Oxide-based chip element merges processing and memory for neuromorphic computing
Researchers have developed a novel electronic device using oxide interfaces that integrates processing and memory functions, advancing neuromorphic computing by mimicking brain synapse behavior.
An international research team, including Professor Victor López-Richard from the Federal University of São Carlos, has created an electronic device that merges processing and memory capabilities, a key goal for neuromorphic computing. This development, published in Nature Communications, aims to replicate the integrated way the brain stores and processes information, unlike conventional computers where these functions are separated.
The device is built upon the interface between two oxides: LaAlO₃ and SrTiO₃. At this junction, a quasi-two-dimensional electron gas forms, creating a conductive channel that can be adjusted electrically. This allows the device to function not only as a transistor, controlling electrical current, but also as a memristor and a memcapacitor. Memristors and memcapacitors are components whose resistance or capacitance, respectively, depends on the history of applied signals, mirroring the behavior of biological synapses.
An unusual aspect of the device is its architecture, featuring control gates on the sides rather than the traditional top-mounted gate found in MOSFETs. Operationally, it functions in an analog manner, allowing for multiple intermediate states beyond the binary on/off of digital transistors. The memory effect is attributed to charges stored in these side gates, which modulate the conduction channel electrostatically, a mechanism distinct from the oxygen vacancy migration often seen in memory devices.
The primary innovation is termed "electronic polymorphism," meaning a single device can perform multiple functions based solely on its electrical connections. This integration of functions promises significant reductions in interconnections and energy consumption, addressing major bottlenecks in current computing. The device has demonstrated effectiveness in brain-inspired tasks like reservoir computing, pattern recognition of digits, and exhibiting synaptic plasticity, crucial for learning processes.
This development in oxide-based electronics is significant for neuromorphic computing by enabling "electronic polymorphism." Integrating transistor, memristor, and memcapacitor functions into a single device on a single substrate reduces energy and component count, mirroring biological neural networks. This is a step towards more efficient, brain-like computation, potentially impacting AI hardware and specialized processors for tasks requiring in-situ learning and pattern recognition.
Edited by the news editor with AI from the original report — please refer to the original source.