Abstract
We propose self-rectifying resistive random access memory (RRAM) synapse to prevent reverse leakage current problem which occurs when RRAM is integrated with integrate and fire (IF) circuit in spiking neural network (SNN). Ni/W/SiNx/n-Si RRAM was fabricated by varying the bottom electrode (BE) doping concentration and their rectifying characteristics were analyzed. Low BE doping concentration device showed self-rectifying characteristics without any additional selector or diode device. Furthermore, hardware-based system-level simulation was conducted to evaluate the effect of self-rectifying RRAM synapse on MNIST classification accuracy. About 93.34% accuracy was obtained using the proposed RRAM.
Original language | English |
---|---|
Article number | 9366932 |
Pages (from-to) | 1610-1615 |
Number of pages | 6 |
Journal | IEEE Transactions on Electron Devices |
Volume | 68 |
Issue number | 4 |
DOIs | |
State | Published - Apr 2021 |
Keywords
- Neuromorphic
- resistive random access memory (RRAM)
- self-rectifying
- synaptic device
- system-level simulation