Why a Same-Chemistry Second Source Isn't a Drop-In Replacement

A same-chemistry second source is not a drop-in replacement, because “chemistry” is a family name, not a specification. The parameters that decide whether a pack still behaves the way it was validated to behave — the charge and discharge cutoffs, the open-circuit-voltage-to-state-of-charge curve, cell impedance, and the temperature window the cell maker is actually willing to warrant — are cell-level properties, not chemistry-level ones. Swap the cell maker and the qualification work restarts at the BMS and in the test file, however identical the top rows of the two datasheets look.

I spend a good part of my year doing second-source qualifications on lithium packs, and every one of them starts with the same sentence, usually from a buyer with a spreadsheet open: “It’s the same chemistry, so it should be a drop-in.”

Come stand at the bench with me the morning the samples land.

There are two cells in front of us, both LFP, both 3.2 V nominal, both in the same 26650 can. The incumbent on the left, the candidate on the right. Two datasheets, printed, side by side. The thermal chamber behind us is holding at -10 C with the door frosted over and a fan running somewhere inside it. A cycler is stepping a channel through a charge sequence, and you can hear the contactors click each time it changes step. The laptop is logging two voltage traces that, right now, at 20% SOC, sit on top of each other almost perfectly.

That’s the trap. They look the same at the bottom of the curve. They separate at the top.

The rows that disagree

The first nine lines of the two datasheets match closely enough to be boring. Nominal voltage, capacity at 0.2C, energy density, operating temperature, storage temperature, dimensions, weight, standard charge, standard discharge. If you stopped reading there, you’d sign the PPAP and move on.

Then come the rows I actually read first.

Charge cutoff: 3.65 V on the incumbent, 3.60 V on the candidate. Max continuous charge current: 1C versus 0.5C. Discharge cutoff: 2.5 V versus 2.0 V. Charge temperature window: 0 to 45 C on one, and 0 to 45 C on the other but with a note that charging above 40 C voids the warranty. AC impedance at 1 kHz: 8 milliohms versus 12. Ten-second DCIR at 50% SOC: 15 versus 22. Cycle life: both claim 2,000-plus cycles, but one is measured at 1C/1C at 25 C to 80% of initial capacity and the other at 0.5C/0.5C to 70%.

Every one of those numbers is honest. They just aren’t the same cell. I’m using illustrative figures here — yours will land wherever they land — but the pattern is reliable: the differences cluster in the rows that govern how the pack is charged and how the BMS decides what “full” and “empty” mean.

Now watch the two traces on the laptop. Bring both cells to 30% SOC, rest them, and read open circuit voltage. The candidate sits maybe 15 to 25 mV above the incumbent. At 70% SOC they might be 40 mV apart the other direction. That curve — OCV as a function of SOC — is the single most consequential thing you can measure, and it is not on either datasheet.

Why the BMS, not the chemistry, decides whether it’s a drop-in

Here’s the part that took me a few painful projects to internalize: the drop-in question is answered by the firmware, not by the cathode.

Where do the protection thresholds in your BMS come from? They come from the incumbent cell. Overvoltage protection sits a little above the incumbent’s charge cutoff plus its tolerance. Undervoltage sits a little below its discharge cutoff, with margin for the IR drop under load at your worst-case current. Charge termination current — typically somewhere around C/20 to C/10 — is tuned to how that specific cell’s current tapers into CV. The temperature thresholds for charge inhibit are set against the window that cell maker is willing to stand behind.

Move the charge cutoff from 3.65 V to 3.60 V and every one of those numbers deserves a second look, because a pack that terminates at a voltage the candidate cell doesn’t want to see is either undercharging or overcharging, and you won’t know which until the cycle data comes back.

Then there’s the fuel gauge. Most gauges on the market are model-based: they learn a relationship between OCV, impedance, coulomb count, and temperature, and they learn it against one specific cell. When you change the cell, the model is wrong in a way that doesn’t show up on a fresh pack. It shows up at the edges. The pack reports 100% and then falls off a cliff. Or it reports 15% when there’s really 25% left, and your customer’s device goes into low-battery shutdown early. Or the gauge slowly walks off over a few hundred cycles and never comes back.

A gauge relearn isn’t optional, and it isn’t fast. It wants full charge, full discharge, rest periods, and temperature variation, and it wants it repeated on more than one pack.

And then there’s the case that no firmware tweak fixes. If the incumbent was warrantied for charging at 0 C and the candidate’s warranty starts at 5 C, you don’t have a firmware change. You have a product behavior change that shows up in your customer’s cold-weather spec. That’s a conversation with product management, not with engineering.

Three tiers, and only one of them is real dual-sourcing

I sort every second-source request into one of three buckets before I agree to anything, because the scope of work is wildly different.

Tier one is the same cell from the same maker, coming from a different plant or a different packaging arrangement. The cell spec is identical, so the requal is mostly paperwork: confirm the cell part number is genuinely the same, confirm the UN 38.3 test summary still covers the configuration you’re buying, and confirm traceability at the serial level so you can tell the two apart in a field return. This one usually takes weeks, not quarters.

Tier two is the same chemistry family from a different maker. This is where I live. Voltage window, charge profile, cutoffs, impedance, OCV curve, thermal derating, tab geometry, weld spec — all of it has to be compared, and none of it can be assumed from the chemistry label.

Tier three is a different chemistry family selected against the same pack requirement, usually because somebody found capacity in LFP when the design was NMC, or the reverse. I’ll be blunt about this one: it is almost never dual-sourcing. It’s a second design. Two voltage windows means two charge profiles, and two charge profiles means two BMS configurations. Two cell families means two sets of safety evidence, two work instructions, two incoming inspection plans, and a decision about whether the same part number can ship with either build inside it. If the answer is yes, you’ve made your BOM ambiguous, and the day a pack comes back from the field you’ll have no fast way to know which cell was in it. If the answer is no, you’re managing two SKUs, and you should say so out loud rather than letting it happen in the ERP.

The reason tier three still gets considered is supply. Which brings up the question I ask before any of this starts.

What a second source actually buys you

Ask what the constraint is. Not what the rumor is — what the constraint is.

If the thing that’s tight is cathode active material or separator, a second cell maker who buys the same material from the same place doesn’t relieve anything. You’ve added a qualification and kept the same bottleneck. If the constraint is fab capacity in one region, then yes, a second maker helps, but only in proportion to how much of their output you can actually get.

That’s the part continuity plans get wrong. Allocation in a tight market goes to the customers with history. A second source you’ve never bought volume from will not hand you a proportional share of their capacity in the quarter you need it. Your realistic split in year one might be 85/15, and the 15% may arrive on a lead time that’s longer than the incumbent’s, not shorter. Different chemistry families have different lead-time shapes and different allocation behavior, and a plan that assumes the second source is a pressure-relief valve rather than a small trickle is a plan that fails in exactly the quarter it was written for.

I write the split down as a percentage, with the lead time attached, and I keep the incumbent’s volume where it is until the second source has proven it can ship on schedule twice.

The evidence you have to reissue, not reuse

Here’s the piece that surprises people who haven’t done it. The test evidence on your incumbent does not transfer. Not the safety file, not the DVT report, not the incoming inspection limits, not the label.

A different cell means a different UN 38.3 test summary — that document is tied to the cell type, and it’s the one a freight forwarder will ask for. If your pack carries a safety listing, the cell is a critical component in that listing, so a cell change is a change to the listed configuration and needs to be evaluated as such, not written into a change notice and filed. Your customer’s own qualification requirements, whatever they are, will likely have to be run again too.

So the internal plan looks roughly like this, in order. Paper comparison first: full datasheet-vs-datasheet on cutoffs, rate limits, temperature derating, cycle life conditions, storage behavior, self-discharge, mechanical tolerances, tab and weld specification. Then cell-level sampling: capacity at a few rates and temperatures, DCIR, OCV points across SOC with proper rest, a self-discharge soak of a couple weeks, and a cycling run at your pack’s actual profile and actual temperatures. That cycling run is where the calendar goes. A month is fast. If you want a life data point you’re looking at considerably longer, and there is no way to compress it except by starting earlier.

Then pack level, with first articles built from the candidate cells. Run the charge profile the BMS will actually use, and watch cell balance over repeated cycles rather than one. Verify the protections by injecting faults, not by reading the config back. Run the gauge learning cycle and then check SOC accuracy across temperature and after some aging. And confirm that the low-temperature charge behavior your product promises is behavior the new cell will actually tolerate.

Only then does it go to production, and only then do you update the work instructions, the label, the incoming inspection plan, and the traceability scheme so that a unit built with the second source can be identified without a teardown.

None of this is exotic. It’s just longer than anyone wants it to be, and the temptation to skip it is strongest in the quarter when you’re short on cells — which is exactly when the cost of skipping it lands.

“Same chemistry” is a sentence that starts a conversation. It has never once been a qualification.