Memory model of the synaptic device. Higher repetition rate of information input causes formation of long-term memory (red line), while lower repetition rate forms short-term memory (blue line) and does not cause formation of long-term memory. The memory level is basically unchanged by the initial few inputs, and corresponds to sensory memory. The results of the device operation are in good agreement with this memory model, showing that the synaptic device can accurately reproduce the multistore model of human memory proposed in psychology.
A Japanese research group of WPI MANA, NIMS, in joint work with UCLA, succeeded for the first time in the world in the development of a new device called a “synaptic device,” which autonomously reproduces two phenomena which are distinctive features of the neural activity of the brain, namely, “memorizing necessary information” and “forgetting unnecessary information,” with a single device.
Present-day computer systems are considered to be approaching the limits of higher performance, the development of neural network systems and brain-type computers is seen as necessary in achieving even higher performance. The newly-developed device is equivalent to a biological synapse, which is essential for realizing this goal. The human brain memorizes information more reliably with the more frequent rehearsal, and conversely, infrequent rehearsal only forms vague memories to be forgotten. It is thought that these mechanisms are realized by changes in the strength of synaptic connections in the brain.
The synaptic device developed in this research is capable of adjusting its own strength of a synaptic connection depending on the frequency of electrical stimulation. Even when the strength of the stimuli and number of repetitions of the stimulation are the same, frequent stimulation causes a strong synaptic connection maintained persistently (i.e., the long-term memory), whereas the infrequent stimulation only causes a temporal increase in the synaptic connection. It was found that this kind of device operation is in good agreement with the changes in the strength of synaptic connections in the brain.
The synaptic device comprises a metal electrode and an ionic/electronic mixed conductor electrode. Using ion movement, which depends on the input frequency of electrical stimuli, the researchers succeeded in controlling the state of the metal atom bridge formed between the electrodes, in other words, the synaptic strength of the device.
Artificial reproduction of synapses, which are key constituent elements in neural circuits, is indispensable for neural network systems and brain-type computers. Conventional artificial synapses, which had been realized by complicated circuits and software, can only operate as designed in advance. Because the new synaptic device will enable diverse operations without prior operational design, it is expected to contribute to the construction of artificial intelligence which becomes wiser with experience, in precisely the same way as humans.
Inorganic synapse showing STP and LTP, depending on input-pulse repetition time. a, Schematics of a Ag2S inorganic synapse and the signal transmission of a biological synapse