- Neuromorphic architecture
Von Neumann Architecture is the structure of current modern computer.
And this modern computer has two main parts for processing data; CPU and storage.
The sequential execution process of a typical computer makes a bottle-neck problem when it comes to dealing with unstructured and complex data.
To overcome this, neuromorphic architecture that works in a similar manner to the human brain has been studied.
Neural network is one of the strongest way to implement the neuromorphic architecture, and it can be easily fabricated in the form of a cross-point array because each interconnection between the pre-neuron and the post-neuron would be a synapse.
- RRAM for synaptic memory device
Conventional synapse device are used as Static Random Access Memory (SRAM).
The SRAM consists of six transistors, which is disadvantageous for high integration.
Due to this, Resistive Random Access Memory (RRAM) has been investigated as a synapse device due to its simple 2-terminal structure, low power operation, scalability, and simple stack.
The resistive switching mechanism of RRAM is oxygen migration between electrode and oxide layer characteristic so that the analog characteristics can be shown in the device.
Human brain can easily recognize various objects and process a large amount of information.
To emulate biological characteristics, neuromorphic system with massive parallelism and fault tolerance has been developed.
Using cross-point RRAM synapse array, various application such as visual image processing and speech recognition can be implemented with high efficiency.
- Burr, Geoffrey W., et al. "Experimental demonstration and tolerancing of a large-scale neural network (165 000 synapses) using phase-change memory as the synaptic weight element." IEEE Transactions on Electron Devices 62.11 (2015):3498-3507.
- S. Park et al., "Neuromorphic speech systems using advanced ReRAM-based synapse." Electron Devices Meeting (IEDM), 2013 IEEE International. IEEE, 2013.
- K. Moon et al., "High density neuromorphic system with Mo/Pr0.7Ca0.3MnO3 synapse and NbO2 IMT oscillator neuron." Electron Devices Meeting (IEDM), 2015 IEEE International. IEEE, 2015.
- Park, Sangsu, et al. "Electronic system with memristive synapses for pattern recognition." Scientific reports 5 (2015): 10123.