A van der Waals Floating-Gate Platform for Broadband Optoelectronic Neuromorphic Computing

The human brain is a master of efficiency, processing vast amounts of information while consuming very little power. It achieves this by merging memory and calculation into a single, interconnected web of neurons and synapses. In contrast, the computers we use today rely on a design that separates these two functions. Data must constantly travel back and forth between a processor that calculates and a memory chip that stores information, a journey that wastes energy and slows down performance. To build the next generation of intelligent machines, scientists are trying to create hardware that mimics the brain's architecture, where a single device can both remember information and learn from new inputs. A key component in this effort is the artificial synapse, a tiny electronic switch that can change its strength based on past activity, much like a biological connection between brain cells. While researchers have built many types of these devices, most struggle to combine long-term memory with the ability to learn from light, a capability that would allow machines to process visual information directly and efficiently.

A team of researchers at Dongguk University in South Korea has developed a new type of artificial synapse that successfully bridges this gap. They created a single transistor that can store data permanently and also learn from a wide range of light, from visible colors to invisible infrared rays. The device is built using a stack of three different atomically thin materials, layered like a microscopic sandwich. The bottom layer is a material called indium selenide, which acts as a storage vault for electrical charges. Separating this vault from the top layer is a thin sheet of hexagonal boron nitride, which serves as a secure tunnel that electrons can pass through only when prompted. The top layer is made of rhenium disulfide, a material that reacts strongly to light. When the researchers apply an electrical pulse or shine light on the device, electrons move through the tunnel and get trapped in the storage layer, permanently changing how easily electricity flows through the top layer. This change in flow represents a new memory state or a learned connection.

The researchers found that this device is exceptionally good at remembering. By applying electrical pulses, they could trap enough charge to create a massive difference between the device's "on" and "off" states, with the current in the "on" state being more than one million times stronger than in the "off" state. Once the information was written, the device held it steadily for over 10,000 seconds without needing any power to refresh it. This stability is crucial for building reliable computer memory. Beyond simple storage, the device behaves like a learning brain cell. When the researchers sent a series of electrical signals, the device's response grew stronger with each pulse, mimicking how a biological synapse strengthens when neurons fire together repeatedly. They could also weaken the connection by sending signals of the opposite polarity, allowing the device to forget or adjust its learning.

What makes this discovery particularly significant is how the device responds to light. Unlike older electronic memories that only understand electrical signals, this new synapse can be trained using light across a broad spectrum. The researchers tested it with green, red, and near-infrared light, and the device responded to all of them. By changing the brightness of the light or how long it shines, they could precisely control how much the device learned. A brief flash of light caused a temporary change, similar to short-term memory, while a longer, brighter exposure created a lasting change, similar to long-term memory. The device could even distinguish between different patterns of light and electricity. In one experiment, the researchers taught the device to associate a specific electrical signal with a flash of light. After training, the device would react to the electrical signal alone, as if it had learned to expect the light, demonstrating a form of associative learning similar to the famous experiments where a dog learns to salivate at the sound of a bell.

The team also showed that this single device could perform complex logical tasks by simply changing how it was stimulated. By adjusting the strength of the electrical and light inputs, they could program the device to act as an "AND" gate, which only produces an output when both light and electricity are present, or an "OR" gate, which produces an output if either one is present. This ability to reconfigure its own logic without changing its physical structure suggests a path toward computers that can adapt their functions on the fly. The researchers confirmed that the device's behavior was driven by the movement of electrons into the storage layer and not by other effects, using detailed measurements to rule out alternative explanations. This work demonstrates a versatile platform where memory, learning, and light sensing are unified in one compact component, offering a promising strategy for creating smaller, faster, and more energy-efficient systems that can process information the way the human brain does.