The Missing Ohm: How a 0.5% Tolerance Decided a Gadget Recall
Archi3 min read
Contents
The spec that looked fine on paper#
A power bank ships with a 10,000 mAh cell, a boost converter, and a resistor that sets the output voltage. The resistor is a 1% part. The design calls for 100k. At 1% tolerance, the voltage stays within 50mV of target. The firmware has a 100mV window. The margin looks comfortable. It is not.
When the resistor drifts high, the output creeps up. At 1.5% drift, the voltage moves 75mV. Add a hot day and a cheap cell, and the protection circuit trips. Users see a dead port. The return rate climbs. The root cause is a part that costs two cents and sits in a corner of the board nobody photographs.
What the tolerance actually means#
Tolerance is a promise about the part, not the system. A 1% resistor can sit at 0.9% high and 0.8% low across a batch. Over temperature, the drift adds. Over time, it adds more. The datasheet gives you a number at 25C. The gadget runs at 45C inside a bag. The number moves.
Engineers who treat tolerance as a worst-case bound get burned. Tolerance is a distribution. Most parts cluster near nominal. A few sit at the edge. In a run of 100,000 units, the edge is not rare. It is a production line.
The feedback loop nobody models#
The boost converter uses the resistor to sense output voltage. The controller adjusts the duty cycle to hold the setpoint. If the resistor reads high, the controller thinks the output is too high and lowers the duty cycle. The output drops. If the resistor reads low, the output climbs. The loop corrects in one direction and amplifies in the other.
This is not a failure of the resistor. It is a failure to model the loop with the resistor's real distribution. A Monte Carlo run with 10,000 samples would have caught it. A single nominal simulation did not.
The fix that cost more than the part#
The company swapped the 1% resistor for a 0.1% part. The unit cost rose by eight cents. The rework and recall cost ran into six figures. The math is lopsided in hindsight. It always is.
A better fix would have been a different topology. A sense resistor in the return path tolerates drift differently. But changing the topology means a new board spin, new certification, and a delay of months. The 0.1% part shipped in weeks.
Where else this hides#
Every gadget with a feedback loop has a resistor like this. Chargers, LED drivers, audio amps, battery gauges. The part is small, cheap, and easy to overlook. The failure mode is not a bang. It is a slow drift that shows up as a warranty claim in month nine.
Consumer gadgets are full of these quiet decisions. The USB-C connector's 5.1k pull-down resistor, the crystal's load capacitance, the ferrite bead's impedance at 100MHz. Each one is a number on a datasheet that a designer picked in an afternoon. Each one can decide whether a product works in the field.
What buyers can learn from this#
You cannot inspect a resistor. You can read the spec sheet and ask a question: what is the tolerance on the parts that set the output? A maker who answers with a number and a test method is a maker who thought about it. A maker who answers with a slogan did not.
A recall is rarely caused by the biggest component. It is caused by the smallest one that nobody checked twice. The 0.5% drift is not dramatic. It is enough.
More articles