The headline number is 29.0%: the maximum external quantum efficiency of a deep-blue phosphorescent OLED built around a new iridium complex called CF3-2. The number that makes it interesting is 1.28 × 10⁶ per second: the complex’s radiative decay rate, the speed at which its triplet excitons turn into photons. Deep-blue OLEDs have spent two decades failing on exactly the combination of those two numbers — efficient or long-lived, rarely both — because the triplets that make phosphorescence possible also linger long enough to destroy each other and the molecules around them. A team from Tsinghua Shenzhen, Kyoto University, and Kyushu University reports in Light: Science & Applications that a trifluoromethyl (CF3) group on one ligand can steer the triplet state toward charge-transfer character and make it emit faster. The result is not one device that wins everything: in a family of six complexes, the efficiency prize and the lifetime prize go to different siblings.
The paper is “High-efficiency and stable deep-blue iridium phosphorescent OLEDs with enhanced charge transfer dynamics,” by Siqi Li, Kai-Ning Tong, Meng Zhang, and colleagues, with Chengcheng Wu, Feiyu Kang, Chihaya Adachi, Takuji Hatakeyama, and Guodan Wei as corresponding authors, in Light: Science & Applications (2026) 15:259, DOI 10.1038/s41377-026-02264-y. A vendor summary from LUMORA — a materials supplier whose catalogue includes some of the stack materials used in the paper — circulated the results in October 2026. Everything quantitative below comes from the paper itself, not the vendor’s telling of it.
Why deep blue is the hard color
A phosphorescent OLED harvests both singlets and triplets, formed in a roughly one-to-three ratio when electrons and holes meet. A fluorescent emitter throws the triplets away; a phosphorescent emitter, typically built around iridium, uses spin-orbit coupling to let triplets emit, so in principle all excitons contribute. Red and green phosphors have delivered on that promise for years. Deep blue has not.
Blue light means high-energy excitons, around 2.8 electron volts for a 443-nanometer photon. High-energy triplets are chemically aggressive: they collide with each other in triplet-triplet annihilation, quench on charges and defects, and break the bonds of the host and transport materials around them. The established deep-blue reference emitters the paper cites, FIrpic and mer-Ir(tfpi_tmBn)₃, reach CIE coordinates around (0.16, 0.08) but suffer maximum external quantum efficiencies below 15% and severe efficiency roll-off at practical luminance levels.
The group’s answer is architectural before it is chemical. Most iridium phosphors use symmetric [2+2+2] coordination: three bidentate ligands around the metal. This group instead uses an asymmetric [3+2+1] arrangement — a rigid tridentate N-heterocyclic carbene (NHC) ligand, a bidentate carbon-carbon ligand, and a cyanide — a geometry it has developed across several earlier papers, with a prior record external quantum efficiency of 30.7% in deep blue. The NHC ligand’s strong electron-donating character and steric bulk raise the ligand field, suppressing non-radiative decay through metal-centered states and widening the HOMO-LUMO gap that sets the blue emission. The new paper asks what happens when the bidentate ligand is decorated with strongly electron-withdrawing groups: CF3 or fluorine.
One substituent, two kinds of triplet
The team synthesized six complexes, CF3-1, CF3-2, and CF3-3 alongside F-1, F-2, and F-3, all of the form [(pbib)Ir(C^C)CN], where pbib is the tridentate NHC ligand and the C^C ligand carries the CF3 or F substituent. In toluene at room temperature the six emit across 416 to 456 nanometers, all in the deep-blue range. What differs is not the color but the character of the triplet state doing the emitting.
Natural transition orbital calculations show the split cleanly: the CF3 complexes have predominantly charge-transfer (CT) excited states, roughly 94.5 to 95.0% CT character, while the fluorine complexes are dominated by locally excited (LE) states. The solvent test is the tell — CF3-2’s emission shifts from 471 nanometers in dimethylformamide to 437 nanometers in toluene as polarity falls, the signature of a charge-transfer state, whereas F-3 moves only 14 nanometers. A CF3 group, bulkier and more strongly electron-withdrawing than a single fluorine atom, also pushes neighboring molecules apart: crystal structures put adjacent tridentate ligands 9.035 angstroms apart in F-2 but 11.637 angstroms apart in CF3-3, weakening pi-pi stacking and suppressing triplet-triplet annihilation and concentration quenching.
Charge-transfer character matters because the radiative rate scales with the square of the transition dipole moment between the triplet and ground states. The calculated transition dipole moments run 0.60, 0.74, and 0.55 Debye for CF3-1, CF3-2, and CF3-3, against 0.45, 0.37, and 0.29 Debye for F-1, F-2, and F-3. CF3-2, with the largest transition dipole moment in the series, has the fastest radiative rate: 1.28 × 10⁶ s⁻¹ in a film of 12.5 weight percent CF3-2 in the CzSi host, with a photoluminescence quantum yield of 83.8% and an excited-state lifetime of 0.66 microseconds. The fluorine analogues are slower, with radiative rates from 0.54 × 10⁶ to 1.11 × 10⁶ s⁻¹ and quantum yields as low as 51.0%. Faster emission is not a vanity metric: a triplet that emits in 0.66 microseconds spends less time exposed to annihilation and quenching than one that lingers for over a microsecond, and at display brightness, where triplet density is high, that exposure is where efficiency goes to die.
The devices, in full
The direct-emission devices use a vacuum-evaporated stack: ITO / HAT-CN (10 nm) / TAPC (40 nm) / TCTA (10 nm) / CzSi (5 nm) / CzSi doped with 12.5% iridium complex (20 nm) / DPEPO (5 nm) / TSPO1 (40 nm) / Liq (2 nm) / aluminum (120 nm). CzSi was chosen as host for its high triplet energy and thermal stability. This is a laboratory stack, not a manufacturing recipe — every layer is vacuum-deposited — but its conventionality is useful: differences between devices are attributable to the emitters, not to exotic architectures.
CF3-2 is the efficiency device. Its direct device, D-CF3-2, peaks at 443 nanometers with CIE coordinates (0.147, 0.089), close to the NTSC deep-blue standard of (0.149, 0.085), and reaches a maximum external quantum efficiency of 29.0%, with a maximum current efficiency of 17.7 candelas per ampere and a maximum power efficiency of 14.4 lumens per watt. Its siblings are close behind on peak efficiency — D-CF3-1 at 24.6% and D-CF3-3 at 23.0% — and all three CF3 devices beat all three fluorine devices, which top out at 16.4%.
But the CF3-2 device’s brightness ceiling is low. Its maximum luminance is just 1,554 cd/m², against 6,542 cd/m² for D-CF3-1, and its efficiency falls to 14.3% at 10 mA/cm² — roughly half its peak. The roll-off the paper set out to beat is still very much present in the direct-emission device. The 29.0% figure is a peak at low current, and any account that reports it without the 1,554 cd/m² ceiling and the halved efficiency at 10 mA/cm² is misleading about what this device can do in a display.
In a hyper-OLED, the phosphor is not the emitter at all: it acts as a sensitizer, harvesting triplets and transferring their energy to a narrowband thermally activated delayed fluorescence (TADF) terminal emitter that produces the final light. With CF3-1 sensitizing the TADF emitter ν-DABNA, the hyper-OLED H-CF3-1 emits at 472 nanometers with a narrow 20-nanometer linewidth, CIE coordinates (0.120, 0.120), and a maximum external quantum efficiency of 23.4%. With CF3-2 sensitizing DOB2-DABNA-A, H-CF3-2 reaches a deeper blue, 452 nanometers, CIE (0.146, 0.067), 28-nanometer linewidth, and 14.3% maximum efficiency — and, notably, the paper reports minimal efficiency roll-off for this device, holding 13.8% at 10 mA/cm².
The lifetimes are where the split becomes stark. A control device using CF3-1 as the direct emitter without any sensitizer, L-CF3-1, manages only 8.5% maximum efficiency but lasts 3,875 hours to half its initial luminance (LT50) at 100 cd/m², with CIE coordinates (0.132, 0.131). That is the paper’s evidence for CF3-1’s intrinsic stability: the same complex that sensitizes a 23.4%-efficient hyper-OLED also survives nearly 4,000 hours on its own. The hyper-OLEDs themselves last 2,127 hours (H-CF3-1) and 373 hours (H-CF3-2) at 100 cd/m². So the honest scorecard is three-way: best peak efficiency (29.0%, CF3-2 direct) with the lowest brightness ceiling; best lifetime (3,875 hours, CF3-1 control) with the lowest efficiency; and best combination of narrowband color, decent efficiency, and long life (23.4%, 20-nanometer linewidth, 2,127 hours) in a device where CF3-1 never emits a photon itself.
A microdisplay, not a television
The paper closes with a demonstration rather than a product claim: top-emitting CF3-2 OLEDs on a thin-film transistor array, a 24-millimeter-diagonal panel with 270 × 270-micrometer pixels at 94 pixels per inch, displaying programmable patterns spelling “Tsinghua.” At 94 PPI this is a proof of integration, not a competitive microdisplay, and what it establishes is that the emitter survives integration with a TFT backplane and patterned driving.
That distinction — materials result versus display product — matters when this work circulates. The vendor summary that brought it to attention sells several stack materials, including the CzSi host and the ν-DABNA terminal emitter, for research use. That does not taint the paper, whose authors declare no competing interests, but the 29.0% device and the 3,875-hour device are different devices, built from different complexes, in different architectures — not one device’s properties.
What it is not — yet
First, brightness. A display pixel routinely operates at hundreds of candelas per square meter, and the CF3-2 direct device’s 1,554 cd/m² ceiling, with efficiency halved by 10 mA/cm², leaves little headroom for the losses encapsulation and a real backplane add. The hyper-OLEDs reach higher luminances — 6,474 cd/m² for H-CF3-1 and 4,093 cd/m² for H-CF3-2 — but at the cost of color (472 nanometers is blue, not deep blue, for H-CF3-1) or lifetime (373 hours for H-CF3-2).
Second, lifetime at display conditions. LT50 figures quoted at 100 cd/m² are the field’s convention, but a phone or headset display does not live at 100 cd/m², and lifetime falls steeply as initial luminance rises — the paper’s own table notes the extrapolation involved, with an acceleration exponent of 1.836. A 3,875-hour figure at 100 cd/m² is not a 3,875-hour display.
Third, everything here is a laboratory device: vacuum-evaporated, measured as single pixels, with no data on batch-to-batch variation, no cost analysis for iridium — a scarce metal whose price and supply chain constrain phosphorescent OLEDs generally — and no panel-level demonstration beyond the 94-PPI test array. The paper’s contribution is a design rule, not a supply chain.
None of this diminishes the design rule, and the design rule is genuinely useful. It says: in this asymmetric [3+2+1] family, the substituent on the bidentate ligand chooses the triplet’s character, the triplet’s character sets the transition dipole moment, the transition dipole moment sets the radiative rate, and the radiative rate tracks device efficiency. That chain, supported by calculation, spectroscopy, and devices, gives the next group a knob to turn: if your blue phosphor is too slow, make its triplet more charge-transfer-like, and if your charge-transfer emitter aggregates, put something bulky on it. The fluorine series is the control group that makes the claim believable — same scaffold, same host, same stack, slower and dimmer, because its triplets stay local.
Why this trade is the one to watch
Deep blue is the bottleneck for phosphorescent OLED displays because it is where the physics fights hardest: the highest exciton energy, the longest-lived triplets, the fastest degradation. Commercial displays have coped by compromise — fluorescent blue emitters that waste triplets but survive, or phosphorescent blue with moderated color and lifetime. A result that moves both efficiency and lifetime, even in different devices from the same family, narrows the compromise. The hyper-OLED result is arguably the more commercially relevant half: a sensitizer architecture already used in industry, a narrowband terminal emitter for color purity, and a lifetime above 2,000 hours at 100 cd/m² with 23.4% efficiency is a combination a display maker can design around, even if the direct-emission record grabs the headline.
The signals to watch: a single device holding efficiency above 20% at luminances a display actually uses; a lifetime measurement at higher initial luminance without extrapolation doing the heavy lifting; and whether the CF3 design rule transfers to other scaffolds or remains a property of this one asymmetric family.
For now, the fair summary is this: one fluorine-bearing group made a deep-blue triplet faster, and faster triplets made a better device — twice over, in two different devices, neither of which is yet the device displays need. The efficiency-lifetime trade has not been beaten. It has been mapped more precisely, with a mechanism attached, and that is how trades like this eventually get beaten.
Sources and further reading
Li, S., Tong, K.-N., Zhang, M., He, W., Wu, C., Ochi, J., Wu, D., Shi, K., Wang, X., Jung, S., Kang, F., Adachi, C., Hatakeyama, T., and Wei, G. “High-efficiency and stable deep-blue iridium phosphorescent OLEDs with enhanced charge transfer dynamics.” Light: Science & Applications 15, 259 (2026). DOI: 10.1038/s41377-026-02264-y. All device figures in this article are from this paper, including its Table 4 device summary. The LUMORA vendor summary (October 2026) provided the initial pointer to the paper; its catalogue relationship to stack materials is noted in the article as context, not relied on for any figure.