Unique memristor design with analog switch shows promise for high-performance neuromorphic computing

The growing use of models based on artificial intelligence (AI) places great demands on the electronics industry, as many of these models require large amounts of storage space and computing power. Therefore, engineers around the world are trying to create neuromorphic computer systems that can help meet these needs, many of which are based on memristors.

Memristors are electronic components that control the flow of electricity in circuits as they “remember” the amount of electrical charge that has passed through them. These features can replicate the function of biological synapses in the human brain, thereby improving the performance of machine learning models in analyzing data and making calculations.

Despite their potential, most of the memories developed so far have shown significant limitations, including small on/off ratios. These small components interfere with the memristors’ ability to represent accurate measurements, thus increasing noise and reducing the accuracy of the algorithm’s estimates.

Researchers at Wuhan University recently developed new memristors with analog switching and high on/off ratios. These memristors, presented in a paper published in Natural Electronicsare made of two-dimensional (2D) van der Waals metal materials as cathodes.

“Analog memristors with multiple performance levels are used primarily in high-performance neuromorphic computing, but their mass-mapping capabilities are often limited by small ratios of on/off,” Yesheng Li, Yao Xiaong and their colleagues wrote in their paper.

“We show that memristors with analog resistive switching and large on/off ratios can be created using two-dimensional van der Waals metal materials (graphene or platinum ditelluride) as cathodes. switching medium.”

The unique memristor design proposed by Li, Xiaong and their colleagues introduces an additional high resistance that prevents the migration of silver ions. Finally this enables analog switching, as well as on/off measurements comparable to those recorded in digital memristors.

“Previous methods have focused on changing the ion mobility using changes in the fixed layer or anode, which can lower the on/off ratios,” wrote Li, Xiaong and their colleagues.

“On the other hand, our method relies on the van der Waals cathode, which allows silver ion intercalation / de-intercalation, creating a high exchange barrier to change the ion movement. The strategy can achieve analog resistive conversion with an on/off ratio of up to 10⁸than 8-bit conductance states and attojoule-level energy consumption.”

To test their memristors, the researchers performed a chip-level simulation of a convolutional neural network (CNN) for image recognition. Their findings were very promising, as in this comparison, the model performed remarkably well, achieving image recognition accuracy of up to 91%.

In the future, the team’s newly developed memristor can be further developed and used to run other advanced types of AI computing. Alternatively, some researchers may aim to create similar memristors using other materials as switching devices or other van der Waals materials such as cathodes.

© 2024 Science X Network