There is no such thing as a perfect battery — the EV safety challenge

With battery recalls in the news and battery fires on everyone’s minds, one of the primary challenges in the transition to a battery-powered world is rearing its head: batteries can be dangerous if not used correctly.

Root-cause analysis by multiple automotive OEMs has traced the issue back to cell-level defects, leading to fingers pointed at cell suppliers and calls for improved cell quality. While these explanations and aspirations may be perfectly valid and noble, it seems these OEMs have run headlong into a problem well understood by Tesla back in the earliest days of the modern EV industry: there is no such thing as a perfect battery.

A game of battery roulette

In the early 2000s, headlines told stories of lithium-ion batteries spontaneously combusting inside laptops and other devices (due to a phenomenon known as thermal runaway), leading to recalls by companies like Dell, HP, Lenovo, and even Apple. As Tesla set out to build some of the biggest lithium battery packs ever made at that time, late night experiments in the parking lot demonstrated how one burning cell could ignite the cells around it, leading to a deadly cascading inferno. This process came to be known as thermal propagation. Naturally, they sought to understand how likely this disastrous occurrence might be.

Tim Higgins’ new book “Power Play: Tesla, Elon Musk, and the Bet of the Century” tells the next part of the story:

Days later, a small group of battery consultants were gathered with what at first seemed like a manageable message: Yes, even the best battery manufacturers produced a random cell that would have a defect, causing it to short and catch fire. But the odds were remote. “It happens really, really infrequently,” one of the consultants said. “I mean like between one in a million and one in ten million cells.”

But Tesla planned to put about 7,000 cells in a single car. Sitting near [Tesla CTO] Straubel, [Tesla engineer] Berdichevsky pulled out his calculator and computed the likelihood that a cell in one of their cars might catch fire by chance. “Guys, that’s like between one in 150 and one in 1,500 cars,” he said.

With this realization, Tesla knew that in order to succeed — in order for EVs to scale — they had to design a battery pack that would remain safe even when the pack included a bad cell. This knowledge has informed their approach to battery design and safety at every level — cell, module and pack — ever since. (And shout out to Voltaiq Advisor Celina Mikolajczak, who was one of those nameless consultants that helped Tesla understand this game of battery roulette.)

How the numbers play out today

Let’s put some of these numbers in a contemporary context. To make the math easier, we’ll assume a worst case where an OEM gets one defective cell in every million from their supplier. (For comparison, the gold standard of quality control — Six Sigma, or 6σ — assumes 3.4 defects per million, so we’re assuming fully 71% fewer defects than that.) Given that defect rate, what would we expect to see across the fleets of some EV models in the news today?

Expected frequency of packs with a defective battery cell

The first thing you’ll notice is that Tesla uses ten to twenty times as many cells per pack relative to the models from Chevy and Hyundai, having based their pack designs around larger numbers of small cylindrical cells from the beginning (a progression in form factors from 18650 to 2170, and soon to 4680). You also probably know that it is the Hyundai and Chevy models, built with smaller numbers of much larger pouch cells, that are currently facing billion-dollar recalls — check out The High Cost of Failure. More on that in a bit.

Now let’s compare these figures to the frequency of fires reported in these recalled models, relative to the number of vehicles sold:

Frequency of reported fires in vehicle fleets

Wow — for these two recalled models, the expected frequency of packs containing bad cells is reasonably close to the frequency of fires reported across the fleet. Let’s not read too much into the precise numbers, recalling that we made a simplifying assumption about cell defect rate. Nonetheless, these numbers raise the concern that pack architecture could be a factor making catastrophic thermal propagation more likely when a bad cell starts to overheat on its own.

The drive toward larger-format cells?

Cell form factor is a topic that inspires something resembling religious fervor in the battery community, along with the accompanying arguments around who’s right and what’s best. We won’t take sides here, but there are a few things we can learn that inform pack safety.

Most auto OEMs have eschewed Tesla’s small-cell approach in favor of using much larger pouch cells, packed tightly together into brick-like structures called “modules.” The seductive appeal of these pouch cells stems from a few perceived advantages:

• Range: Flat, slab-like pouch cells can be packed together more tightly than cylindrical cells, offering an inherent energy density advantage at the pack level. Similarly, a larger cell uses relatively less of its volume for packaging, leading to a cell-level energy density advantage as well.
• Cost: Using fewer, larger cells reduces the total number of electrical connections, component count, pack complexity, and ultimately cost.
• Reliability(?): It follows logically that using fewer cells will reduce the likelihood that any given pack will have a bad cell, thus improving the overall quality and safety of the fleet. However there are reasons to believe that the safety and reliability case for large pouch cells is not so cut and dried.

There are a couple intrinsic properties of these larger format cells that may mitigate some of the perceived advantages, in regards to safety and reliability in particular:

1. More places for things to go wrong. In practice, the types of defects that lead to thermal runaway tend to scale with the total amount of materials inside the cell. For example, a fold or tear in the separator material, or at the edge of one of the electrodes, will tend to occur at some very low, but relatively steady rate per length of electrode material. A large-format cell with ten times the capacity of a smaller one will have roughly ten times the length (and area) of weld, separator, electrode, etc. and will thus be ten times more likely to contain a defect. The bottom line is that there is no free lunch, quality-wise, when using larger cells.
2. More stuff to burn. By definition a larger format cell contains more energy. Once thermal runaway begins in a cell, all of the energy in that cell is pretty much guaranteed to be released, even if the incident is confined to that one cell. So larger cells release proportionally more energy, increasing the severity of a single thermal runaway event.
3. More energy-dense packs. As discussed above, part of the appeal of large-format pouch or prismatic cells is that they can be packed together tightly, thus increasing energy density and range. However with all of these large, highly energy-dense cells packed together, a single thermal runaway incident is much more likely to spread to adjacent cells, ultimately igniting the whole pack.

This is not to say that OEMs using larger format cells are doomed to a future of balance-sheet-busting product recalls. Even Tesla has dramatically increased cell size over time, with their newest 4680 cells being roughly eight times the size of the 18650s in the original Roadster and Model S/X. Ultimately EV battery design is an optimization between cost, performance (range, acceleration, fast-charge capability), and safety, and there are likely multiple valid approaches.

However at this point it should be clear that OEMs must take steps to mitigate the impact of thermal runaway events when they do occur, because they are guaranteed to happen in some fraction of vehicles. The goal of eliminating bad cells from packs is a noble one, but in practice this is impossible to accomplish. Packs must be engineered with this game of battery roulette in mind from the start.

Okay, there will be defective cells. What do we do about it?

There are a number of things OEMs can do across the full lifecycle of an EV battery pack to minimize the likelihood of a catastrophic fire in one of their vehicles. Most, if not all, of these practices are already employed by Tesla, who has been working on this problem the longest. But there is plenty the rest of the industry can learn.

1. Clean up your cell supply chain. Penny pinching is still common in cell manufacturing, with many cell suppliers under-investing in data collection, quality control, and basic equipment like X-ray scanners to spot poor welds, tears, or misaligned electrodes. OEMs can seek suppliers that demonstrate a commitment to cell quality up front, and require as part of their supply contracts that they receive the full “digital thread” of data describing each cell through manufacturing and cell formation. With that said, an ongoing shortage in cell supply (which is likely to worsen), may deprive OEMs of the leverage to make these demands of their cell suppliers. In this case, a “trust but verify” mindset is called for, up to and including performing rigorous incoming QA testing on every cell before it is built into a production vehicle.
2. Design packs to limit thermal propagation. Again taking some lessons from Tesla, packs can be designed to incorporate physical spacing between individual cells, with padding to mitigate wear and tear from vibrations over the vehicle’s life. Packs can further be designed to electrically disconnect individual cells from the overall pack upon cell failure.
3. Monitor the battery throughout its life. The fact is that some of these problems develop over the course of a pack’s service life, even if it passed QA as it left the plant. Battery cells expand and contract with every charge-discharge cycle, leading to mechanical wear and tear inside the cell. Vibrations from normal driving can have a cumulative effect as well. These problems can often be detected in situ, however, with the appropriate investment in instrumentation inside the pack, and telemetry to analyze operational data at scale and recognize patterns of early failure across entire fleets of vehicles.
4. Close the loop. When problems are detected in the field, perform root-cause analysis, identify the cause, implement a selective (not fleet-wide) recall if appropriate, and remediate upstream to ensure it doesn’t happen again.

Notwithstanding these challenges, and the high-stakes game of battery roulette we’ve described here, we’re optimistic that there can be a bright future ahead for the broader transportation sector as fleets electrify. Achieving this vision will require substantial investments in engineering battery packs for safety across the full lifecycle, and acquiring the Enterprise Battery Intelligence capability that will enable many of these advances. We’ve seen this story before in the semiconductor industry, as over the last few decades investments in analytics and quality control powered dramatic increases in volume, yield, Moore’s law increases in transistor density, and exponential growth overall. It’s time for the battery industry to follow suit.