KAIST PAN Film Cuts Crude Refining Energy by 31.6%

A bare polyacrylonitrile membrane cut modeled crude distillation energy by 31.6% after the oil itself narrowed the film’s pores into a sieve. Researchers at KAIST and Georgia Tech published the result in Nature on June 24, 2026, using a cheap support polymer the field had treated as too open to sort crude. Process models also put carbon dioxide down 37.6% and cooling water down 20.7% against a standard atmospheric tower.

The energy math is a computer model of a retrofit, not a meter on a live crude unit. The lab fact underneath it is stranger, and it is the part the 31.6% headline skips: the film works because it clogs.

A Support Film Turns Itself Into a Sieve

Polyacrylonitrile, or PAN, is the porous backing in many industrial filters, not the selective skin. Crude molecules are small, and a PAN sheet starts with surface mesopores about 15 nanometers across, wide enough that nobody expected it to sort naphtha from residue on its own.

Under tangential flow, heavy hydrocarbons built up inside those pores. Depth-resolved chemistry in the Nature paper found n-alkanes piling up until the openings shrank to sub-2-nm channels, then stopping. Lighter cuts such as naphtha and kerosene kept moving. The heavies that formed the lining were, after that point, largely kept out.

In water plants, that deposit would be fouling, and operators would clean or replace the module. Here the deposit is the filter. Once it stabilized, the team ran raw crude for 4 weeks without the selectivity collapsing. Permeance reached 0.591 liters per square meter per hour per bar, more than 23 times the previous whole-crude benchmark, which sat under 0.1 in the same units.

This study reveals a new scientific principle in which a membrane interacts with a complex mixture and spontaneously forms its own separation channels. Working with real crude oil supplied by HD Hyundai Oilbank allowed us to validate the technology under conditions relevant to industrial operation.

Dong-Yeun Koh, associate professor, KAIST

KAIST’s public account put the same claim in one line: the oil writes its own molecular pathways, and the energy cut in the model is 31.6%.

Crude oil, no boiling required? 🚀 KAIST and Georgia Tech developed a room-temperature membrane that lets crude oil create its own molecular sieving pathways, cutting refining energy use by 31.6%. #KAIST #Professor Dong-Yeun Koh #EnergyTech #MembraneTechnology #Sustainability pic.twitter.com/J5Ry3CUffv

— KAISTPR (@kaistpr) June 29, 2026

Why Earlier Crude Membranes Stayed in the Lab

Custom coatings could split hydrocarbons in a flask. They could not move enough oil to interest a refinery, and the ultrathin skins that make those coatings selective are hard to make large without defects. That throughput gap, not a lack of climate language, is why the idea stayed on the bench after 2020.

Ryan Lively, the Thomas C. DeLoach Jr. Endowed Professor at Georgia Tech and a corresponding author, advised the KAIST group. Koh led the study after a postdoc in Lively’s lab. Six years after their first whole-crude membrane paper, Lively said the open problem was still how little product those units made.

One of the key challenges that the KAIST team set out to tackle was the very low oil productivities of the membrane units, which has limited the ability of this concept to leave the lab. Along the way, we not only increased the productivities, but we also uncovered a surprising new mechanism that could make membrane-based crude oil separations even more practical.

Ryan Lively, professor, Georgia Tech School of Chemical and Biomolecular Engineering

FROM CUSTOM POLYMERS TO A BARE SUPPORT

1. July 2020: A Science paper from Lively’s circle shows N-aryl-linked spirocyclic polymers can enrich molecules lighter than about 170 daltons from light crude, with whole-crude permeance under 0.1 liters per square meter per hour per bar.
2. 2023: Designer “DUCKY” polymers in Nature Materials still rely on new chemistry rather than a stock support, aiming at the same heat-and-water bill in the atmospheric tower.
3. June 24, 2026: Jihoon Choi and Hyeokjun Seo, co-first authors at KAIST and KRICT, report that a bare PAN support, with no selective coating, holds a 4-week crude run and lifts permeance more than 23-fold.

Andrew Livingston, vice president of research at Queen Mary University of London and not on the paper, called it work that rewards curiosity. He noted that it goes after heavy hydrocarbons, where most membrane papers have stayed with lighter oils, and that it uses a simple film already on the market.

The Model Leaves the Distillation Tower in Place

Atmospheric and vacuum distillation still burn more than 1,100 TWh a year and emit more than 160 million metric tonnes of CO2 equivalent, according to the paper’s framing. KAIST compared that load to about 130 gigawatt-scale nuclear plants running all year. Georgia Tech compared the same 1,100 TWh to power for 100 million U.S. homes for a year.

A working tower does not vanish in this scheme. Atmospheric distillation units separate fractions by boiling point after the crude has already gone through hot furnaces, with lights rising and heavies sinking. KAIST notes those furnaces push the oil above 350 C. The PAN step runs at room temperature upstream of that heat.

The flowsheet splits the crude into two streams before the furnace. A lighter cut, enriched in naphtha, gasoline, and kerosene, still goes to conventional firing and the column. A heavier cut skips the furnace and is distilled on its own. Less mass is heated to the top temperature, which is where the model harvests its energy, water, and carbon cuts.

MODELED CUTS AGAINST A STANDARD TOWER

Those four lines come from process simulations of PAN pre-fractionation and the team’s cost model, not from a billed utility invoice. The dollar pair is a 36% drop, matching the operating-cost cut KAIST issued with the paper. Georgia Tech also said the same membranes could incrementally increase refinery capacity without building another giant column, which existing hardware cannot do in small steps.

HD Hyundai Oilbank Already Shares the Author Line

Woong-Chul Shin of HD Hyundai Oilbank is a co-author. The acknowledgements thank the refiner for supplying and characterizing the oils. Koh’s KAIST listing also includes the Aramco-KAIST CO2 Management Center. The first industrial reader of this paper is already in the byline, not waiting on a press tour.

HD Hyundai Oilbank states a 690,000 barrels of crude oil per day refining capacity, with more than 2,500 stations in Korea and 2025 revenue of 28.0249 trillion won. That is a full atmospheric system, not a bench loop. The crudes in the study, tagged AXL and AL in the public dataset, are the company’s oil, run in both dead-end and cross-flow cells.

The authors also compared PAN against commercial organic-solvent films already sold for other jobs, including Puramem Selective, oNF-2, and Puramem Flux. The permeance data for PAN and commercial films sit on Figshare with the simulated distillation curves and GC×GC traces. A supplement clip shows a 40-bar cross-flow run as the permeate lightens over time.

South Korea’s Ministry of Science and ICT put the work out on June 25, 2026, naming Shin alongside Jae W. Lee of KAIST and Lively. The ministry did not announce a unit on a live pipestill.

What a 31.6% Cut Would Mean at Plant Scale

If every Korean refinery and petrochemical site adopted the pre-cut, KAIST put the greenhouse-gas drop at about 10 million tonnes a year, on the order of 4 million cars. That figure is an all-in national scaling of the model, not a measured stack test. Georgia Tech ran a separate scaling on U.S. atmospheric capacity of about 18 million barrels a day and translated the same class of savings into power for about 2.2 million homes, about 3 million passenger vehicles, and water for about 660,000 people each year.

IF THE MODEL WERE SCALED

• Korea, KAIST: About 10 million tonnes of greenhouse gases a year, compared with 4 million cars, if the pre-cut ran across the country’s refining and petrochemical sector.
• United States, Georgia Tech: About 18 million barrels a day of atmospheric capacity, with the modeled cuts compared to 2.2 million homes, 3 million vehicles, and water for 660,000 people.
• Single-plant cost book: Annual operating cost in the team’s model falls from $140.3 million to $89.8 million.
• Global tower load: More than 1,100 TWh a year and more than 160 million metric tonnes of CO2 equivalent still sit on atmospheric and vacuum distillation before any membrane is installed.

None of those translations is a purchase order. They are what the published percentages do if someone multiplies them by a nation’s stills. Lively’s stronger industrial claim is narrower: the PAN flux is finally high enough that a refiner should look at a module, and a module can bolt onto pipe that already exists.

Room-Temperature Flow Still Has to Survive a Refinery

The 4-week run is long for a journal figure and short for a crude unit, which is expected to hold a campaign for months. Choi and Seo want to control the pore-narrowing on purpose rather than wait for it to happen. Jae W. Lee, a co-corresponding author, said the next work is large-area modules and long-term reliability.

Georgia Tech reported similar deposit-and-sieve behavior in a second membrane chemistry, which hints the trick is not unique to PAN. It also means the team has not yet named a single commercial SKU a buyer can order by catalog number for crude service.

WHAT STILL HAS TO BE PROVED ON A UNIT

• Area: Sheets and modules large enough for refinery pipe, without the defects that killed coated films at scale.
• Time: Stable flux and cut points well beyond 4 weeks, through shutdowns, crude switches, and cleaning cycles.
• Crudes: Performance off AXL and AL, including heavier, sourer, and high-asphaltene barrels a Korean or U.S. still actually runs.
• Stage cut: Enough light product from the film that the furnace truly sees a smaller load, matching the modeled split.
• The tower: The column and its fired heater remain in the flowsheet; the membrane is a pre-cut, not a spare still.

Through the first week of October 2026, neither KAIST nor Georgia Tech had announced a module on a live atmospheric unit. Official posts in late June restated the paper and then went quiet. That silence is the adoption test Lively invited when he said industry should consider the technology.

Plastic Pyrolysis Oil Is the Next Feed They Named

Choi and Seo said they want the same pore-constriction trick across the rest of the refinery, then on plastic-recycling oil, biofuel cleanup, and other carbon-cutting chemical work. KAIST also listed pyrolysis oil from waste plastic, solvent recovery in battery plants, and drug purification as targets. Those feeds are messy in the same way crude is messy: many molecule sizes, a few that stick, and a strong reason not to boil the whole tank.

If the deposit layer can be steered, a stock PAN roll becomes a platform instead of a one-off crude gadget. If it cannot, each new mixture is another 4-week bet on whatever happens to stick in the pores. That is the live engineering argument, and it is more specific than a generic race to replace distillation.

The paper’s own close is already a hybrid. Lights still go through the furnace. Heavies still go to a column. The cheap film’s job is to stop the still from heating barrels it does not need to heat. Until a 690,000-barrel plant puts that film on a header, the 31.6% remains a model of a clog that, for 4 weeks in Daejeon, behaved like a sieve.