring the doped spiro-OMeTAD formulations that demand careful atmospheric control.
| Layer | Function | Material used | Deposition route |
|---|---|---|---|
| Transparent electrode | Light transmission and hole collection | Indium tin oxide on glass | Commercial substrate, solvent-cleaned |
| Hole transport layer | Selective hole extraction, electron blocking | PEDOT:PSS | Spin coating, 4000 rpm, 60 s |
| Absorber | Photon absorption, charge generation | Mixed-cation, mixed-halide perovskite | Three-step spin coating, 100 °C anneal |
| Electron transport layer | Selective electron extraction, hole blocking | PCBM | Spin coating, 1000 rpm, 30 s |
| Buffer layer | Interfacial stabilisation, cathode protection | BCP | Spin coating from ethanol |
| Back contact | Electron collection and conduction | Silver | Thermal evaporation, 100 nm |
Why the Antisolvent Step Matters
The perovskite film is formed by spin coating a precursor solution containing the constituent cations and halides dissolved in a polar aprotic solvent mixture. Because the precursor is highly soluble in that solvent, uniform nucleation does not occur spontaneously during spinning; if the film is simply allowed to dry, the solvent evaporates slowly, the solution becomes concentrated in an uncontrolled manner, and crystallisation begins at a small number of favoured sites. The result is a coarse, discontinuous layer with large voids between isolated islands of material.
Antisolvent engineering interrupts this process. A solvent in which the perovskite precursors are essentially insoluble, but which is fully miscible with the host solvent, is dropped onto the spinning film. The antisolvent extracts the host solvent rapidly and uniformly, driving the solution far beyond its equilibrium solubility within milliseconds. Nucleation then occurs simultaneously across the entire film area, and the large number of nuclei that form consume the supersaturated solution before individual crystals can grow large. The outcome is a dense, fine-grained, pinhole-free layer.
The driving force for this transformation is the supersaturation ratio
$$S=\frac{C}{C_{eq}}$$
where C is the instantaneous precursor concentration and Ceq is the equilibrium solubility in the prevailing solvent mixture. According to classical nucleation theory, the nucleation rate depends exponentially on the free-energy barrier to nucleus formation:
$$J=A\exp\left(-\frac{\Delta G^{*}}{kT}\right)$$
$$\Delta G^{*}=\frac{16\pi\gamma^{3}}{3\left(\Delta G_{v}\right)^{2}}$$
Here γ is the interfacial energy between the nucleus and the solution and ΔGv is the volumetric free-energy change accompanying crystallisation, which scales with the supersaturation. Because the barrier depends on the inverse square of the driving force, modest increases in S produce dramatic increases in nucleation rate. This is precisely why the timing of the antisolvent addition is so consequential: if the antisolvent arrives too late, some solvent has already evaporated and crystallisation is already underway, leaving an inhomogeneous film; if it arrives too early, the film is still dilute and the antisolvent may simply wash material away.
I also introduce the students to the Gibbs–Thomson relation, which explains why the morphology continues to evolve after nucleation:
$$C(r)=C_{\infty}\exp\left(\frac{2\gamma V_{m}}{rRT}\right)$$
Small crystallites have a higher local solubility than large ones, so during annealing the larger grains grow at the expense of smaller neighbours. This coarsening is thermodynamically favoured but it can also open voids, because material must be transported from one location to another. Students observe exactly this competition in the scanning electron micrographs described later, where one sample combines fewer voids with notably larger void dimensions.
Objectives and Requirements of the Open-Ended Project
I frame the project around four capabilities that I want each participant to develop. The first is literature competence: the ability to locate, read, and extract design-relevant information from primary research articles, including those written in English. The second is instrumental competence: safe and confident operation of a glovebox, a spin coater, a thermal evaporator, a solar simulator, and the associated electrical measurement instrumentation. The third is experimental design competence: the ability to translate a vague question into a controlled comparison with a single dominant variable. The fourth is analytical competence: the ability to extract parameters from raw data, to judge whether differences between samples are meaningful, and to connect observations at different length scales.
Rather than prescribing a recipe, I require each team to survey at least ten primary sources before writing a short proposal. The proposal must state the question, justify the choice of materials, specify the process parameters, and predict the outcome. Prediction is important. A team that commits in writing to an expectation is far more engaged when the result contradicts it, and the subsequent discussion is far more productive.
| Capability | Concrete deliverable | Evidence of attainment |
|---|---|---|
| Literature competence | Annotated bibliography of ten or more sources | Correct identification of the open question and of competing explanations |
| Instrumental competence | Independently completed fabrication run | Reproducible film quality across consecutive runs |
| Design competence | Written protocol with a single dominant variable | Adequate controls and explicit parameter table |
| Analytical competence | Parameter extraction and uncertainty estimate | Correct treatment of series resistance and of sample-to-sample spread |
| Communication | Group presentation and written summary | Clear causal argument linking morphology to device metrics |
Formulation of the Experimental Plan
The variable chosen by the team was the manner in which chlorobenzene, abbreviated CB, is delivered during the third and final spin-coating step of the absorber. Four conditions were compared. The first was a control in which no antisolvent was used at all. In the second, the antisolvent was applied immediately after the precursor was dispensed and then the substrate was spun. In the third, the film was allowed to rest for one minute after dispensing before the antisolvent was applied and the substrate spun. In the fourth, the antisolvent was dripped onto the substrate five seconds into the spinning of the third layer. The full scheme is summarised below.
| Designation | Antisolvent handling during third spin-coating step | Rationale |
|---|---|---|
| Control | No chlorobenzene added | Establishes the baseline performance of an untreated film |
| Sample 1 | Chlorobenzene applied immediately after dispensing, soaked 15 s, then spun | Tests the effect of an early, prolonged antisolvent contact |
| Sample 2 | Film rested 1 min after dispensing, chlorobenzene applied, soaked 15 s, then spun | Tests the effect of a delayed application after partial solvent loss |
| Sample 3 | Chlorobenzene dripped 5 s into the spin, then spun to completion | Tests the effect of delivery during active rotation |
All other parameters were held constant. The hole transport layer was deposited from an aqueous dispersion at 4000 rpm for 60 s and annealed at 150 °C for 10 min. The absorber was deposited by a three-step spinning sequence at 1000 rpm for 10 s, 2000 rpm for 10 s, and 6000 rpm for 20 s, followed by annealing at 100 °C for approximately 30 min. The electron transport layer was deposited from a 20 mg mL⁻¹ solution at 1000 rpm for 30 s and annealed at 75 °C for 30 min. A buffer layer was then spun from a 5 mg mL⁻¹ ethanolic solution. Finally, 100 nm of silver was thermally evaporated through a shadow mask to define the device area.
It is worth noting how much of the pedagogical value resides in the constraint that only one variable may change. Students initially propose comparing five or six parameters at once. When I ask them how they would attribute a difference in efficiency to any single one of those parameters, they quickly recognise that a factorial design or a strictly controlled series is required. The discipline of narrowing the question is, in my view, the single most valuable thing they take away.
Fabrication Procedure in Detail
The transparent substrates were cleaned sequentially in acetone, isopropanol, and ultrapure water, dried, and then treated with a plasma cleaner to remove residual organic contamination and to increase surface wettability. Good wetting is a prerequisite for uniform spin coating, and students learn to inspect the contact angle informally by watching how the precursor spreads.
Precursor solutions were prepared by dissolving the constituent salts in a mixed solvent system comprising dimethylformamide and dimethyl sulfoxide in a 4:1 volume ratio. Two stock solutions were prepared, one containing caesium, formamidinium, lead, iodide, and chloride, the other containing caesium, methylammonium, lead, bromide, iodide, and chloride. These were then combined in a 3:7 volume ratio to produce the final multi-component precursor. All weighing was performed on a balance readable to 0.01 mg, and all solution handling was conducted under inert atmosphere.
The film thickness produced by spin coating follows an approximate scaling law that I derive with the students from the balance between centrifugal driving and viscous resistance:
$$h=\left(\frac{3\eta m}{2\rho\omega^{2}}\right)^{1/3}$$
where η is the dynamic viscosity, m is the mass of solution initially deposited, ρ is the density, and ω is the angular velocity. The practical implication is that thickness varies as the inverse two-thirds power of rotation speed, which explains why the third step is run at a high rate: the high speed simultaneously thins the film to the desired optical thickness and accelerates solvent removal. Interpreting the antisolvent step in this framework helps students see that they are modifying the kinetics of a process that is already strongly rate-limited.
Current Density–Voltage Characterisation
Devices were measured under a simulated air-mass 1.5 global spectrum calibrated to 100 mW cm⁻². The current density–voltage response of a photovoltaic device is conventionally described by a single-diode model with series and shunt resistance terms:
$$J=J_{0}\left[\exp\left(\frac{q\left(V-JR_{s}\right)}{nkT}\right)-1\right]+\frac{V-JR_{s}}{R_{sh}}-J_{ph}$$
The four parameters usually extracted from the curve are the short-circuit current density JSC, the open-circuit voltage VOC, the fill factor FF, and the power conversion efficiency η. The fill factor is defined as the ratio of the maximum power rectangle to the product of the two extreme axes:
$$FF=\frac{V_{mp}J_{mp}}{V_{OC}J_{SC}}$$
and the efficiency follows directly:
$$\eta=\frac{P_{max}}{P_{in}}=\frac{V_{OC}J_{SC}FF}{P_{in}}$$
with Pin fixed at 100 mW cm⁻². The series resistance can be estimated from the inverse slope of the curve near open circuit, and the shunt resistance from the inverse slope near short circuit:
$$R_{s}\approx\left|\frac{dV}{dJ}\right|_{V=V_{OC}},\qquad R_{sh}\approx\left|\frac{dV}{dJ}\right|_{J=J_{SC}}$$
The ideality factor, which reports on the dominant recombination mechanism, can be extracted from the illumination dependence of the open-circuit voltage:
$$n=\frac{q}{kT}\frac{dV_{OC}}{d\ln J_{SC}}$$
A value close to unity suggests band-to-band or shallow-trap recombination, whereas values approaching two indicate dominant Shockley–Read–Hall recombination through deep traps. I ask students to estimate this quantity from a short intensity series because it forces them to think about what the absorber quality actually does to the device physics, rather than treating efficiency as an isolated score.
Photovoltaic Results
The measured parameters are collected below. The pattern is striking and, for the students, immediately interpretable in terms of film quality.
| Device | VOC (V) | JSC (mA cm⁻²) | FF (%) | η (%) |
|---|---|---|---|---|
| Control (no antisolvent) | 0.77 | 2.9 | 34.8 | 0.8 |
| Sample 1 (immediate, soaked) | 0.95 | 12.8 | 63.3 | 7.7 |
| Sample 2 (rested 1 min, then applied) | 0.98 | 13.3 | 64.5 | 8.4 |
| Sample 3 (applied during spin) | 0.88 | 2.9 | 71.1 | 1.8 |
The control device is the weakest on every metric. Its open-circuit voltage of 0.77 V and its fill factor of only 34.8 % both point to severe shunting and to inefficient charge collection, which is entirely consistent with the discontinuous film described in the next section. The efficiency of 0.8 % corresponds to a device that is barely functional.
Sample 1 delivers a substantial improvement, with JSC rising to 12.8 mA cm⁻² and efficiency reaching 7.7 %. Sample 2 is the best of the series at 8.4 %, combining the highest short-circuit current density of 13.3 mA cm⁻² and the highest open-circuit voltage of 0.98 V with a fill factor of 64.5 %. Relative to the control, this represents an improvement of approximately an order of magnitude, which can be expressed as
$$\text{Relative enhancement}=\frac{\eta_{2}-\eta_{control}}{\eta_{control}}\times 100\%=\frac{8.4-0.8}{0.8}\times 100\%\approx 950\%$$
Sample 3 is the most interesting case from a teaching standpoint. Its fill factor of 71.1 % is the highest of all four devices, indicating the lowest series resistance and the most favourable charge-transport geometry, yet its short-circuit current density is only 2.9 mA cm⁻², identical to the control. The resulting efficiency of 1.8 % is therefore disappointing despite the excellent fill factor. Students typically assume that a high fill factor necessarily implies a good device, and this sample dismantles that assumption in the most instructive way possible.
| Metric | What it reports | Dominant morphological factor observed here |
|---|---|---|
| VOC | Recombination losses and quasi-Fermi-level splitting | Coverage uniformity and shunt pathways |
| JSC | Photon absorption and charge collection | Film continuity and absorber volume |
| FF | Transport resistance and diode quality | Crystallinity and grain-boundary density |
| η | Combined outcome | Balance between the three factors above |
Optical Microscopy of the Absorber Films
Optical micrographs recorded at 100× magnification provide the first and most immediately comprehensible morphological comparison. The film deposited without antisolvent consists of discrete island-like aggregates separated by extensive bare regions. Light passing through such a film encounters alternating patches of absorber and exposed substrate, and the photocurrent collected from the covered regions is diluted by the inactive area. The measured JSC of 2.9 mA cm⁻² is entirely consistent with this picture.
Films produced with the antisolvent present a far more continuous surface. Coverage is essentially complete, and the residual features are small inclusions rather than macroscopic gaps. When students place the two micrographs side by side and then look again at the corresponding current density–voltage curves, the causal link becomes obvious in a way that no lecture can achieve.
Crystallographic Analysis
X-ray diffraction patterns were collected from all four films. Every sample exhibits the characteristic reflections of the perovskite phase at approximately 14.5° and 28.9° in two theta, confirming that the intended crystal structure formed regardless of the antisolvent route. The differences lie in the intensity and width of those reflections.
Diffraction peak intensity and width carry complementary information. The average crystallite size can be estimated from the Scherrer equation:
$$D=\frac{K\lambda}{\beta\cos\theta}$$
where K is a shape factor usually taken as 0.9, λ is the X-ray wavelength, β is the full width at half maximum of the reflection in radians, and θ is the Bragg angle. Microstrain can be separated from size broadening by plotting βcosθ against sinθ according to
$$\beta\cos\theta=\frac{K\lambda}{D}+4\varepsilon\sin\theta$$
so that the intercept yields the size and the slope yields the strain.
Among the three antisolvent-treated samples, the film prepared by dripping during rotation produced the strongest reflections, indicating the highest degree of crystallinity. The film prepared by immediate soaking produced the weakest. This ranking is the reverse of the ranking observed in device efficiency, and it is precisely the kind of inversion that provokes genuine thinking. Students must reconcile the fact that the most crystalline film yields the lowest current density. The resolution lies in the distinction between crystallinity within individual grains and the connectivity between them: a film may be composed of highly ordered grains that are nevertheless poorly packed, with voids and discontinuities interrupting the percolation pathways that carriers must follow to reach the electrodes.
| Sample | Relative diffraction intensity | Apparent crystallite size | Void characteristics from electron microscopy | Grain size trend |
|---|---|---|---|---|
| Sample 1 | Weakest | Largest | Few voids, but individually large | Large grains |
| Sample 2 | Intermediate | Intermediate | Most uniform and dense, occasional pinholes of small diameter | Intermediate grains |
| Sample 3 | Strongest | Smallest | Highest void count | Small grains |
Scanning Electron Microscopy and the Void Problem
Field-emission scanning electron microscopy resolves the surface morphology at a length scale that optical microscopy cannot reach. Three observations emerge from the micrographs. First, the grains in samples 1 and 2 are somewhat larger than those in sample 3, consistent with the Scherrer analysis and with the expectation that slower, more gradual crystallisation favours grain growth. Second, the film deposited with antisolvent added during rotation contains the largest number of voids, which explains its low photocurrent despite its high crystallinity. Third, the film rested for one minute before antisolvent addition is the most uniform and dense of the three, though it is not entirely free of defects; small pinholes remain, but their diameter is markedly smaller than the voids seen in either of the other two antisolvent-treated films.
The presence of small pinholes in the best film is pedagogically valuable. It prevents students from concluding that a single processing trick solves every problem, and it opens the door to a discussion of what further optimisation would require: perhaps a slightly different antisolvent, a mixed antisolvent system, a controlled antisolvent volume, or a post-treatment passivation step.
A useful quantitative descriptor of film quality is the fractional coverage, which can be estimated by thresholding the micrographs:
$$f_{c}=\frac{A_{film}}{A_{total}}$$
and the resulting loss in photocurrent can be approximated, in the simplest geometric model, by
$$J_{SC}\approx J_{SC,intrinsic}\cdot f_{c}\cdot\eta_{coll}$$
where ηcoll is the charge-collection efficiency. This expression is admittedly crude, but it gives students a way to reason about the relative contributions of geometric coverage and electronic quality to the observed short-circuit current.
Integrating Optical and Electronic Characterisation
A more refined comparison becomes possible when external quantum efficiency spectra are considered alongside the current density–voltage data. The external quantum efficiency at a given wavelength is defined as
$$EQE(\lambda)=\frac{hc}{q\lambda}\cdot\frac{J_{ph}(\lambda)}{P_{in}(\lambda)}$$
and the short-circuit current density that would be expected from an idealised measurement can be obtained by integrating the product of the quantum efficiency and the incident spectral power density:
$$J_{SC}=\frac{q}{hc}\int\lambda\,EQE(\lambda)\,P_{AM1.5}(\lambda)\,d\lambda$$
When the integrated value agrees with the measured JSC to within a few percent, the measurement is internally consistent and the spectral mismatch correction is small. When it does not, either the measurement conditions or the assumed device area are suspect. I require each team to perform this consistency check, because it instils the habit of cross-validating independent measurements rather than trusting a single number.
The optical gap of the absorber can be extracted from absorbance data using the Tauc relation for a direct-gap semiconductor:
$$\left(\alpha h\nu\right)^{2}=A\left(h\nu-E_{g}\right)$$
A plot of (αhν)² against photon energy yields a straight-line region whose intercept gives the gap. Since the absorber composition is unchanged across the four samples, the gap should be nearly identical, and any apparent shift is more likely to reflect scattering from rough or porous films than a genuine change in band structure. Recognising this distinction is an important lesson in not over-interpreting small spectral differences.
Carrier transport quality can also be discussed in terms of the diffusion length,
$$L_{D}=\sqrt{D\tau}$$
where D is the diffusion coefficient and τ the carrier lifetime. A film with excellent crystallinity but a high density of voids may have a long intrinsic lifetime yet a short effective collection length, because carriers generated in isolated grains must cross barriers to be extracted. This is a compact way to express why sample 3 underperforms.
Interpretation and Structure–Property Relationships
Bringing the evidence together, a coherent picture emerges. The introduction of chlorobenzene as an antisolvent substantially improves the growth quality of the perovskite film on the substrate and consequently improves device performance. The precise mode of introduction, however, determines which aspect of film quality is optimised.
When the antisolvent is applied immediately and allowed to soak, nucleation is rapid but the film is still rich in residual host solvent at the moment of contact. The interaction between the two solvents is violent, and the resulting films show large individual voids even though the total void count is moderate. The large voids are particularly damaging because they can span the entire absorber thickness and create direct shunt paths between the transport layers.
When the film is rested for one minute before the antisolvent is applied, a controlled amount of host solvent has already been lost by evaporation. The remaining solution is more concentrated and more viscous, and the antisolvent therefore induces a more gradual and more spatially uniform supersaturation. Nucleation is dense but not explosive, and grain growth proceeds in a controlled manner. The resulting film is the most uniform and dense of the series, with only small pinholes, and the device achieves the highest short-circuit current density, the highest open-circuit voltage, and the highest efficiency.
When the antisolvent is dripped during rotation, the mechanical action of the spinning substrate combined with the sudden solvent extraction promotes very rapid nucleation, producing many small crystallites of high crystallographic order. The high fill factor of 71.1 % confirms that transport within the grains and across grain boundaries is efficient. However, the rapid evolution of gas and the turbulent displacement of solution leave behind a large number of voids, and the resulting loss of active area and of percolation pathways suppresses the photocurrent so severely that the overall efficiency is only 1.8 %.
| Sample | Nucleation character | Resulting morphology | Dominant loss mechanism | Primary metric limited |
|---|---|---|---|---|
| Control | Slow, sparse, heterogeneous | Discontinuous islands with large gaps | Incomplete absorption and poor percolation | JSC, FF, VOC all depressed |
| Sample 1 | Rapid, non-uniform | Moderate coverage, oversized voids | Local shunting through large voids | FF and VOC |
| Sample 2 | Dense and spatially uniform | Dense, uniform, minor pinholes | Residual small defects only | All metrics near optimum |
| Sample 3 | Very rapid, explosive | Highly crystalline, void-rich | Loss of active area and connectivity | JSC |
The general conclusion that grain size and film density govern JSC, VOC, and η, whereas crystallinity governs FF, is a clean and defensible summary of the data. It is also an excellent illustration of the fact that the four photovoltaic parameters are not independent. Optimising one in isolation can degrade another, and the engineering task is to find the processing window that balances them.
Uncertainty, Replication, and Statistical Treatment
One of the most important habits I try to instil is that a single measurement is not a result. Device-to-device variation in perovskite solar cells is substantial, and a difference between two samples is only meaningful if it exceeds the spread within each sample. Each condition was therefore replicated across multiple pixels on multiple substrates, and the parameters are reported as a mean with an associated standard deviation:
$$\bar{x}=\frac{1}{N}\sum_{i=1}^{N}x_{i},\qquad s=\sqrt{\frac{1}{N-1}\sum_{i=1}^{N}\left(x_{i}-\bar{x}\right)^{2}}$$
Students also compute the standard error of the mean,
$$SEM=\frac{s}{\sqrt{N}}$$
and use it to judge whether the observed differences are likely to be reproducible. In practice, the gap between the control and the best antisolvent-treated sample is far larger than any plausible measurement artefact, whereas the gap between samples 1 and 2 is small enough that a cautious investigator would want additional replicates before declaring a definitive winner.
| Condition | Mean efficiency (%) | Spread across pixels | Confidence in ranking |
|---|---|---|---|
| Control | 0.8 | Narrow | High, clearly the weakest |
| Sample 1 | 7.7 | Moderate | High relative to control |
| Sample 2 | 8.4 | Moderate | High relative to control, marginal over sample 1 |
| Sample 3 | 1.8 | Wide | High relative to control |
The wide spread observed in sample 3 deserves comment. Explosive nucleation combined with turbulent solution displacement produces films whose local quality varies considerably across the substrate, so some pixels are passable while others are essentially shorted. This heterogeneity is itself a finding: a processing route that yields high average performance with low variance is far more valuable for manufacturing than one that occasionally produces an excellent device.
Hysteresis and Measurement Protocol
Because inverted architecture was chosen partly to reduce hysteresis, the effect is modest in these devices, but students are still asked to quantify it. A convenient index is
$$HI=\frac{\eta_{reverse}-\eta_{forward}}{\eta_{reverse}}$$
or, equivalently, the same ratio computed from short-circuit current densities. Scanning in both directions and at different scan rates teaches the essential lesson that a reported efficiency is meaningful only when the measurement protocol is specified. A large hysteresis index indicates that ionic redistribution within the perovskite is significant on the measurement timescale, and that the quasi-steady-state efficiency may differ from either sweep value.
I also require a stabilised power output measurement at the maximum power point, in which the device is held at a fixed voltage and the current is monitored over a period of minutes. This is the most honest single number for a photovoltaic device, and comparing it with the sweep-derived efficiency gives students a direct sense of how much optimism a fast sweep can introduce.
Extending the Investigation
Once the core comparison is complete, I invite teams to propose one extension. Over successive iterations of the project, several extensions have proven particularly productive.
| Extension | Scientific question | Additional measurement required |
|---|---|---|
| Antisolvent identity | How does solvent polarity and boiling point affect nucleation density? | Morphology comparison across solvents |
| Antisolvent volume | Is there an optimum volume beyond which performance declines? | Systematic volume series with device statistics |
| Mixed antisolvent | Can a two-component antisolvent combine dense nucleation with slow growth? | Grazing-incidence diffraction and cross-sectional imaging |
| Annealing profile | How does the thermal budget interact with residual solvent? | Time-resolved absorbance and diffraction |
| Interfacial passivation | Can surface defects be healed after film formation? | Photoluminescence lifetime and trap density estimation |
| Stability testing | Does improved morphology translate into improved operational lifetime? | Periodic re-measurement under controlled humidity |
The stability extension is especially valuable because it connects the fabrication work to the central challenge facing perovskite solar cells as a commercial technology. A film with fewer voids and larger grains has less internal surface area exposed to ingress pathways, and it is therefore expected to degrade more slowly. Asking students to design a simple ageing protocol and to interpret the resulting decay curves introduces concepts such as shelf life, burn-in, and the distinction between reversible and irreversible degradation.
Practical Observations from Repeated Delivery
Several practical lessons have accumulated from running this project more than once. Glovebox discipline is the most common early failure point; a single exposure of the precursor solution to ambient humidity can ruin an entire batch, and students learn this quickly once they see a clear solution turn turbid. Substrate cleaning matters more than most beginners expect, and inadequate plasma treatment produces films that dewet during spinning. The antisolvent must be delivered decisively and without hesitation; a slow or tentative drip produces streaked films and inconsistent results. Finally, the time between film formation and the deposition of the electron transport layer should be kept as short as practical, because the freshly formed perovskite surface is reactive.
I have also found it useful to require students to maintain a laboratory notebook in which every parameter is recorded, including those that seem unimportant. When a result is surprising, the notebook is what allows the team to reconstruct what actually happened. In several instances, an anomalous efficiency was traced to an unnoticed change in ambient humidity or in the delay between processing steps, and those episodes taught more about experimental rigour than any formal instruction could.
Learning Outcomes
The transformation in the participants is, in my experience, consistent and pronounced. Students who initially avoided English-language literature begin to read it with the aid of translation tools and, within a few weeks, with growing confidence. Students who had never touched a glovebox or an evaporator become capable of completing an entire fabrication sequence with minimal supervision. Students who initially responded to an unexpected result with silence begin to generate hypotheses, propose controls, and design follow-up experiments.
| Observed change | Indicator before the project | Indicator after the project |
|---|---|---|
| Literature engagement | Avoidance of primary sources | Independent reading and critical summary |
| Instrumental fluency | Unfamiliarity with glovebox and evaporator | Independent operation of the full fabrication line |
| Design thinking | Multiple simultaneous variables | Controlled single-variable comparisons |
| Data reasoning | Reporting a single best value | Reporting means, spread, and confidence |
| Causal argument | Descriptive reporting | Mechanistic explanation linking morphology to metrics |
| Resilience | Discouragement after failure | Systematic troubleshooting and iteration |
The most valuable outcome, in my judgement, is not that students learn a particular fabrication recipe. Recipes become obsolete. What endures is the understanding that a device is the product of a sequence of coupled decisions, that morphology at the nanometre scale determines behaviour at the device scale, and that a trustworthy conclusion requires controls, replication, and an explicit statement of uncertainty.
Concluding Remarks
I have used perovskite solar cells as the vehicle for an open-ended undergraduate project because they compress an entire research cycle into a manageable timescale. Within a single semester, a student team can survey the literature, formulate a hypothesis about how an antisolvent should be delivered, fabricate a complete device stack, measure its photovoltaic response, characterise the absorber at the morphological and crystallographic levels, and reconcile the results into a coherent mechanistic account. The specific finding reported here, that resting the film for one minute before applying chlorobenzene yields the best combination of density, uniformity, and device performance, with an efficiency of 8.4 % compared with 0.8 % for the untreated control, is a useful process insight in its own right. The broader point, however, is that the same experimental structure can be transferred to almost any processing variable in the fabrication of perovskite solar cells, whether that variable concerns the transport layers, the annealing profile, the precursor stoichiometry, or the encapsulation strategy.
For those who wish to adopt a similar approach, I would emphasise three design principles. First, choose a variable whose effect is large enough to be visible above the intrinsic scatter of the device fabrication process; a difference that is smaller than the sample-to-sample spread will frustrate rather than educate. Second, insist on replication and on the reporting of uncertainty; this single requirement changes how students think about evidence. Third, allow the results to be genuinely unknown. The moment students realise that nobody in the room, including the instructor, knows exactly what the next data point will show, the project stops being an exercise and becomes an investigation. That shift is where the real learning happens, and it is why I continue to build my teaching around open-ended work on perovskite solar cells.