In response to the growing demand for more energy-efficient computer systems, researchers are exploring innovative solutions in artificial intelligence (AI) hardware. Neuromorphic computing, inspired by the human brain, is a key area of focus, aiming to develop AI semiconductors that can efficiently process large volumes of data. Dr. Joon Young Kwak and a team at the Center for Neuromorphic Engineering at the Korea Institute of Science and Technology (KIST) have made strides in this direction by implementing an integrated element technology for artificial neuromorphic devices.
The team’s approach involves developing devices that mimic biological neurons and synapses, using vertically-stacked memristor devices made from hexagonal boron nitride (hBN), a two-dimensional material known for high integration and ultra-low power implementation. What sets this research apart is the integration of artificial neuron and synaptic devices with the same material and structure, providing ease of process and scalability for large-scale artificial neural network hardware.
The researchers have likened the integration of these devices to “Lego blocks,” allowing them to connect neurons and synapses seamlessly. By implementing the “neuron-synapse-neuron” structure, which represents the basic unit block of an artificial neural network, the team demonstrated hardware capable of spike signal-based information transmission, mirroring the functioning of the human brain.
The use of hBN-based emerging devices holds promise for developing low-power, large-scale AI hardware systems. Dr. Joon Young Kwak highlighted the potential applications, stating, “Artificial neural network hardware systems can be used to efficiently process vast amounts of data generated in real-life applications such as smart cities, health care, next-generation communications, weather forecasting, and autonomous vehicles.” Furthermore, the integration of these systems could contribute to reducing energy usage and addressing environmental concerns, such as carbon emissions, while surpassing the scaling limits of existing silicon CMOS-based devices. The study’s findings were published in the journal Advanced Functional Materials
