By Kyle Proffitt
April 25, 2025 | Troy Hayes, Director of Quality at Tesla, gave one of the plenary presentations at last month’s International Battery Seminar. Hayes talked passive propagation resistance in battery design—how to keep a single thermal runaway event from growing into a major fire. Freer discussed the wider UK battery strategy and how they are bolstering research and innovation alongside an energy transition.
“You need to develop a battery pack that, if you have a single thermal runaway event of a cell inside the pack, it needs to passively be able to stop that reaction from spreading to the entire battery pack,” Hayes said as the thesis for his keynote address. He provided some gratifying data. “At Tesla, we’ve accumulated over 200 billion miles of data worldwide… electric vehicles experience fires far less frequently than their ICE counterparts.” In the US, the vehicle fire rate for ICEs is about 1 per 18 million miles; that rate falls to about 1 fire per 120 million miles for EVs, nearly 7-fold reduction. He supported this claim with data published by various other countries demonstrating everything from 4- to 83-fold reduction in fires for EVs.
“It’s pretty clear that EVs catch fire less frequently than internal combustion engines, which isn’t surprising since internal combustion requires… combustion,” Hayes said. However, EV fires make headlines because they can be much more difficult to put out. He added a confounder in the data: some EVs are just in fires and not causative. An example was shown of a burned-out Tesla vehicle from which an intact, functional battery was subsequently retrieved.
Still, the goal is to stop any EV-initiated fires. Hayes listed several possible sources in a lithium-ion battery: torn electrodes, folded separator, lithium plating (which can be related to temperature, speed of charging, electrolyte depletion, etc.), core impingement in cylindrical cells, and metallic contamination of the cathode or anode. For Tesla, he says they are applying an analogy to protection from a gun; they are removing bullets, and they’re wearing a bullet-proof vest, just in case.
Remove the Bullets
“There’s a number of solutions; a lot of them are quite low tech and pretty easy to implement, and then there’s a number that are much higher tech,” Hayes said. For materials like iron, if it’s homogeneously distributed throughout the cathode material, they need to control the level of incoming contamination to around 15 parts per million. However, for larger particles in a 100-200 micron range, “it needs to be controlled to like 30 parts per billion.” He pointed to enclosed physical spaces or virtually enclosed spaces kept clean with fan and filter units to limit contamination. Then there are simple concepts, he said. “You need to avoid metal-on-metal contact in lithium-ion batteries like the plague.” However, that’s difficult, because all of your equipment is metal. He walked through several examples of seemingly benign environmental sources, including a door rubbing a doorframe and personal protective equipment that uses metal hooks for tying onto when working at heights. He also mentioned the example of abrasive metal oxides moving through a closed plumbing system as a powder, creating wear in any area of impingement. Hayes champions a pretty simple add-on for reducing metal contamination. “One of the low-tech methods that I like to use… all over the world I preach: magnets, magnets, magnets.” He said this is a really cheap way to clean iron, nickel, and cobalt-based contaminants. He added that 316 and 304 stainless steel, which are not magnetic, become magnetic with a small energy input and phase transition. This means that any abrasive process creating burrs or small particles causes stainless steel to transform to a magnetic structure, which can be easily removed.
Math and Failure Rates
Hayes said another tactic to help find and remove risk is offline product sampling. However, if your defect rate is low, many samples are required to have a reasonable chance of finding the defect. In an example, if the defect rate is 75 parts per million, you need about 10,000 samples to have a 50% chance of detection. Instead, Hayes favors 100% inspection techniques, including cameras, x-rays, beta-rays, and soon, extensive CT scans. He predicts that within two years, we’ll have high-volume manufacturing that uses 100% CT with resolution of 40 microns or less.
Wear a Vest
Even with all that, there are some “bullets” left in the gun. Hayes showed statistics indicating that from 2005-2015, the spontaneous runaway rate in lithium-ion batteries was between one in 1 million and 1 in 10 million, but has since improved to about 1 in 40 million. That’s great, but because EVs have thousands of cells in them, “you’re expecting to have a single cell experience thermal runaway in every few thousand vehicles; that’s why passive propagation resistance is critical.” Restating the thesis, “you have to build a pack that will not turn one cell fire into a car fire.”
Tesla’s solution is to use a smaller unit of energy, the cylindrical cell. The annular space between cylindrical cells allows you to manage thermal characteristics, using cooling tubes or potting material. Then they put plates of aluminum or steel on both sides of the pack.
Hayes talked about how they predict propagation events, starting with individual cells, analyzing runaway with high-speed x-ray imaging, adding thermal modeling that begins to consider larger groups of cells, and then moving on to actual pack testing. At the pack level, he says the testing gets really expensive, and you need to think about all the different kinds and combinations of initiation events that might occur, such as punctures, overheating, and overcharging. Once again, he turned to a mathematical complication: “failures are really stochastic, meaning even if you put it under the same exact condition ten times, you can get ten different results.” He said if you really want to have statistical confidence an event is less likely than 1%, about 300 tests have to be run. You don’t want to burn 300 battery packs if you can avoid it, but Hayes says you can avoid that by running your tests carefully. If you can initiate individual runaway events without those propagating to neighboring cells, you can run the next test using the same pack.
Hayes pointed to one somewhat counterintuitive and interesting finding. When a thermal runaway event occurs, the more mass that stays inside that cell, the more likely it will propagate to its neighbors. He explained that if the contents of the cell are actually expelled, that heat is distributed among the pack more evenly. How to use this information for battery design was less clear.
In closing, Hayes said that pack propagation is something you can mitigate with good design and quality control. He added that for large pouch cells, this is really difficult and that prismatic cells also present challenges. In every case, extensive testing is necessary.
