How much can a computer do with borrowed energy?

Battery-free shipment tracking

I came across battery-free tracking while learning how retailers monitor shipments. I wondered how this was different from an AirTag. If both help locate things, what makes one suitable for my keys and the other for boxes moving through a supply chain?

Then I learned that these tags harvest energy from radio signals. That made me wonder if we could collect enough of that energy to light a city.

I wanted to understand how much energy was available, what we could do with it, and where the limits were.

Wiliot describes a Walmart collaboration involving millions of its IoT Pixels across the retailer’s supply chain. That scale comes from Wiliot’s announcement; it does not tell us how many individual products are tagged. Wiliot’s announcement.

Why not put an AirTag on it?

An AirTag helps you find something you’ve lost. Apple’s system combines Bluetooth, the Find My network and, with compatible hardware, nearby Precision Finding. It uses a replaceable battery. Apple’s AirTag overview.

With a shipment, finding the box is only part of the job. I would want to know if it had been sitting somewhere too long or exposed to the wrong temperature. Wiliot’s system uses battery-free Bluetooth tags, receiving equipment and cloud software to collect and process observations along the way. Its current Gen 3 description includes sub-1 GHz and 2.4 GHz energy harvesting. IoT Pixel specifications.

Comparing the price of an AirTag with a bulk tag price would leave out the equipment and services needed to make shipment tracking work. I would want to price the whole installation.

The energy has to come from somewhere.

An antenna receives RF energy, a rectifier converts it to DC, and stored energy supports a short operation. The device collects energy until it has enough to take a reading and send a message.

Wiliot describes bridges that energize Pixels and collect their transmissions, with gateways forwarding information to its platform. The tag can run without a battery because powered equipment around it supplies energy and receives its messages. A phone or Wi-Fi access point alone may not provide what the installation needs. Network infrastructure.

You can see that wait in the model below: a capacitor must accumulate enough energy before each report. Move the source farther away and the wait grows. Turn the source off and the model stops harvesting energy.

The model simplifies the radio environment. In an installation, turning the antenna or putting the tag inside different packaging could change the result. Walls and other obstructions would matter too. The graph shows how energy builds up between reports and drops when the device sends one.

So why couldn’t it light a city?

A tag can wait between reports while it collects energy. A light needs power the whole time it is on. We could store energy and turn the light on later, but we would still have to collect enough to keep it running.

Suppose, for this calculation, the RF power density is 0.1 µW per square centimetre. That is 0.001 W per square metre. At 40% collection-to-DC efficiency, a 10 W electrical load would require 25,000 square metres of collection area—even before asking whether the same field could exist across that whole area.

Even that calculation assumes the same RF power density across the entire collection area. Actual conditions would vary. And if we installed transmitters to supply the energy, we would have to power those first. At these assumed levels, collecting radio energy to run city lighting would require an enormous area.

The calculator uses the electrical power a lamp draws, in watts. Its brightness in lumens is a different measurement and depends on the light’s spectrum. NIST’s explanation of photometry.

How much would better computing help?

That made me wonder how much more a tag could do if computation needed less energy. It would still have to power its sensor and radio.

Suppose an operation costs 90 microjoules: 5 for sensing, 10 for computation and 75 for transmission. These are example values, not measurements from a Wiliot tag. Make computation ten times more efficient and the total falls to 81 microjoules. The total energy saving is 10%.

In this example, the radio uses most of the energy. Sending fewer messages could save more than improving computation. If we wanted more frequent reports instead, we would also have to account for other tags sharing the radio channel. A faster processor would not give them more bandwidth.

Landauer’s principle provides a lower bound associated with irreversible information erasure. That limit alone cannot tell us how much better a tag could become. We would need to account for its memory, sensors and radio as well as its computation. Shrinking the processor would also leave the antenna’s size and performance to consider. IBM’s history of Landauer’s work.

What would we ask ordinary objects to notice?

If a shipment regularly waits too long at one depot, more frequent readings could help locate the delay. Temperature readings could help someone decide which cases to inspect first. I can see why a retailer would want that information.

I also wonder what we could do with reusable containers. Could we follow them through several trips and find out where they keep getting held up? At home, could packaging help us keep track of what is left inside? That would need sensors suited to the contents and equipment to read them. I would want to know how reliably it worked before depending on it.

Could a retailer still read a package’s tag after I bring it home? If so, what could they learn about me? I would want a way to stop it being read. Selling or giving the object to someone else would raise the same problem for its next owner.

Wiliot describes encryption and privacy measures in its system. I would want to check who could access the records and how long they would be kept. Encrypting a message would not answer those questions. Wiliot’s privacy overview.

And what if a reading never arrives? That might reflect inadequate energy, a failed gateway, interference, shielding or a damaged tag. A missing reading alone cannot tell us whether something was stolen.

Try the shipment scenario: the tag went silent. You can check the gateway, nearby tags and the package itself. Even after those checks, there may not be enough evidence to explain the missing reports.

I would also want to know what happens to the tags when the packaging is thrown away. Avoiding batteries helps with maintenance, but large deployments would still use materials and powered equipment. Those costs need to be counted too.

Could the tag decide what to send?

There is another possibility I want to look into: doing more of the processing on the tag itself.

A sensor could send every reading, apply a simple threshold, or use a small model to recognise a pattern worth reporting. I want to compare the energy those approaches use and see what each might miss. If a model filters out an event, we may never get the reading that would tell us it made a mistake.

I’m saving that for a separate essay on AI miniaturization. I want to understand what a small model could do with the energy and memory available on a device like this.