Industries / Energy, water and waste / Batteries and energy storage
Cell manufacturing
Cell manufacturing turns the materials into finished cells: mixing them into a paste, coating it on metal foil, drying and pressing the foil, winding or stacking the layers, filling the cell with electrolyte, sealing it, and charging it for the first time under control.
6
Milestones achieved, from 1887 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1887
A paste instead of liquid
Carl Gassner Jr. of Mainz, Germany, received US patent 373,064 for a cell whose electrolyte is a paste of zinc oxide, sal ammoniac (ammonium chloride), plaster and water that "after awhile becomes comparatively hard and solid". The patent says the zinc oxide makes the paste porous, which lets gases move through it and reduces polarization, the fall in voltage as a cell works. The patent was filed on 20 May and granted on 15 November 1887.
Credit
Carl Gassner Jr
1896
Batteries become consumer goods
The National Carbon Company of Cleveland, Ohio, introduced the Columbia, a paper-lined, six-inch, 1.5 V dry cell. The American Chemical Society, which named it a National Historic Chemical Landmark in 2005, calls it the first battery marketed for consumer use: dry cells had been made in Europe before, but the Columbia did not spill, did not break easily and was cheap to produce. Its carbon-zinc chemistry was the basis of all dry cells for the next 60 years.
Credit
National Carbon Company, a predecessor of Energizer
1991
Lithium-ion goes on sale
Sony's battery plant in Koriyama, Japan, then run by Sony Energytec, began making lithium-ion batteries in 1991, by the history of Murata, which owns the plant today. The Nobel committee writes that in 1991 "a major Japanese electronics company" started selling the first lithium-ion batteries, after which mobile phones shrank and computers became portable.
Credit
Sony Energytec (Sony)
2017
Cells by the gigawatt-hour
In January 2017 Tesla and Panasonic began mass production of cylindrical "2170" cells, 21 mm across and 70 mm long, at the Gigafactory outside Reno, Nevada, while the building was less than 30% complete. Tesla said the plant would make 35 GWh of cells a year by 2018, "nearly as much as the rest of the entire world's battery production combined". By the IEA's count, Panasonic still supplied almost half of the US electric car battery market in 2024.
Credit
Tesla and Panasonic
2023
A dry-made anode confirmed
Most electrodes are coated as a wet paste, and the solvent is then dried off; a dry process makes the electrode from powders with no solvent. Researchers at the Technical University of Munich took apart a first-generation Tesla 4680 cell (46 mm across, 80 mm tall) and found that its anode was made and coated without solvent. They also found a cathode of NMC811, a nickel-rich NMC, and a pure graphite anode, and published all their measurement data openly.
Credit
Tesla (cell); Manuel Ank and colleagues, Technical University of Munich (teardown)
2025
Northvolt goes bankrupt
The Swedish cell maker Northvolt filed for bankruptcy in Sweden on 12 March 2025, after a US Chapter 11 restructuring failed to secure financing. Besides rising capital costs and shifts in demand, it cited "significant internal challenges in its ramp-up of production", and said output from its production lines in Skellefteå had doubled and its production yield had improved by 50% since September 2024. The IEA wrote that Northvolt "faced serious difficulties in scaling up production, leading to insufficient manufacturing yield and high production costs".
LessonA new cell plant has to raise its yield, the share of cells good enough to sell, before its money runs out.
Credit
Northvolt (company); IEA (assessment)
Watch for
Dry-made cathodes at scale
Tesla says it now makes "dry-electrode for 4680 cells with both anode and cathode made in Austin", Texas, where it lists 40 GWh a year of installed 4680 capacity, and that it has begun building packs from these cells for certain Model Y cars. The Munich teardown found a dry-made anode only, in first-generation cells, and no outside teardown of a dry-made cathode was found.
Achieved whenan outside teardown of a current 4680 cell shows a cathode made without solvent.
Claimed by
Tesla, January 2026
Northvolt's plant restarts
Lyten, a battery company based in Silicon Valley, completed its purchase of Northvolt's plant in Skellefteå and its research labs in Västerås in February 2026, with 16 GWh of existing capacity. Lyten says the plant will make NMC cells again, restarting one production line at a time, and deliver commercial cells in the second half of 2026, first to Lyten's own battery plant in Gdańsk, Poland. No report of deliveries was found by mid-November 2026.
Achieved whena customer outside Lyten confirms deliveries of cells made at Skellefteå, or an outside party reports the plant's output.
Claimed by
Lyten, February 2026
Open
01
Defect finders across machines
A test of whether electrode-defect finders trained on one public image set still work on the other, and on a third small set made by hand. Map the two sets' defect labels onto one shared list, train on each set and score on the other. Then spread a water-based graphite paste (graphite, a little carbon black, and the binders CMC and SBR) on copper foil, make defects on purpose (air bubbles, specks of dirt, blade scratches, cracks from drying too fast), photograph and label them, and score both sets' models on these images too.
The briefCredit · unclaimed02
Dry films made twice
An open recipe for making dry-processed electrode films at lab scale, and a test of whether a second team following it gets the same films. Mix graphite or LFP powder, carbon black and PTFE binder powder; shear the mix so the PTFE stretches into fine fibers (fibrillation) that hold the powder together; then roll it into a free-standing film. Publish every setting (mixing time and speed, temperature, roll gap, number of passes) and each film's thickness, density, tensile strength and electrical resistance.
The briefCredit · unclaimed
Open 01 · the brief
Defect finders across machines
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of whether electrode-defect finders trained on one public image set still work on the other, and on a third small set made by hand. Map the two sets' defect labels onto one shared list, train on each set and score on the other. Then spread a water-based graphite paste (graphite, a little carbon black, and the binders CMC and SBR) on copper foil, make defects on purpose (air bubbles, specks of dirt, blade scratches, cracks from drying too fast), photograph and label them, and score both sets' models on these images too.
Why it is still open
Two public sets now exist, from different machines. CoatingVision from Argonne National Laboratory (Scientific Data, February 2026) holds over 2,200 images of lab slot-die coatings labeled for surface cracks, delamination cracks, pinholes and unclassified defects. LIBAD, posted in August 2026 by Wenbo Sui and colleagues, holds patches from real roll-to-roll production lines with matched visible-light and X-ray images; its authors found that existing methods carried over poorly and gave high false-alarm rates even within their own set. Projects on GitHub train and test within one set, such as BatteryVision on CoatingVision. A search found no test across the two sets.
Built with what exists
The CoatingVision images and code; LIBAD's data on Hugging Face (free after accepting its conditions, CC BY 4.0) and its open BSD-licensed code; PatchCore-style anomaly detection, Ultralytics YOLO or a U-Net; a small vision model fine-tuned on one graphics chip, such as Liquid AI's LFM3 (see the Note "One GPU to fine-tune"); for the hand-made set, a doctor-blade film applicator (a hand-pulled blade set to a fixed gap), battery-grade graphite, CMC and SBR, copper foil and a lab oven, with a dust mask for the powders, a USB microscope or a phone camera with a macro lens, and Label Studio. The paste holds no lithium and no cells are built.
First step
Download both sets, write the label mapping, and score one model trained on CoatingVision on LIBAD's visible-light images.
Done looks like
A published label mapping between the two sets; a table of each model's per-class F1 score (a measure that balances missed defects against false alarms) on its own set, on the other set and on the hand-made set, so the drop between machines can be seen; at least 500 labeled hand-made images released under an open license, with a second labeler's agreement reported on 100 of them; code to reproduce it all.
Open 02 · the brief
Dry films made twice
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open recipe for making dry-processed electrode films at lab scale, and a test of whether a second team following it gets the same films. Mix graphite or LFP powder, carbon black and PTFE binder powder; shear the mix so the PTFE stretches into fine fibers (fibrillation) that hold the powder together; then roll it into a free-standing film. Publish every setting (mixing time and speed, temperature, roll gap, number of passes) and each film's thickness, density, tensile strength and electrical resistance.
Why it is still open
Tesla says it now makes both electrodes of its 4680 cells this way, but makers do not publish their settings. A 2025 University of Oxford study of dry graphite sheets found that the degree of PTFE fibrillation changed the finished sheet, and that adding 3.5 percentage points more PTFE raised ultimate tensile strength by 900%. Fraunhofer ICT has published code for a model that grades dry-mixed LFP powder from microscope images on its own continuous dry-coating line, which judges one line's material, not whether a recipe repeats in another lab. A search found no open dataset of dry-film settings and results and no test of one recipe in two labs.
Built with what exists
Graphite or LFP powder, carbon black and PTFE binder powder; a high-shear lab blender; a small lab roll press; a micrometer; a load cell on a 3D-printed tensile jig, or a university tensile tester; a four-point probe for resistance; Python for the analysis. Work with dust extraction and keep the rolls far below the temperatures at which PTFE breaks down and gives off toxic fumes. No lithium metal, electrolyte or cells are used.
First step
Make five graphite films with 5% PTFE by weight at one setting, measure them, and write the recipe so that a stranger could follow it.
Done looks like
30 films from the first team and 30 from a second team working only from the published recipe, with the mean and spread of thickness, density, tensile strength and sheet resistance for each; the second team's means within 10% of the first team's, or the cause of each difference found and written down.