Photovoltaic Cells and Intermittent Computing

By Louis Case

Ask whether a solar cell can replace a computer chip and the answer is a flat no, for a reason worth stating plainly because it clears the ground for the more interesting question. A photovoltaic cell is a transducer. It turns light into a voltage, and the voltage it produces rises and falls in proportion to how much light lands on it. A transistor is a switch. Its whole usefulness is that it snaps between two discrete states, on and off, one and zero, and holds them cleanly enough to build logic on top. Light intensity is a smooth analog quantity. Computation is built from crisp digital ones. You cannot make a solar cell be a logic gate any more than you can make a thermometer be an abacus. They answer different questions.

So the naive substitution fails. But there is a second question hiding behind the first, and it is one of the genuinely elegant ideas in embedded systems. Not can a solar cell be a computer, but can a computer be built to live on nothing but what a solar cell happens to catch. That question has a rich and growing answer, and it goes by the name intermittent computing.

Designing around failure instead of against it

The ordinary way to power a device from an unreliable source is to fight the unreliability. Put a battery in. Size a capacitor for the worst-case dark stretch. Engineer the storage large enough that the computer never notices the supply flickering. This works, and it is how nearly everything is built, but it carries a cost that becomes intolerable at scale. The battery. Batteries are expensive, they are environmentally nasty, they degrade, and eventually they must be replaced. And when a device is a sensor glued inside a bridge or scattered by the thousand across a field, replacement is not a service call. It is the end of the device's life and the beginning of its life as waste.

Intermittent computing inverts the whole premise. Instead of hiding the intermittency of a harvested source, it embraces it. The device harvests light into a small capacitor, waits until it has banked enough charge for a short burst of work, computes furiously for as long as the charge lasts, and then, this is the part that sounds absurd until you see it work, simply powers off when the energy runs out, and waits to recharge and resume. Power failure is not an error condition. It is the normal operating cycle, arriving many times a second.

The obvious objection is the one everyone raises first. If the power keeps dying mid-calculation, how does the thing ever finish anything, or avoid corrupting itself when it stops in the middle of writing to memory? That objection is exactly what turns a party trick into a research field. The answer is a family of techniques for state retention across power loss. The device writes checkpoints of its progress into non-volatile memory, FRAM or MRAM, memory that keeps its contents with no power at all, so that when the lights come back on, it can pick up precisely where it left off rather than starting over. Whole systems have been designed around this single problem. Mementos and Hibernus are the canonical early ones, architectures that decide when to snapshot and how to resume so that forward progress is guaranteed even though the power is not. The device also needs brownout detection, a circuit that senses the voltage falling and triggers a clean shutdown a hair before the collapse, so the checkpoint is never taken mid-write. Get these pieces right and you have a computer that treats a thousand power failures a second as nothing more than a stutter.

The battery-free proof

If this sounds theoretical, the most charming demonstration is a video game. In 2020 a team from Northwestern and TU Delft built a battery-free Game Boy, a real, cartridge-playing Game Boy with no battery in it at all, powered by small solar cells and by the kinetic energy of the player mashing the buttons. It loses power constantly. On a moderately cloudy day with active play, it browns out for less than a second in every ten. And it keeps playing, holding your position in Super Mario Land across each blackout by saving state to non-volatile memory the instant the voltage drops. It is a toy, deliberately, because a toy makes the point unmissable. Here is a device with no battery, harvesting ambient energy, computing through relentless power interruption, and it works well enough to play a platformer. The researchers were blunt about why it matters. Batteries are costly, hazardous, and destined for a landfill, and a world of trillions of small connected devices cannot be a world of trillions of batteries to mine, ship, and eventually bury.

That last figure is the real stakes. The vision of ubiquitous sensing, structural monitors in every bridge, environmental sensors across every field, tags on every object, collapses under its own logistics if each node needs a battery that a human has to change. The "Internet of Batteryless Things," as the researchers who work on this call it, is not an aesthetic preference. It is the only version of pervasive sensing that is physically and environmentally sustainable at the scale being proposed.

I have argued before, in the context of defense infrastructure, that scale without resilience is just a larger attack surface. The same logic governs hardware. A network of a trillion nodes, each depending on a battery that a person must eventually reach and replace, is enormous scale resting on a single brittle assumption, and the first dead cell in an inaccessible place is where that scale starts becoming waste. The batteryless design is not a trick to save money on power. It is what makes the scale resilient in the first place, by removing the one component whose failure was guaranteed.

Why photovoltaics are the right harvester

Among the ways to scavenge ambient energy, vibration through piezoelectrics, temperature gradients through thermoelectrics, stray radio waves through rectennas, photovoltaics have a particular fit to this model, and it is worth being precise about why. Solar cells have no moving parts, they scale down cleanly to tiny form factors, and, most importantly, their energy is intermittent by its own nature. Light comes and goes with the day, with shade, with a hand passing over a sensor. A system that is already architected to expect irregular power is not fighting the sun's variability. It is designed in harmony with it. And the harvesting no longer depends on standing a panel in direct sunlight. A wave of recent work on indoor photovoltaics, much of it in perovskite materials, is pushing the efficiency of cells under ordinary indoor lighting to the point where a sensor can run on the light already in the room, harvesting from a ceiling fixture the way an older device drew from a coin cell. Cells tuned for indoor spectra have begun outperforming, under indoor light, the silicon panels optimized for the roof.

I will resist overselling it, because the honest boundaries matter. Intermittent computing is not a general-purpose replacement for the machine on your desk. It is for the low-duty-cycle edge, the device that wakes, senses, computes a little, radios a result, and sleeps. Anything that needs continuous power or low latency still wants storage and, often, a battery. The checkpointing overhead is real, and a workload that never gets an uninterrupted stretch long enough to make progress simply will not run. This is a technique with a shape, and the shape is tiny, occasional, patient computation at the ragged edge of the network.

But that edge is where the coming decades of computing are quietly headed, into the walls, the infrastructure, the objects, and the batteries that would otherwise power it are the part that does not scale. The elegant answer is not a better battery. It is a computer redesigned so completely around the rhythm of a fickle energy source that it no longer needs one, holding its thoughts through each small darkness and resuming when the light returns. A photovoltaic cell cannot be a computer. But it can, it turns out, be the only thing a certain kind of computer ever needs.

References

1. de Winkel, J., Hester, J., Pawełczak, P. et al. "Battery-free Game Boy: Sustainable Interactive Devices." Northwestern University & TU Delft, 2020. https://news.northwestern.edu/stories/2020/09/battery-free-game-boy-runs-forever
2. "The Internet of Batteryless Things." Communications of the ACM. https://dl.acm.org/doi/10.1145/3624718
3. Ransford, B., Sorber, J., Fu, K. "Mementos: System Support for Long-Running Computation on RFID-Scale Devices." (Foundational intermittent-computing work.)
4. Balsamo, D. et al. "Hibernus: Sustaining Computation During Intermittent Supply for Energy-Harvesting Systems." https://www.researchgate.net/publication/273395519
5. "Fast and Energy-Efficient State Checkpointing for Intermittent Computing." ACM Transactions on Embedded Computing Systems. https://dl.acm.org/doi/10.1145/3391903
6. "Reliable perovskite indoor photovoltaics for self-powered devices." National Science Review, 2025. https://academic.oup.com/nsr/article/12/8/nwaf242/8161003
7. "Indoor Photovoltaics Spark a Sustainable Revolution." Tech Briefs, 2026. https://www.techbriefs.com/component/content/article/55545-indoor-photovoltaics-spark-a-sustainable-revolution
8. "A Survey of Prototyping Platforms for Intermittent Computing Research." ACM. https://dl.acm.org/doi/pdf/10.1145/3698384.3699612