A neuromorphic core based on threshold switching memristor with asynchronous address event representation circuits

The full memristive network hardware features high density and excellent scalability. However, recent researches on the full memristive network have been limited to a single-layer network, due to the lack of effective and flexible communication between neurons. In this design, we demonstrate a neuromorphic core based on Ag/SiO2/Au threshold switching memristor, which has built-in asynchronous address event representation (AER) circuits to provide flexible communication between neurons. Since temporally sparse spikes are the medium of communication between neurons, the AER circuits are designed to transmit spikes serially which have been encoded with neurons' addresses before transmission. With the asynchronous circuits design, the AER circuits will detect neurons' output in real-time. To test the performance of the neuromorphic core, we designed a behavioral simulator for the neuromporphic core to simulate the liquid state machine (LSM) network, which achieves a 100% recognition rate in the free spoken digital dataset. The simulation results show that the neuromorphic core obtains 35 times higher performance than the CPU and 111 times higher energy efficiency than the GPU.

Automated summary

In this study, a neuromorphic core based on threshold switching memristors is designed to enable flexible communication between neurons. The experimental results show significant improvement in performance and energy efficiency of the neuromorphic core.