Jack Ganssle is an internationally recognized embedded systems engineer, author, and speaker. He has published more than 700 articles on different aspects of embedded development as well as six books on the subject. Ganssle lectures at symposiums and conferences on electronics and software all over the world.
Ultra-low-power design is all about current: You need to provide the required operating current for long periods at a sufficient voltage, typically from a small battery. In Jack Ganssle’s paper “Hardware and Firmware Issues in Using Ultra-Low Power MCUs,” a 10-year operating life was assumed for designs using a CR2032 lithium coin cell with a 2-V cutoff. To put the problem into perspective, the average current for that period of time cannot exceed 2.5 µA.
When designing for extremely low current levels, secondary effects become very important. For example, a decoupling capacitor must be chosen for its shunt resistance rather than the series resistance and inductance that would be more important in a high-speed application. Similarly, test setups and measurement techniques must be designed to maximize any impedances shunting the battery to minimize leakage current.
Specifications
Incomplete, inconclusive, or misleading specifications further complicate the job of the ultra-low-power designer. Ganssle’s paper is particularly relevant because he performed a series of tests to determine realistic values for several parameters. Some conclusions are very easy to understand. Claims that a certain MCU can run for decades in sleep mode, for example, are irrelevant because coin-cell manufacturers only project a maximum 10-year life.
Beyond 10 years, chemicals may deteriorate and batteries become unreliable. Besides, you’re interested in the work the MCU does while it’s awake, not while it’s asleep. Other factors, such as how an MCU should be operated during the brief time it’s active, are more complex. Almost all low-power design recommendations are based on battery fundamentals.
Typical specifications
Ganssle’s views are succinctly summarized in his paper. “The bottom line is that the meaning of a ‘typical’ spec is devoid of engineering value. If a spec isn’t guaranteed, it’s not a spec, it’s an aspiration.”
Batteries
Capacity
Ganssle presents a graph captioned “mA available as a function of sleep time and sleep current” that assumes a 10-year CR2032 battery life and a 225-mAh capacity, verified through actual 0.5-mA discharge tests. The horizontal axis represents duty cycle, 3% at the left edge and 0.3% at the last labeled gridline. Sleep current is the amount of current that the entire system consumes when inactive, so it includes leakage current as well as true MCU sleep current.
Even with zero sleep current, expanding the right-hand end of the graph shows that 10-mA active current only is available if the system is inactive at least 99.975% of the time, equivalent to being active for a little less than one second per hour. It’s also easy to see that 20-nA sleep current doesn’t improve the situation much beyond that for 200 nA or even 1 µA. The important relationship is between the duty cycle and the required active-mode current because it is orders of magnitude larger than the sleep current.
Dynamic effects
Batteries are complex electrochemical systems, although to a first order they can be modeled by an ideal voltage source with a series output resistance. This means that the voltage delivered to the load will be lower than the battery voltage by the product of the load current and battery impedance—the internal resistance (IR). Importantly, as the battery discharges, its impedance increases, and the load voltage can easily drop below the 2.0-V cutoff well before the battery charge is depleted.
A graph captioned “Internal resistance and its standard deviation” shows how the IR of three brands of CR2032 coin cells increased from about 10 Ω to 60 Ω as they were discharged. It includes curves for the average IR and the standard deviation for each brand—Ganssle tested several batteries of each type.
A graph captioned “Voltage delivered from battery depending on load” starts with an overall average IR curve and adds 3x a representative standard deviation to display the worst-case effective loss of capacity. With a 10-mA load, only 88% of the battery’s capacity—about 200 mAh—is available before the load voltage drops below 2.0 V. Clearly, the situation gets much worse for higher current loads.
In addition, there is a second-order electrochemical effect that Ganssle noticed when recording load voltage as the load was switched on—how does the battery behave when the load suddenly increases from low current to 10 mA or 30 mA? For this test, he discharged several batteries at a constant 0.5-mA rate with a step increase to 10 mA for one second after two hours and then to 30 mA for one second after four hours. The cycle repeated every four hours.
The final graph in the paper shows that the IR was lowest immediately after the load increase but within several milliseconds had significantly increased. The effect was largest near the end of life where the IR already was high. Although the exact cause of the phenomenon isn’t clear, as Ganssle concluded, you need to figure on another 10% reduction in capacity because of it.
This effect comes into play along with basic IR limitations if you try to run an MCU very fast for a short period of time as some manufacturers recommend. At higher speed, the current increases, the load voltage drops, and the effective battery capacity reduces even further. Ganssle recommends coming out of sleep mode at a slower MCU clock rate and electing to increase speed or not based on measured load voltage. The object is to run the system for as long as possible without crashing, which could occur were it to immediately begin running at a high speed.
Leakage
PCB
Ganssle conducted a series of tests on bare PCBs that indeed proved humidity and contamination could dramatically increase leakage current. However, when these tests were repeated on a clean board that had solder mask applied, the effects were negligible. So, although solder mask won’t be applied to solder joints, you can design a PCB to largely avoid additional unwanted battery drain.
The section in the original paper on measuring leakage current is well worth reading. Ordinary general-purpose test equipment is of little use at subpicoamp levels, so Ganssle built a sample-and-hold circuit with femptoamp-level leakage. With this tool, the effect of various contaminants on bare PCB traces could be compared. How he verified the circuit’s actual performance is straightforward but an important detail.
Capacitor
Capacitors leak, especially the large value ones recommended for use with some MCUs. The best low-leakage capacitors with the C values needed are MLCC. MLCC leaks are specified in ohm-farads, as shown in Table 1.
It’s clear that Y5V dielectrics can’t be used because their leakage is too high. You must select a more expensive X7R device. And, be careful which X7R is selected because some leak as badly as Y5Vs. It turns out that MLCC leakage is a function of temperature and voltage, in particular the ratio of the applied voltage to the rated voltage.
Kemet’s Application Notes for Tantalum Capacitors shows how derating the voltage can help. By operating at a third of the rated voltage, the leakage typically will decrease by an order of magnitude. Go to 10% of the part’s rated level and leakage dives by another factor of five. But in a 3-V system that means a 30-V capacitor—bigger and pricier. Applied voltage also affects actual capacitance as shown in Figure 5.
This graph includes the usual 20% tolerance for large value X7R MLCCs and shows that applied voltage has much less effect on capacitors with large case sizes than for small cases sizes. All these effects add up to requiring a larger, higher value, and more expensive capacitor.
In addition to decoupling, large value capacitors often are recommended by MCU manufacturers as a way to source current for pulses 10 ms or more in length. The capacitor may have a much lower ESR than the battery’s IR, but this advantage is offset by the effect of leakage on the overall system operating time. Ganssle concluded that trying to boost current in this way is impractical.
Conclusion
Extremely low-current design is painstaking. In addition to the topics already covered, Ganssle’s paper discusses several topics also related to battery capacity and leakage:
- Dropping Vdd to cut current and increase battery life
- The high current drain of an MCU’s brownout reset circuit
- Reverse battery protection
- UL rules for use of coin cells
All of these factors either reduce the available battery voltage or increase the leakage current, making ultra-low-power design even more difficult. However, there are a few straightforward things you can do, such as using a battery holder that mechanically eliminates the possibility of reversed polarity. That approach has no effect on either the load voltage or system operating life.
Finally, Ganssle comments that given all the performance limitations he actually measured, it’s unlikely that a coin-cell-powered system can achieve a 10-year operating life.