Pioneering Major Advances in Brain-Inspired Artificial Intelligence

The human brain, with its many nuances around short- and long-term memory, is roughly a million times more efficient than a computer. Thus, brain-inspired artificial intelligence is poised to be an important technology in computing. Synaptic transistors, which function like synapse connections between neurons, represent key building blocks for developing these computing systems.

Fu et al. examined synaptic transistors under single and multiple pulses and under mild (1 Volt) and strong (up to 8 V) stimuli, revealing the devices can provide a vast range of memory retention time. They also conceptually demonstrated pattern learning and memorizing functions based on experimental data.

“So far, only relatively short memory times have been achieved,” said author Aimin Song. “For the first time, our work achieved an extremely wide range of memory retention time from approximately 2 milliseconds to approximately 20,000 seconds, over seven orders of magnitude.”

Synaptic transistors generally are based on soft ion conductors called electrolytes, such as ion-liquids and ion-gels. These are, however, difficult to microfabricate into well-defined shapes and structures such as in silicon chips; and they are usually incapable of operating at voltages beyond 2 V to enable prolonged memory retention.

“We adopted a solid-state electrolyte, sputtered SiO₂, that is already familiar to and compatible with the silicon industry, to produce a solid-state synaptic transistor capable of operating at a record of 8 V, enabling a seven orders of magnitude span of memory time like the human brain,” said Song.

Such a development may facilitate artificial neuromorphic chips that process vast amounts of unstructured data, accomplish high-throughput image analysis, conduct fast machine learning, and construct artificial intelligence systems.

Source: “Synaptic transistors with a memory time tunability over seven orders of magnitude,” by Yang Ming Fu, Tianye Wei, Joseph Brownless, Long Huang, and Aimin Song, Applied Physics Letters (2022). The article can be accessed at http://doi.org/10.1063/5.0095730 .