Main Parameters of Strengthening Synaptic Appliances

The disadvantage of the high power consumption of the traditional von Neumann computing method was overcome by the development of neural computing systems, which simulate the human brain.

Main Parameters of Strengthening Synaptic AppliancesArticle concept image. Image credit: Korea Institute of Science and Technology (KIST).

Implementation of a semiconductor device using brain information transfer technology requires a high-performance analog synthetic synapse device that can express different synaptic connection strengths. This technology takes advantage of the signals sent when neurons produce a spike signal.

When considering conventional variable-resistance memory devices frequently used as synthetic synapses, the electric field builds as the filament grows with a fluctuating resistance, generating a feedback phenomenon that leads to rapid filament growth.

As a result, it is difficult to incorporate too much plasticity and maintain analog (gradual) resistance variability depending on the type of filament.

Dr. YeonJoo Jeong’s team at the Neuroengineering Center in korean institute of science and technology Overcome long-standing problems with analog synaptic properties, plasticity, and information preservation in memristors and neuronal semiconductor devices.

Announced the creation of an artificial synaptic semiconductor device capable of relying on highly neural computing.

The performance of current neuron semiconductor devices was hampered by the modest synaptic plasticity, which the KIST research team addressed by modulating the redox properties of active electrode ions.

In addition, several transition metals have been used in the synaptic apparatus to alter the reduction potential of active electrode ions. The high reduction potential of ions has been shown to be a critical factor in creating high-performance synthetic crosslinked devices.

Therefore, the study team added a titanium transition metal with a high ion-reduction potential to an already existing synthetic crosslinker.

This preserves the analog properties of the synapse and the plasticity of the device at the biological synapse of the brain, which is approximately five times the difference between high and low resistance.

The team also created a high-performance neuron semiconductor that is about 50 times more efficient.

Compared with the current synthetic crosslinking device, the information retention is enhanced up to 63 times due to the higher alloy formation reaction that includes the titanium-doped transition metal. More accurate simulations of certain brain processes, such as long-term depression and potentiation, can also be performed.

Using the artificial synaptic device they created, the team attempted to implement an artificial neural network learning pattern for image recognition. As a result, compared with the current synthetic synaptic device, the error rate has been reduced by more than 60%.

In addition, the accuracy of handwriting image pattern recognition (MNIST) increased by more than 69%. By optimizing the synthetic synaptic apparatus, the research team demonstrated the feasibility of a high-performance neural computing system.

This study significantly improved the synaptic range of motion and information preservation, which were the greatest technical barriers to existing synaptic mimicry. In the developed artificial synapse device, the analog operating area of ​​the device has been maximized to express different connection strengths in the synapse, thus the performance of AI computing based on brain simulation will be improved.

Dr. YeonJoo Jeong, Senior Researcher, Center for Neuroengineering, Korea Institute of Science and Technology

Dr. Jeong added, “In the follow-up research, we will manufacture a neuron semiconductor chip based on the developed artificial synapse device to realize a high-performance artificial intelligence system, thus enhancing the competitiveness of the domestic system and the field of artificial intelligence semiconductor.

magazine reference

Kang, c. et al. (2022) A cluster-type analog memristor by engineering redox dynamics for high-performance neural computing. Nature Communications. doi: 10.1038/s41467-022-31804-4.


Leave a Reply

Your email address will not be published. Required fields are marked *