Three years earlier, 趙清煙 had spoken about perovskite materials at the same forum. At the time, inverted devices still lagged behind the conventional architecture in efficiency. On returning, he began with what had changed: inverted devices had caught up, some records had moved ahead, and lifetime performance had improved.

He then moved through one high-efficiency paper after another, repeatedly asking what the researchers had actually changed. As the efficiency gap narrowed, the answers shifted from the light-absorbing formulation toward the contacts through which charge leaves the absorber.

A perovskite solar cell is a stack of thin layers. The absorber in the middle converts light into charge; contact layers above and below extract holes and electrons. An inverted, or p–i–n, architecture reverses the order of those contacts. How the bottom contact is laid down, how surface defects are treated and whether materials move during operation all appear in voltage and lifetime.

趙清煙 began on the hole-transport side. The talk placed the transport material beside changes in film structure and emission after heating. A contact layer affects how efficiently charge can leave, while also influencing what happens to the film under heat and illumination.

He then moved to the upper surface of the perovskite. If atoms at the crystal surface are not fully bonded, charge can be lost there and mobile ions can interact with those sites. Differences in emission, voltage and forward–reverse scans before and after defect treatment can therefore reflect not only a newly fabricated device but also ion movement during operation.

Below that lies the interface between the absorber and the bottom contact. Inverted devices commonly coat the bottom with a self-assembling layer only about one molecule thick. If the molecules aggregate before reaching the surface, gaps remain where contact is needed. During aging, grain boundaries may also provide pathways for them to move upward. Charge must leave through this interface, so uneven contact becomes another source of loss.

A 2026 Nature Communications paper separated this problem more precisely. The researchers added a bulky phenyl group to one side of the molecule, preventing the otherwise flatter carbazole cores from stacking face to face and reducing aggregation. A second, crosslinkable component moved along grain boundaries into the film and fixed those migration pathways at low temperature.

Devices in the study achieved a certified efficiency of 27.03%. After 2,000 hours of continuous maximum-power-point tracking at 65°C, they retained more than 96% of their initial efficiency. The coverage problem described by 趙清煙 acquires a specific location in this case: the molecule must not only enter the formulation; it must remain at the bottom after the film forms.

趙清煙 did not stop at one molecule. Later examples added anchoring sites, allowed two molecules to organize together, or crosslinked the contact layer in place. Some approaches improve initial coverage, some divide functions between two molecules, and others ask whether the contact loosens under heat.

At the supply end, the task is more than adding a new molecule to a catalogue. A research team designs a material for a particular interface; a supplier has to synthesize, purify and analyze it repeatedly while maintaining quality between batches. Lumtec's public company page lists synthesis, purification, sublimation and materials analysis, along with 50–500 L reactors and company-wide synthesis capacity of one metric ton per month. That figure covers Lumtec's overall materials business and cannot be treated as monthly production of perovskite materials.

Lumtec formally entered the talk only at the end. The aggregation, gaps at the bottom interface and movement under heat described earlier become practical supply requirements: can one batch be purified to the same degree, will the next remain consistent, and can a scaled material still solve the original interface problem? The costs, lifetime and process integration left by 趙清煙 at the close all rest on that work.

Sources and further reading

Talk and industry context

Primary research extending the interface problem