Table of Contents
An RTC can retain time through a power interruption and still fail the product requirement. The clock may drift too far before the next synchronization, the backup current may exceed the battery budget, or firmware may treat an invalid timestamp as valid after startup.
Select a real-time clock by separating those three tasks: time accuracy, backup operation, and recovery of trustworthy time. The example calculations below show how to turn a product requirement into a purchasing specification.
The offline time-error budget
The allowable accumulated time error determines the oscillator requirement. For a constant frequency error:
Time error in seconds = ppm × 10⁻⁶ × elapsed seconds.
Over 30 days, 1 ppm produces 2.592 seconds of error. If a product must remain within ten seconds for that interval, the frequency-error allowance is about 3.86 ppm before allocating other error sources.
That budget also needs synchronization uncertainty, aging, voltage effects, and temperature behavior. A nominal room-temperature specification cannot represent an outdoor sensor’s entire environment.
The DS3231 product information, checked September 15, 2026, specifies ±2 ppm over 0°C to +40°C and ±3.5 ppm over −40°C to +85°C. A constant 3.5 ppm offset would correspond to about 9.07 seconds in 30 days. That conversion illustrates the scale; it is not a complete end-product accuracy guarantee.
Oscillator architecture and temperature range
An RTC using an external crystal introduces crystal selection, loading, layout, and assembly variables. An integrated crystal simplifies some of those dependencies. Temperature compensation can reduce temperature-related error, but its limits and power cost still need evaluation.
| Design need | Selection question | Evidence to retain |
|---|---|---|
| Long disconnected operation | How much accumulated error is allowed? | Error budget over the full interval |
| Large temperature variation | What accuracy is specified across that range? | Exact grade and temperature conditions |
| Very low backup consumption | Which functions run from the backup supply? | Backup-current conditions and feature settings |
| Small manufacturing variation | Is the resonator integrated or external? | Crystal/load requirements and calibration process |
| Field synchronization | How is time updated without inconsistent reads? | Firmware sequence and timestamp policy |
For a clock-source comparison beyond calendar timekeeping, see the precision oscillator guide. An RTC adds state and calendar behavior that a frequency source alone does not provide.
Backup charge demand
Suppose an illustrative RTC backup mode draws 1 µA for 365 days. The ideal charge requirement is 8.76 mAh: 0.001 mA multiplied by 8,760 hours.
That number excludes battery self-discharge, temperature effects, usable capacity at the system’s cutoff voltage, leakage elsewhere on the board, and any additional active functions. It is a charge-budget input, not a predicted service life.
Use maximum or suitably bounded current values for the actual configuration. Include protection and switching components in the board-level leakage review. Verify whether the design charges its backup source and whether the chosen battery or capacitor is suitable for that circuit.
Power-loss and recovery behavior
The DS3231 datasheet describes battery switchover and oscillator-status behavior. Other RTCs may handle those events differently, even if they share an I2C interface.
An application-level test plan should include main-power removal, slow brownout, repeated startup, absent or depleted backup supply, and initial assembly with no valid time set. Verify the status bits and the software response to invalid time.
Check alarms and outputs separately in main and backup modes. A function working while the host is powered does not establish that it behaves the same after switchover. Avoid clearing a status indication before firmware has recorded why the timestamp may be invalid.
The approved RTC configuration
Capture the exact RTC ordering code, temperature grade, resonator arrangement, backup-source requirements, and firmware driver revision. Preserve calibration settings and the procedure used to initialize the time.
For an IoT production BOM, a replacement RTC should pass both electrical checks and timestamp-recovery tests. The IoT sourcing solution can be supported by that controlled record.
Acceptance is complete when the product keeps time within its defined budget and recognizes when it cannot. Calendar registers continuing to count are only one part of the requirement.
Frequently Asked Questions (FAQ)
How much clock error does 1 ppm produce?
A constant 1 ppm frequency error accumulates 0.0864 seconds per day, or 2.592 seconds over 30 days. Actual RTC error also depends on temperature, aging, voltage, and calibration.
Does an RTC backup battery keep the whole board running?
Normally it powers only the functions defined by the RTC's backup mode. Check whether alarms, outputs, and the serial interface operate in that mode; do not assume the host MCU remains powered.
Can two I2C RTCs be exchanged without firmware changes?
Not by interface name alone. Register maps, addresses, data formats, status flags, alarms, and power-loss behavior must all match or be handled by revised firmware.
