The interrupt that was never asleep.
Two lines of division killed the part.
A connected personal-care device had to detect use within one second and last a sixty-day study on a 137 mAh cell. The proposed wake-up source was a time-of-flight sensor with an interrupt output. Peter Sords, then a senior electrical engineer at Kaleidoscope Innovation, ran the battery arithmetic and made the no-go call.
The problem
A wake-up source that had to last sixty days.
The program needed presence detection: the device had to recognize that it was in use within one second. It ran from a 137 mAh cell at 3.7 V, and it had to last a sixty-day study.
The proposed wake-up source was an ST VL6180X time-of-flight sensor. It has an interrupt output, and that had been taken to mean it was an event-driven, low-power choice.
What made it hard
Interrupt-capable is not the same as low power.
The VL6180X can raise that interrupt only while it is ranging continuously, actively emitting and receiving infrared. The interrupt spares the microcontroller from polling the sensor. It does not spare the sensor from polling.
That is the trap. Low power is a property of the duty cycle, not of the part, and an interrupt pin on a datasheet says nothing about what the part draws while it waits for the event.
How it was found and solved
The battery arithmetic, run against the real constraints.
Peter ran the part against the program's actual requirements rather than its feature list. Three rows settled it.
Poll at 2 Hz
137 mAh ÷ 0.17 mA = 805 hours, about 33 days.
0.17 mA
Barely half the study
Apply the documented low-power settings
Early Convergent Estimate, and a readout-average sampling period of zero. Distance accuracy is destroyed.
~0.1 mA
Still short of 60 days
Replace the sensor with a mechanical switch
Positioned to sit closed while the device is docked.
Virtually zero
No sensor left to budget
Polling at 2 Hz gave 33 days, barely half the requirement. The documented low-power settings reached about 0.1 mA only by destroying distance accuracy, and still fell short of sixty days. The answer was not a better sensor. It was no sensor.
No-go, with the fallbacks named.
Peter called no-go on the VL6180X. The switch came first. If a sensor proved unavoidable, he named the alternatives to evaluate: VCNL3020, SI1143-MOI-GM, SX9310, ISL29147 and Si1102.
Getting this wrong would not have been a respin. It would have been a study that died on day 33, in the field, with the data half collected. Two lines of division settled a part choice that would otherwise have been discovered at week five.
What it teaches a buyer
Do the battery arithmetic before the schematic.
An interrupt-capable sensor is not a low-power sensor
Before budgeting a wake source as event-driven, check whether raising the event requires continuous emission.
Run the division early
Capacity over average current, against the real requirement. It takes two lines, and here it ended a part choice.
Ask whether the job needs a sensor at all
The lowest-power answer here was no sensor at all: a switch that draws virtually nothing.
Say no with alternatives
The no-go named five fallback parts. A rejection that comes with options keeps a program moving.
How Kyros applies it now
Every wake source is a current draw until proven otherwise.
On a battery-powered product, Kyros profiles current across the real duty cycle and audits every wake source before a sensor or a schematic is fixed, because one unnecessary periodic wake can dominate the average. It is how we build power budgets for battery-powered connected devices.
Choosing parts for a battery-powered product?
Bring the requirement and the battery. We will tell you which choices the arithmetic rules out first. To run the numbers yourself, use the free Power Budget Calculator.
Schedule an Architecture CallFrequently Asked Questions
- Is an interrupt-capable sensor a low-power sensor?
- Not necessarily. Some sensors can raise an interrupt only while measuring continuously, so the sensor keeps drawing current while the microcontroller sleeps. Check what the part draws while it waits for the event before budgeting it as event-driven.
- How do you check battery life before committing to a sensor?
- Divide battery capacity by the average current over the real duty cycle and compare the result with the requirement. Here, 137 mAh at 0.17 mA gave 805 hours, about 33 days, against a 60-day study, which ended the part choice on paper.
- Who did this work, and for whom?
- Peter Sords, founder of Kyros Engineering, as a senior electrical engineer at Kaleidoscope Innovation in 2021. It was a prior role, not a Kyros engagement, and the end client is not named.