Several days ago, I was researching some old computers, machines that required substations' worth of power. Now I'm thinking about systems that run for years off a single coin cell. How much things have changed!
Ultra-low power embedded systems can run off a CR2032 for ten years, if they spend most of that time in a deep sleep, waking only occasionally and for brief periods to do something useful. But it's not easy to squeeze the last coulomb from these small cells.
We know that a processor's power consumption is proportional to the voltage squared times the operating frequency. But we're not really interested in power, since battery capacity is rated in amp-hours. Current is the issue, and current in a resistive network is linear with voltage.
Is it linear with an active (very active!) component like a microprocessor? Here's what the datasheet says about Microchip's PIC18LF46K22, a typical very low power part:
TI's MSP430F2013 also exhibits the same behaviour, though, it's unclear if this is the pretty-much-meaningless "typical" data or worst-case:
This linear Vdd vs current data is a bit puzzling. Depending on clock rate, running at a lower Vdd results in substantial energy reductions. A number of vendors are locked in a battle for dominance in the ultra-low-power space. One would think they'd put a regulator on-board to operate the chip at the lowest possible Vdd, optimising their current numbers. Obviously, there may be some pin issues, though those could be powered pre-regulator.
The processor will, in these applications, mostly be sleeping. But we really don't care much about the relationship between Vdd and current when it's in a snooze mode. For a ten year life from a CR2032 the average current draw over that decade cannot exceed 2.5 uA. With sleep currents of very-low power MCUs in the tens to hundreds of nanoamps, sleep is practically irrelevant to these issues.
To reduce the processor's needs why not add an external LDO (low drop out) regulator?
Unfortunately, during the processor's long periods of sleep the typical LDO will require tens or more of microamps. I can't find one that is frugal enough with coulombs.
Touchstone Semiconductor has a boost converter that looks like a perfect solution. Their TS3310 will do all sorts of cool things to crank low voltage sources up to MCU levels. But it will also regulate a battery down to lower levels. When the processor is sleeping it only needs 180 nA from the battery, which is just 7% of the 2.5 uA budget. Regulate from 3 to 1.8V and you'll gain, even accounting for the regulator's needs, around a third more energy. Using the regulator will boost the amount of current a system can consume by around a third for the same battery lifetime.
The 2mm square IC itself is $0.99 in 2000-unit quantities. It does require a small inductor, but those are about 7 cents in volume. A couple of capacitors are needed as well, which would also be the case with an LDO.
It may be tempting to assume one can use this trick to build a system that will run for more than ten years off a coin cell. Alas, that's not to be. The vendors specifically claim a ten-year life on their batteries. It was just last summer that Duracell changed their ratings from seven to ten. If a part supplier says "don't use this part in this mode" (in this case for more than a decade) it's poor engineering practice to ignore the warning. Sure, the system might work, but it's a crap shoot.
Of course, a cynic might note that he will probably not be employed at the same place after all those years go by, so can't be held accountable...
(Note: Touchstone has fallen on hard times. Silicon Labs recently acquired their IP and will hopefully continue selling this part.)