TY - GEN
T1 - A circuit model for a Si-based biomimetic synaptic time-keeping device
AU - Ostwal, Vaibhav
AU - Rajendran, Bipin
AU - Ganguly, Udayan
N1 - Publisher Copyright:
© 2015 IEEE.
PY - 2015/10/5
Y1 - 2015/10/5
N2 - Spike time dependent plasticity (STDP) is synaptic conductivity modulation which is dependent upon the time difference between pre- and post-synaptic neuronal spikes. Superposition of extended waveforms from pre- and post-synaptic neurons has been proposed to apply a pre- vs. post-neuron spike time difference dependent voltage across an RRAM based synapse to implement STDP. Such long waveforms are power-intensive. We have recently proposed an equivalent synaptic time-keeping based on impact ionization based NPN diode (I-NPN).Here the time difference between sharp pre- and post-neuronal spikes enables conductance change due to the internal charge carrier dynamics of I-NPN device. Such an approach is low power and biomimetic. In this paper, we propose a SPICE model of the I-NPN device. The model is able to emulate DC, transient and pulse train response of I-NPN diode. STDP is demonstrated. The SPICE simulations are much faster than TCAD, thus enabling circuit level simulation of neural network. Such a circuit model is essential for the evaluation of the impact of device characteristics on the learning performance of neuromorphic circuits which will provide essential device specification and enable learning efficiency benchmarking.
AB - Spike time dependent plasticity (STDP) is synaptic conductivity modulation which is dependent upon the time difference between pre- and post-synaptic neuronal spikes. Superposition of extended waveforms from pre- and post-synaptic neurons has been proposed to apply a pre- vs. post-neuron spike time difference dependent voltage across an RRAM based synapse to implement STDP. Such long waveforms are power-intensive. We have recently proposed an equivalent synaptic time-keeping based on impact ionization based NPN diode (I-NPN).Here the time difference between sharp pre- and post-neuronal spikes enables conductance change due to the internal charge carrier dynamics of I-NPN device. Such an approach is low power and biomimetic. In this paper, we propose a SPICE model of the I-NPN device. The model is able to emulate DC, transient and pulse train response of I-NPN diode. STDP is demonstrated. The SPICE simulations are much faster than TCAD, thus enabling circuit level simulation of neural network. Such a circuit model is essential for the evaluation of the impact of device characteristics on the learning performance of neuromorphic circuits which will provide essential device specification and enable learning efficiency benchmarking.
UR - http://www.scopus.com/inward/record.url?scp=84959371953&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84959371953&partnerID=8YFLogxK
U2 - 10.1109/SISPAD.2015.7292324
DO - 10.1109/SISPAD.2015.7292324
M3 - Conference contribution
AN - SCOPUS:84959371953
T3 - International Conference on Simulation of Semiconductor Processes and Devices, SISPAD
SP - 321
EP - 324
BT - 2015 International Conference on Simulation of Semiconductor Processes and Devices, SISPAD 2015
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 20th International Conference on Simulation of Semiconductor Processes and Devices, SISPAD 2015
Y2 - 9 September 2015 through 11 September 2015
ER -