RMIT
Engineers from the Royal Melbourne University of Technology in Australia have created a tiny chip that can recognize movements, process information, and store visual memories without the use of an external computer.
It is noted that this neuromorphic chip simulates the work of the human brain. According to the lead author of the study, Professor Sumit Walia, the development paves the way for ultra-fast visual data processing when used in unmanned vehicles, autonomous robotic systems, and other technologies aimed at human interaction. Unlike traditional digital systems that consume large amounts of power, this chip uses analog processing similar to that of the human brain.
«Neuromorphic vision systems are designed based on analog processing similar to that used by our brains, which can significantly reduce the amount of energy required to perform complex visual tasks compared to digital technologies used today», — emphasizes Sumit Walia.
The device is made of molybdenum disulfide (MoS2) with a thickness of several atoms. According to one of the latest studies, the developers have demonstrated that tiny defects at the atomic level in MoS2 can be used to detect light and generate electrical signals from it. This resembles the work of neurons that are activated and exchange signals. The chip is capable of capturing and processing visual information in real time.
«This experimental device mimics the human eye’s ability to detect light and the brain’s ability to process visual information, allowing it to instantly sense changes in the environment and create memories without the need for huge amounts of data and energy. Today’s digital systems, on the other hand, are very voracious and unable to cope with the growing volume and complexity of data, which limits their ability to make the right decisions in real time», — explains Sumit Walia.
In the experiments conducted, the device recorded changes in movement during arm waving without the need for frame-by-frame fixation. After detecting changes, the device saved these events as memories.
«We have demonstrated that atomically thin molybdenum disulfide can accurately reproduce the behavior of the leaky integrating and activating fiber (LIF) neuron
One day, this development will improve response times automated vehicle systems and modern robotic systems. As emphasized by According to Sumit Walia, neuromorphic vision technology will only appear in such applications in many years to come, allowing autonomous devices to instantly respond to changes in the environment without the need for additional processing of large amounts of data.
The team is currently scaling up the experimental single-pixel device to a larger MoS2-based pixel array. The team plans to develop hybrid systems, that will combine analog technology with traditional digital electronics
The group is also exploring materials alternative to MoS2 to expand its infrared capabilities. This would make it possible to track global emissions in real time and identify pollutants such as toxic gases, pathogens, and chemicals.
The results of the study are published in the journal Advanced Materials Technologies
Source: SkiTechDaily