Choosing a Battery for Your Hardware Product: Chemistry, Safety, and UN 38.3

A hardware team we talk to every few months has the same story: the PCB works, the firmware is tuned, the enclosure looks great, and then a freight forwarder asks for the UN 38.3 test summary and the whole shipment stops. Or worse — the device browns out and resets every time the cellular modem transmits, and nobody can figure out why until they put a scope on the battery and watch it sag 800 mV under a 2 A pulse. The battery is almost always the last thing a team designs and the first thing that bites them, because it is not a component you drop in at the end. It is a system decision that reaches into your PCB, your certification budget, and your ability to legally put product on a plane.

The Battery Is a System Decision, Not a BOM Line

The battery couples to almost every other subsystem: the load profile your firmware creates, the copper and cutoff voltage on your board, the charger IC, the thermal path through your enclosure, and a stack of transport regulations. Pick the cell first, in isolation, and you will re-spin the board when it can't source peak current, re-cost the project when certification lands, and re-plan your launch when shipping paperwork isn't ready. Decide it alongside your power architecture. If you are still shaping the load side, our guide to low-power design for battery devices covers the current-budget work that has to happen before you can size a cell at all. And once you have sleep and active currents in hand, run them through our free Battery Life Calculator — it folds in usable-capacity derating and self-discharge, the two corrections that separate a datasheet estimate from field life.

Lithium-Ion and LiPo: The Default, With Strings Attached

Rechargeable lithium-ion (cylindrical cans) and lithium-polymer (pouch) are the default for most connected products, and for good reason: ~3.6–3.7 V nominal, roughly 150–260 Wh/kg at the cell level, and a mature supply chain. That range is wide for a reason — high-drain cylindricals like the Molicel P42A sit near the bottom (~200 Wh/kg), most consumer LiPo pouches run ~150–220 Wh/kg, and only high-energy cells reach the top of the band. But 'default' does not mean 'drop-in.' A bare Li-ion cell is a hazard without three things around it:

• A protection circuit (PCM) that trips on overcharge above ~4.25 V, over-discharge below ~2.5 V, and overcurrent/short — never ship a bare cell to a consumer.
• A certified CC/CV charger IC that respects the 4.2 V ceiling to within a few tens of millivolts; overshoot degrades cycle life fast and, at the extreme, plates lithium.
• A cell from a real manufacturer (Samsung, LG, Panasonic, Molicel) with its own IEC/UL certification and datasheet — not an unmarked pouch from a marketplace, which will fail your own certification and can't be shipped.

Cylindrical vs Pouch: 18650, 21700, and Custom Packs

Cylindrical cells (18650 at ~2.5–3.5 Ah, 21700 at ~4–5 Ah) are rugged, thermally well-understood, cheap per watt-hour, and easy to source and swap. The 21700 has largely displaced the 18650 in new designs for its better energy-per-cell. Pouch (LiPo) cells win when you need a specific thin form factor — wearables, handhelds, anything where a cylinder won't fit — at the cost of mechanical fragility and swelling margin you must design an enclosure around. If you are building something worn on the body, the flex-PCB, form-factor, and battery constraints stack up quickly; our wearable device PCB design breakdown walks through those trade-offs together.

• Cylindrical: use when volume allows, when you want a mechanically robust cell, or when field-replaceable batteries matter.
• Pouch: use when thickness or a custom outline drives the industrial design — and budget for 8–10% swelling clearance over life so the enclosure doesn't crack or bulge.

LiFePO4: Safer, Flatter, Fewer Cycles Lost

Lithium iron phosphate sits at ~3.2 V nominal with a notably flat discharge curve and much better thermal/abuse tolerance than standard Li-ion — it is far less prone to thermal runaway. You pay for that safety in energy density (~90–120 Wh/kg, roughly half of Li-ion), so it is the wrong pick for a pocket product where every gram matters. It is the right pick where cycle life and safety dominate: 2,000–5,000+ cycles is routine, versus 300–500 for consumer Li-ion. Think stationary devices, solar-charged nodes, and anything that cycles daily for years.

LiSOCl2 and the Pulse-Current Trap

For a 5–10 year primary (non-rechargeable) design — an asset tracker, a utility sensor, a device that gets sealed and forgotten — lithium thionyl chloride is hard to beat: ~3.6 V, exceptional energy density (~500+ Wh/kg), and self-discharge under ~1% per year, which is what makes decade-long life plausible. The trap is pulse current. A bare LiSOCl2 cell has high internal impedance and cannot deliver the 1–2 A bursts an LTE-M/NB-IoT modem demands; the voltage collapses and the MCU resets mid-transmit. The fix is standard and non-optional: pair the cell with a hybrid-layer capacitor (HLC) or supercapacitor that supplies the pulse while the cell slowly recharges it between transmissions. Anyone who promises multi-year cellular life on a bare thionyl-chloride cell is either not transmitting much or hasn't scoped the rail.

Coin Cells, Alkaline, and NiMH

• CR2032 (primary, 3 V, ~220 mAh): fine for an RTC, a remote, or a BLE beacon that blinks occasionally — but its pulse limit is even worse than thionyl chloride, so a BLE advertising burst can brown out an undersized coin cell.
• LIR2032 (rechargeable, 3.6 V, ~40 mAh): a coin-cell footprint you can recharge, at roughly a fifth the capacity — useful for backup rails, not primary power.
• Alkaline (1.5 V) and NiMH (1.2 V): user-replaceable and globally available, which is a real product advantage; the cost is a sagging voltage that your boost converter and firmware brownout thresholds must tolerate down to ~0.9 V/cell.

Sizing: From Average Current to Real Capacity

Start from the load, not the cell. Compute a true average current from your duty cycle — sleep current, active current, transmit current, and how long each lasts — then divide the target runtime into it to get amp-hours. Then derate hard, because the datasheet number is a lie in the field. Apply all four of these together:

• Temperature: capacity drops sharply below 0°C; a cell rated for a room-temperature bench can lose 20–40% of usable capacity in the cold.
• Aging: design to end-of-life, typically ~80% of rated capacity after your target cycle count, not to a fresh cell.
• Cutoff voltage: you only get the capacity above your system's brownout point, so a 3.0 V cutoff leaves usable energy on the table that a 2.7 V-tolerant design would capture.
• Peak load headroom: an undersized cell can meet the average and still fail the peaks — size for both.

A practical rule: after honest derating, plan for 1.5–2x the naive capacity you first calculated. Teams that skip this ship a device that hits its runtime spec on the bench in July and misses it by half in a customer's garage in January.

Peak Current Is the Number That Kills You

Average current sets your runtime; peak current sets whether the device works at all. A cellular transmit burst, a motor inrush, an LED flash, or a Wi-Fi association can pull an amp or more for tens of milliseconds. Multiply that peak by the cell's internal resistance and you get the voltage sag; if that dip crosses your regulator's dropout or the MCU brownout threshold, the device resets — and it will look like a firmware bug for a week before someone scopes the rail. This is exactly why radio-heavy MCUs like the ESP32-S3 have such spiky profiles; our ESP32 vs STM32 vs Arduino for production comparison gets into how much those transmit peaks differ by part. Design the battery, the copper, and any bulk/HLC capacitance for the peak, not the average.

Protection and Battery Management

The protection you need scales with cell count. A single Li-ion cell needs at minimum a PCM handling over/under-voltage, overcurrent, and short-circuit; add a fuel gauge (coulomb counter) if the product needs an accurate state-of-charge readout rather than a crude voltage guess. Multi-cell packs (2S and up) need a real BMS with cell balancing — because series cells drift apart over cycles, and without balancing your weakest cell defines the whole pack's capacity and can be over-discharged into damage. Thermal protection (an NTC on the cell, monitored during both charge and discharge) is not optional for lithium; it is the last line before a runaway event.

Charging: CC/CV, Thermistors, and Wireless

Standard lithium charging is constant-current then constant-voltage: ramp current until the cell reaches 4.2 V, then hold voltage and taper current to termination. Charge rate matters — 0.5C is gentle and long-lived, 1C is faster and hotter; going above what the cell's datasheet allows trades cycle life and safety for speed. Wire an NTC thermistor into the charger IC so it follows JEITA temperature windows and refuses to charge a cold or hot cell — charging a lithium cell below 0°C plates lithium and is a genuine safety failure, not just a longevity one. USB-C plus a purpose-built charger IC (TI's BQ family, Microchip, and similar) is the well-trodden path. Wireless (Qi) is a lovely UX for a sealed IP67 product, but it adds a rectifier, roughly 30% charging inefficiency as heat right next to the cell, and a thermal problem you must design around.

UN 38.3: The Test That Gates Your Shipment

UN 38.3 is mandatory transport testing for every lithium cell, battery, and pack before it can legally move by air, sea, or road. It is a battery of eight tests — altitude simulation, thermal cycling, vibration, shock, external short circuit, impact/crush, overcharge, and forced discharge — and you need a passing test summary before a freight forwarder will accept the shipment. Two things surprise teams: it applies to the whole pack as assembled (not just the cell, so a certified cell does not automatically clear your pack), and testing takes weeks at an accredited lab. Discover this the week before your first container and you have just added a month to your launch. Put it on the schedule the day you freeze the pack design.

IEC 62133, UL 2054, and Cell-Level Safety Certs

UN 38.3 governs transport; a separate stack governs product safety. IEC 62133 is the international safety standard for rechargeable cells and batteries and is what many market approvals reference. In North America, UL 1642 covers lithium cells and UL 2054 covers the finished battery pack. Your cell vendor should already hold IEC 62133 / UL 1642 on the cell — this is a concrete reason to buy branded cells — but the pack-level certification is yours to earn. Reusing a certified cell shortens the path; it does not eliminate the pack testing.

Shipping Classification and the MSDS

Every lithium battery ships under a UN classification that dictates packaging, labeling, and documentation — and which number you use depends on the chemistry, not just the product. For rechargeable lithium-ion, the two you will meet most are UN3480 for cells/batteries shipped alone and UN3481 for lithium-ion packed with or contained in equipment (the finished-product case). Primary (lithium-metal) chemistries ship under a different pair: UN3090 for lithium-metal cells/batteries alone and UN3091 for lithium-metal contained in or packed with equipment. That distinction matters here, because the LiSOCl2 asset tracker and the CR2032 coin cell in this article are lithium-metal, not lithium-ion — declare either under UN3480/UN3481 and you have misclassified the shipment. You also need a battery Safety Data Sheet (SDS/MSDS) on hand for customs and carriers. These are not paperwork you can improvise at the dock — get the classification and SDS lined up alongside UN 38.3, because a carrier will reject an incorrectly declared or undocumented lithium shipment outright.

The Failure Modes We See Repeatedly

• Undersizing for pulse current — the device resets on every transmit and the team burns a week blaming firmware before scoping the rail.
• Shipping a bare cell with no protection circuit, turning a normal over-discharge into a field safety event.
• Buying uncertified marketplace cells to save a few cents, then failing pack certification and being unable to source a drop-in replacement.
• Declaring a primary lithium-metal product (LiSOCl2, coin cell) under the lithium-ion UN numbers and having the shipment bounced at the dock.
• Discovering UN 38.3 the week before a shipment and adding a month to the launch while a lab runs the eight-test sequence.
• Sizing to a fresh, room-temperature datasheet number and missing the runtime spec by half once cold and aging derating hit in the field.

Frequently Asked Questions

How early do we need to start UN 38.3 testing?

The moment you freeze the pack design — cell, PCM/BMS, and mechanical assembly. Accredited labs typically need several weeks, and any change to the pack after testing (a different cell, a new BMS) can invalidate the summary and force a re-test. Treat it as a long-lead item on par with tooling, not a checkbox before shipping.

Can we just buy cheaper unbranded cells to hit our cost target?

For anything you certify and ship, no. Unbranded cells rarely carry their own IEC 62133 / UL 1642, their real capacity and internal resistance don't match the label, and you can't get a consistent second source. You'll spend far more failing certification and chasing field failures than you saved. Buy branded cells and optimize cost elsewhere in the BOM.

Do we need a full BMS, or is a protection circuit enough?

A single-cell product is usually fine with a PCM plus, optionally, a fuel gauge for accurate state-of-charge. The moment you go to 2S or higher — multiple cells in series — you need a real BMS with cell balancing, because series cells drift and the pack is only as good as its weakest, worst-balanced cell.

Talk to an Engineer About Your Power System

The right battery decision is made early, alongside the load profile, the PCB, the charger, and the certification schedule — not bolted on at the end. If you are choosing a chemistry, sizing a pack for real peak current, or staring down UN 38.3 with a ship date looming, inteRFace R&D can help you make the call before it becomes a re-spin or a delayed launch. Tell us about your device and your runtime target at https://interfacernd.com/#contact and we'll walk the power system through with you.