Low Power Wake Up Circuit Design

A battery powered product spends most of its life asleep, and the circuit that wakes it up decides both how long the battery lasts and how quickly the product responds. The two requirements pull in opposite directions, and the design is the reconciliation of them.

What Wakes a System Up

The sources of a wake up event are a real time clock alarm, an external signal from a sensor or a user, and the internal peripherals of the processor such as a comparator or a watchdog. Each has its own quiescent current and its own latency.

The real time clock is the most common source, because it keeps time while everything else is off and can be programmed to raise an alarm at a chosen interval. Its own consumption is small but not zero, and over a year it adds up.

An external signal wakes the system immediately, and the circuit that detects it has to run continuously. That detector is often the largest consumer in the whole design, which is why a mechanical switch or a passive sensor is preferred wherever it is possible.

quiescent current and Its Sources

quiescent current is the current drawn while the system is asleep, and it flows through the regulator, the processor in its lowest power state, the real time clock and every pull up resistor on the board.

A regulator with a quiescent current of a microamp is already significant for a product that runs for years on a coin cell. The leakage of the output capacitors and the reverse leakage of protection diodes add to it and are frequently forgotten.

Pull up resistors are the classic error. A ten kiloohm pull up on a line that is held low draws three hundred and thirty microamps at three point three volts, which is more than the whole sleep budget of many designs.

interrupt latency and Response Time

interrupt latency is the time between the event and the code that handles it starting to run. It includes the wake up time of the oscillator, the start up of the regulator if it was shut down, and the interrupt entry of the processor.

A processor that wakes from a deep sleep in a few microseconds still needs its clock to be stable, and a crystal oscillator takes milliseconds to start. Designs that need a fast response keep a fast internal oscillator running or use a smaller processor that never fully stops.

The latency added by the power supply is often overlooked. A regulator that has to ramp up from zero adds its own start up time, and a circuit that gates the supply to a sensor has to wait for the sensor as well.

real time clock and Its Support

A real time clock is a small counter with its own crystal or an integrated oscillator, and it keeps time from a coin cell or from the main supply with a very low current. Its accuracy is set by the crystal, and a crystal intended for timekeeping has a specified load capacitance and a tuning fork frequency of thirty two kilohertz.

The crystal for a real time clock is as sensitive to stray capacitance as any other, and its load capacitors are chosen the same way. The difference is that the accuracy of a timekeeping crystal is measured in seconds per month, so a few picofarads of error is visible.

Battery backup is usually a small cell or a supercapacitor, and its leakage is part of the budget. A supercapacitor that self discharges in a few days is useless for a product that has to keep time for a year.

power gating and Domains

power gating removes the supply from a part of the circuit rather than putting it into a low power state. It reduces the leakage to zero and it costs a switch, which has its own leakage and its own voltage drop.

A load switch has a leakage of a fraction of a microamp and a resistance of a few tens of milliohms. Both matter: the leakage adds to the sleep current and the resistance reduces the voltage available to the gated circuit.

Domains should be chosen so that the wake up logic is never gated. A design where the circuit that has to detect the wake up event is itself powered down cannot wake up, which sounds obvious and happens regularly when a design is partitioned late.

Designing the Wake Up Path

The path from the event to the processor has to be active at all times, and it should consist of as few components as possible. Every component in the path adds leakage and adds a delay.

A comparator with a built in reference is a common choice, and its quiescent current of a few microamps may be the dominant term. A part with a shutdown pin that keeps only the reference alive is a useful compromise.

Where the event is a slow change, a comparator with hysteresis prevents a burst of wakes at the threshold. The hysteresis also filters the noise from a long cable, which would otherwise wake the processor repeatedly.

Software Contribution to Power

The firmware determines how long the system stays awake after a wake up, and that time is multiplied by the active current, which is hundreds of times the sleep current. A wake up that lasts a millisecond at ten milliamps costs as much as a thousand seconds of sleep at ten microamps.

Dealing with the event quickly and returning to sleep is therefore the single most effective measure. Logging, transmission and peripheral initialisation should be arranged so that each one is done only when it is needed.

Wake up sources should be disabled when they cannot be serviced. A peripheral left enabled that wakes the processor on an event the firmware ignores produces a stream of interrupts that consume the battery without doing anything useful.

Verification and Measurement

Measure the sleep current with an instrument that can resolve microamps and with the product in its final state. A multimeter on the microamp range with a long cable adds its own burden voltage, and a measurement that looks wrong is often the instrument.

Record the energy of one complete cycle: the sleep, the wake, the measurement and the transmission. The average current is the total charge divided by the period, and that number predicts the battery life far better than the sleep current alone.

A product that meets its specification on the bench and not in the field usually has a wake source that fires more often than expected, or a temperature effect on the leakage. Logging the wake count over a week answers the question directly. The release checks that keep such a design consistent are collected in our PCB design release checklist, and the assembly points are listed in judging PCB quality. The board surface leakage that dominates at these currents is covered in our guide to conformal coating board protection.

FAQ

How low can the sleep current be? A few hundred nanoamps is achievable with a good real time clock and no pull up resistors. Anything below that needs the regulator and the board leakage to be addressed as well.

Why does my product wake up unexpectedly? A floating input, a slow edge at a comparator or noise on a long cable. Add hysteresis and a pull up or pull down that does not cost sleep current.

Does a deeper sleep always save more energy? Not if the wake up takes longer. Compare the energy of the whole cycle rather than the sleep current alone.