A 1MW perovskite array outyielded a co-sited 3.5MW TOPCon plant, per watt, for three months

Nanjing University says it installed a 1 MW perovskite photovoltaic system and a 3.5 MW crystalline silicon TOPCon system in the same large-scale ground-mounted power plant and monitored both for three months ^[1]. The perovskite side generated 3.42 percent more daily electricity per unit of installed capacity in March, 3.79 percent more in April and 5.81 percent more in May ^[2].

The metric is the interesting part. Perovskite has been sold on standardised test numbers, and this work has one of those too: modules of 0.72 square metres produced 158.4 watts of steady output, certified at 22.0 percent full-area efficiency, which the team describes as a record for meter-scale perovskite hardware ^[3]. That works out to 220 watts per square metre of aperture ^[1]. But energy delivered per installed watt over a season is the quantity a yield model and a PPA actually settle on, and that is what the co-sited test produced.

Two caveats sit inside the comparison. It is normalised to nameplate, and the perovskite array is 28.6 percent the size of the silicon one ^[2], so the silicon plant still made far more electricity in absolute terms. And normalising to capacity says nothing about land per megawatt-hour, which is where a lower-efficiency module gets expensive. The gap also widened across the window, by 2.39 percentage points from March to May ^[3]; Interesting Engineering reports the lead grew as spring temperatures rose ^[4], which points at temperature behaviour rather than a fixed advantage.

The manufacturing change behind it was published in Nature on August 12, 2026, led by Xiao Ke and Hairen Tan at Nanjing University with engineers from Renshine ^[5]. The commercial pitch for perovskite has always been process cost, since the material can be printed from liquid inks rather than refined like silicon ^[6], but scale-up loses power to surface flaws that trap charge; the group's earlier Science work grew flat films over large areas and still capped out at 17.2 percent ^[7]. The standard repair, ammonium halide salts, works in a lab and fights the factory: the salts degrade in humid air, forcing inert-gas production lines, and they coat large sheets unevenly ^[8].

The fix is a three-solvent blend of 2-methoxyethanol, 1,3-dioxolane and dimethyl sulfoxide that controls drying in a vacuum chamber and drives a formamidinium iodide layer to the top surface ^[9], followed by organic lead-carboxylate salts that coat the full sheet evenly and block moisture without an inert atmosphere ^[10]. Removing an inert-gas line from the process flow is a capex claim, not a physics claim, and it is the part a factory buyer should price.

Durability was screened rather than proven. At 85 degrees Celsius and 85 percent relative humidity for 1,300 hours, roughly 54 days ^[4], ammonium-coated modules lost 39 percent of power and the lead-treated ones lost 2 percent, a 37 point spread ^[11][5]. The modules also completed 300 cycles between minus 40 and 85 degrees Celsius with no measurable loss ^[12].

What to watch: whether the field comparison continues past spring into summer and winter, whether anyone publishes a field degradation rate rather than a chamber result, and whether the solvent and lead-carboxylate steps survive production volumes at Renshine. Bankability is a degradation curve and a warranty counterparty, and neither exists yet.