In what might be the lowest power sensor ever deliberately made, engineers at Washington University in St Louis have used quantum tunnelling as a data logger to keep a record of vibration.

UWashStL-QTSensor
The recording device is a floating gate in a modified transistor structure.
When charged, the floating gate is designed to gradually discharge through Fowler-Nordheim tunnelling. Remaining charge on the floating gate can be deduced at any time by powering the transistor’s channel and measuring channel resistance – like an analogue version of flash memory.
“By building the barrier in a certain way, said Professor Shantanu Chakrabartty, “you can control the flow of electrons. You can make it reasonably slow, down to one electron every minute and still keep it reliable.”

The gradual discharge carries on at a predictable rate whether the read-out transistor is powered or not, with the floating gate taking days to empty through tunnelling.
If charge is coupled into the floating gate form an external source through a capacitor, it puts a small predictable measurable step in the discharge curve – once again, the step is the same whether the read-out transistor is powered or not.
Due to non-linearities in the system (whose equations are known), if the same amount of charge is then removed capacitively, the floating gate returns to close to, but not exactly the same as, the original tunnelling discharge curve – the equal charge and discharge has left a permanent mark on the curve, that remains on the transistor for days.
In use, the floating gate of the transistor would be programmed to a known state (taking about 50 million electrons, according to the University), then left un-powered for hours or days to be affected by an external quantity – the engineers used a piezo vibration sensor as a source of capacitively-coupled charge in the proof-of-concept – and then the transistor could be powered and read to reveal something about the sensor during the un-powered time.
That something is a tricky function of (at least) sensor signal amplitude and time – what extras need to be done to make realistic use of this data is under investigation.
“For some applications, this final result is sufficient,” according to the university. “The next step for Chakrabartty’s team is to overcome the computational challenge of more precisely recreating what happened in the past.”
Multiple devices running together to gather more data points might help. “The information is all stored on the device, we just have to come up with clever signal processing to solve this,” Chakrabartty said.
One thing that does not have to be calculated out of the result is the influence of temperature – even though F-N tunnelling is significantly temperature dependent – as the sensing transistor in the demonstration device is paired with an identical twin whose floating gate is protected from external capacitive influence. To cancel temperature effects, the pair of transistors are read simultaneously through a differential signal chain.
Power consumption is limited to setting the transistors to an initial state, waiting while results accrue with out additional power, then reading them hours or days later. In the proof of concept, power for the initial setting was derived from the piezo sensor.