Does heat kill your battery? It depends on the chemistry

Almost every rule of thumb about batteries comes from consumer electronics, because that is where people have had batteries in their hands for twenty years. Phones and laptops taught a generation that heat is the enemy, that you should not leave the thing in a hot car, and that a warm battery is a dying one.

Then the same people buy a home storage system or spec a grid installation, and they carry the rule of thumb across without noticing that the cell chemistry changed underneath it. The battery in your phone is almost certainly a nickel-based cell. The battery on your wall, and roughly nineteen out of twenty new grid installations, is lithium iron phosphate. Those are different materials with different failure chemistry, and the published data says they do not respond to temperature the same way.

I write about this because I work on the control side of it, and thermal setpoints are a decision somebody makes, usually by importing an intuition rather than reading a study.

The study

The source here is a good one, and it is open. In 2020, a team at Sandia National Laboratories published "Degradation of Commercial Lithium-Ion Cells as a Function of Chemistry and Cycling Conditions" in the Journal of The Electrochemical Society. Yuliya Preger and colleagues explain why they bothered, in the abstract:

appropriate selection of cells for different applications is difficult due to limited public data comparing the most commonly used off-the-shelf Li-ion chemistries under the same operating conditions.

That phrase, "under the same operating conditions," is the whole value of the paper. Plenty of degradation data exists. Very little of it is comparable, because every lab and every manufacturer tests under its own protocol, so you cannot line two chemistries up next to each other and say which one aged faster without the comparison being contaminated by the test design.

Sandia ran commercial LFP, NCA and NMC cells through a common matrix: temperatures from 15 to 35 °C, depths of discharge of 40-60%, 20-80% and 0-100%, and discharge rates from 0.5C to 3C. It is a multi-year study, and the resulting dataset has become one of the reference sets in the field.

The result that reverses

Start with the headline number people usually quote, which is cycle life. Measured in equivalent full cycles, the paper reports LFP at 2500 to 9000, NCA at 250 to 1500, and NMC at 200 to 2500.

That spread is why LFP won the grid, and I have written about that before. But look at the width of those ranges. LFP's own cycle life varies by a factor of more than three depending on nothing but the conditions it was run under. The chemistry sets the range. The operating conditions decide where in the range you land.

Now the finding that made me want to write this. On temperature, the paper says:

capacity fade rate increased with increasing temperature for LFP cells but decreased for NMC cells

Read that twice. Over 15 to 35 °C, warming the cells made the LFP degrade faster and the NMC degrade slower. The third chemistry sat it out: "The NCA cells did not exhibit a strong temperature dependence in the examined range."

So there is no universal direction. "Cooler is better" was true for one of the three chemistries tested, false for another, and roughly irrelevant for the third.

Why it reverses, and why "cold is good" is also wrong

The mechanism matters here, because otherwise this reads as a curiosity rather than something you can reason with.

Degradation is not one process. Two families of it run at once and pull in opposite directions with temperature. Chemical side reactions, the ones that grow the interface layer on the electrode and consume lithium, speed up when things get hot, in the ordinary way that chemistry speeds up when heated. But the transport processes, lithium actually moving into the electrode structure during charging, get sluggish when things get cold. When transport is too slow to keep up with the charging current, lithium stops intercalating and starts depositing as metal on the anode surface. That is lithium plating, and it is both a fast capacity killer and a safety problem.

So every cell has a U-shaped curve with a best temperature somewhere in the middle, and the whole question is where the bottom of that U sits relative to the range you actually operate in. The paper puts numbers on it from the wider literature. For LFP: "previous reports on cycle aging of LFP cells indicate a tipping point at temperatures of 5 °C-10 °C, with degradation rates increasing both above and below this temperature." For NMC: "Studies of NMC cells consistently indicate a tipping point around 35 °C."

That single pair of numbers explains the reversal completely. Sandia tested from 15 to 35 °C. For LFP, with its optimum down at 5 to 10 °C, the entire test range sits above the bottom of the U, so warming only ever made things worse. For NMC, with its optimum up around 35 °C, the entire test range sits below the bottom, so warming was moving toward the sweet spot the whole time.

Both chemistries obey the same physics. They were just being observed on opposite sides of their own minimum. And note what this kills: "colder is safer" is not a general rule either. Push an LFP cell below roughly 5 °C and, on those same reports, degradation starts climbing again. Charging a cold battery hard is one of the more reliable ways to damage it.

The clock that runs while nothing happens

There is a second half to this, and for anyone buying a battery for backup rather than for daily cycling it may be the more important one.

A cell degrades even when you never use it. That is calendar aging, and it is driven mostly by temperature and by the state of charge the cell sits at. A later Sandia presentation by Reed Wittman, using cells from the same program, reports the effect on LFP directly. Holding cells at 25 °C, the ones parked at 90% state of charge lost about 9% of capacity, while the ones parked at 25% lost about 2.5%. Holding state of charge at 50% instead and varying temperature, cells at 35 °C lost roughly 8% over three years, against roughly 1% at 15 °C.

Sitting still at a high state of charge, in a warm place, cost several times more capacity than sitting still cool and half empty. Nothing was drawn from those cells at all.

This is the tension at the centre of any backup system, and almost nobody names it when selling one. A battery bought for outage protection wants to be full, because a battery at 40% when the storm arrives is a battery that gave you 40% of what you paid for. But full and warm is precisely the condition that ages it fastest while it waits. Every day of readiness costs a little life, and the bill arrives years later as a system that no longer carries the house through the night.

The same presentation makes a further point that is easy to miss: a meaningful share of what looks like cycling wear is really calendar wear happening underneath the cycling. Wittman estimates calendar aging may contribute around 33% of the fade seen in the cells cycled at 35 °C, dropping to roughly 25% and 10% as the cycling temperature falls. Some of what you were blaming on use was just time, spent hot.

Why this is a control problem

Put the two halves together and you get something that is genuinely a decision rather than a fact.

If degradation depends on temperature, state of charge, depth of discharge and rate, and if the direction of the temperature term flips depending on your chemistry, then there is no setpoint you can write on a sticker. There is only a trade, made continuously, between what the system is for and what keeping it ready costs. Holding a backup battery at 100% buys certainty today and spends life every day. Holding it at 60% keeps it healthier for longer and gives you less margin the one night it matters. Cooling an LFP installation helps it; cooling an NMC one, within this range, apparently did not.

This is the work I do on the energy side, and I will be exact about my role: I help design the AI battery-cycling systems for a veteran-owned (HUBZone) energy-storage integrator. I do not own that company and earn nothing from this link. Full policy here.

And the honest version of what that work can do is smaller than the pitch you will hear elsewhere. A controller that models degradation does not make cells outlive physics. What it can do is stop treating pack life as free: hold a lower resting state of charge when no weather is coming and top up when it is, decline a marginal arbitrage hour that does not pay for the wear it causes, and use the correct sign on the temperature term for the chemistry actually installed. Those are real gains and they are bounded ones. I have made this argument before with ORNL's simulation work, and it points the same way: anyone selling an AI that extends battery life without a tradeoff is selling you something, because the tradeoff is physical.

What I could not confirm

These are small cells, not your system. Both sources tested 18650 format commercial cells, the size of an oversized AA. A home or grid battery is a pack of many cells with thermal management, a management system, and its own internal gradients, and pack-level behaviour is not a simple scaling of cell-level behaviour. Read this as evidence about chemistry, not as a prediction for a specific product.

One cell per chemistry, one manufacturer each. The LFP cells were A123, the NCA Panasonic, the NMC LG Chem. Cells from other manufacturers with the same nominal chemistry can behave differently. The reversal is a real measured result in these cells; I am not in a position to say it holds for every LFP or NMC cell ever made.

The second source is a conference presentation, not a peer-reviewed paper. The calendar-aging figures come from Sandia slide deck SAND2022-6841C. It is a credible national-laboratory document from the same program and I read it in full, but it has not been through peer review, and its own final slide lists "better quantify the contribution of calendar aging in different cycling conditions" as future work. The percentages are read from the deck's stated bullets and are approximate, as the deck itself marks them. The three-year duration is stated on the temperature slide; the state-of-charge slide does not state its own duration, so I have not attached one to those figures.

I did not independently verify the claim that LFP now takes roughly 95% of new utility-scale awards; that figure circulates in trade coverage and I have used it only as loose framing, not as a load-bearing claim. The tipping-point temperatures are the Preger paper's characterisation of prior literature, not a measurement made in that study, and I did not open the underlying references.

None of this rests on my own research. My published work is in microwave spectroscopy, not electrochemistry.

The signal

The useful habit here is older than batteries. A measurement is the answer to a specific question asked under specific conditions, and the mistake is almost never in the number. It is in carrying the number somewhere it was not measured.

"Heat kills batteries" was measured, and it is true, on the hot side of the curve, for the chemistry most people learned it from. Carried across to a different chemistry sitting on the other side of its own minimum, the same sentence points backwards. Not because anyone lied, but because the rule of thumb lost the conditions that made it true somewhere on the way over.

So the question to bring to an installer is not whether the system will be kept cool. It is which chemistry is in the box, what resting state of charge the controller will hold it at between outages, and who decided that number. If nobody can answer the last one, then it was decided by default, and defaults are where the life of the asset quietly goes.