Researchers developed energy-efficient phototransistors that could help computers process visual information more like the human brain and be used as sensors in things like self-driving vehicles. The structures rely on a new type of semiconductor — metal-halide perovskites — that has proven to be highly efficient at converting sunlight into electrical energy and has shown tremendous promise in a range of other technologies.
The researchers combined perovskite nanocrystals with a network of single-walled carbon nanotubes to create a material combination for photovoltaics or detectors. When they shined a laser at it, they found a surprising electrical response. Normally, after absorbing the light, an electrical current would briefly flow for a short period of time. In this case, however, the current continued to flow and did not stop for several minutes, even when the light was switched off.
Such behavior is referred to as persistent photoconductivity and is a form of optical memory where the light energy hitting a device can be stored in memory as an electrical current. The phenomenon can also mimic synapses in the brain that are used to store memories. Often, however, persistent photoconductivity requires low temperatures and/or high operating voltages and the current spike would only last for small fractions of a second. In the new technology, the persistent photoconductivity produces an electrical current at room temperature and flows current for more than an hour after the light is switched off. In addition, only low voltages and low light intensities were found to be needed, highlighting the low energy needed to store memory.
The research provides previously lacking design principles that can be incorporated into optical memory and neuromorphic computing applications. Visual perception accounts for the vast majority of input the brain collects about the world and these artificial synapses could be integrated into image recognition systems. Such systems could potentially improve energy efficiency, performance, and reliability in applications such as self-driving vehicles.
The researchers tried three different types of perovskites — formamidinium lead bromide, cesium lead iodide, and cesium lead bromide — and found each was able to produce a persistent photoconductivity. To build a neural network requires integrating an array of these junctions into more complex architectures where more complex memory applications and image processing applications can be emulated.