Table 1 Parameter values for the initialization of the SRNN

From: Self-organization of an inhomogeneous memristive hardware for sequence learning

Neurons

Synapses

SRNN

 

Excitatory

Inhibitory

    

Number of neurons

160

40

τ

1 ms

pEE

2%

R2

trained by IP

1 GΩ

Weight

(trained by SDSP)

pII

0%

R1

400 MΩ

600 MΩ

  

pEI

2%

C1

10 pF

10 pF

  

pIE

10%

τCa

100 ms

100 ms

    

Vth

0.2 V

0.2 V

    

R3

1 GΩ

1 GΩ

    

C2

2 pF

2 pF

    

IP

SDSP

  

FT

50 Hz

Vth

0.2 V

  

σ

15 Hz

Vθ

0.1 V

  

VRESET

2 V

Learning Rate

0.01–0.1

  
  1. Such values are defined with small-to-absent tuning, with the only aim to guarantee a minimal activation of the network, so to fully rely on the plasticity mechanisms (SDSP and IP) to improve performance. Some parameters, such as the magnitude of RRAM resistance in HRS and the Membrane Capacitance, are forced by technological constraints.
Back to article page