Scientists led by Professor Nazek El-Atab from the Smart Advanced Memory Devices and Applications (SAMA) Laboratory, School of Computer Electronics and Mathematical Science and Engineering, King Abdullah University of Science and Technology (KAUST), Thuwal, Saudi Arabia, have developed a MOSCap device using hafnium diselenide (HfSe2) that mimics neuron-like adaptive behavior and memory retention . Their findings were published in a recent study in the journal Light Science & Applications .
Separating sensing, processing, and memory operations makes traditional computing systems difficult to dynamically adapt, resulting in latency and excessive power consumption. Neuromorphic computing offers a viable approach to faster, more energy-efficient, and more adaptable data processing by emulating biological brain networks.
Neuromorphic devices can overcome traditional architectural limitations by integrating sensing, processing, and memory capabilities in a single device. Artificial neurons and synapses are often implemented using materials and optoelectronic devices with programmable electrical properties, providing a flexible platform for developing innovative computing solutions.
Their findings add to the expanding field of neuromorphic technologies, which aim to mimic the brain’s highly efficient data-processing and adaptive capabilities.
The researchers achieved this by incorporating two-dimensional HfSe2 nanosheets into the MOSCap structure , which, as the researchers note, enables the device to ” sense and store optical information both by charge trapping and by operating as a memory capacity within the same MOSCap device, with threshold voltage and capacitance varying with light intensity . “
Electrical characterization experiments have shown that the memory window is large, memory retention is strong, and data stability is maintained even under harsh conditions such as high temperatures.
The researchers noted that “the device’s memory window remained above the failure threshold for 106 seconds at temperatures between 60 and 80 degrees Celsius,” demonstrating reliability in real-world scenarios.
Due to an efficient charge retention mechanism, MOSCap has demonstrated the ability to retain data even after the light stimulus is removed, demonstrating its potential as an energy-efficient nonvolatile optoelectronic memory.
According to the scientists, the MOSCap architecture allows for the reconfiguration of the device by “adjusting the volatility of the capacitor based on the bias conditions, allowing a switch from volatile optical sensing to non-volatile optical data storage . “
According to the KAUST researchers, this is a major step forward in the development of neuromorphic devices, demonstrating the capabilities of optogenetic synapses, enabling “stimulus-associated learning,” and ” the device’s ability to respond to light across the visible spectrum is remarkable.”
A major advantage of this refinement is the use of capacitive synapses that operate in the charge domain, which results in lower power consumption and leakage current compared to memristive synapses. According to KAUST researchers, capacitive synapses require very little static power, allow 3D stacking, and have negligible sneak-path current leakage, making them ideal for small, dense memory applications.
The researchers propose a particularly interesting application of the adaptive MOSCap in astronomy: detecting exoplanets using changes in light intensity. By incorporating the device into a leaky integrate-and-fire (LIF) neuron model, the researchers demonstrate that the MOSCap can change its firing pattern in response to light fluctuations, potentially simplifying the detection of exoplanets passing by distant stars.
“These dynamic optogenetic neurons show an extraordinary ability to detect exoplanets based on their light intensity,” the researchers note , highlighting their integration of the neurons into a spike-generating neural network (SNN).
The MOSCap device’s multiple functionalities represent a major achievement in the field of neuromorphic technology that has the potential to spur future innovation in the development of artificial systems that can dynamically respond to environmental signals and learn in a manner similar to biological neurons.
This research was funded by the King Abdullah University of Science and Technology (KAUST) Foundation Fund and the KAUST Semiconductor Transformation Award (award number FCC/1/5939).
Journal References:
Koolen, CD, et al. (2024) Scalable synthesis of Cu cluster catalysts by spark combustion for electrochemical conversion of CO2 to acetaldehyde . Photoscience and Applications . doi.org/10.1038/s44160-024-00705-3
sauce:
Chinese Academy of Sciences
