Abstract
Hardware-based spiking neural networks (SNNs) inspired by a biological nervous system are regarded as an innovative computing system with very low power consumption and massively parallel operation. To train SNNs with supervision, we propose an efficient on-chip training scheme approximating backpropagation algorithm suitable for hardware implementation. We show that the accuracy of the proposed scheme for SNNs is close to that of conventional artificial neural networks (ANNs) by using the stochastic characteristics of neurons. In a hardware configuration, gated Schottky diodes (GSDs) are used as synaptic devices, which have a saturated current with respect to the input voltage. We design the SNN system by using the proposed on-chip training scheme with the GSDs, which can update their conductance in parallel to speed up the overall system. The performance of the on-chip training SNN system is validated through MNIST data set classification based on network size and total time step. The SNN systems achieve accuracy of 97.83% with 1 hidden layer and 98.44% with 4 hidden layers in fully connected neural networks. We then evaluate the effect of non-linearity and asymmetry of conductance response for long-term potentiation (LTP) and long-term depression (LTD) on the performance of the on-chip training SNN system. In addition, the impact of device variations on the performance of the on-chip training SNN system is evaluated.
Highlights
Artificial neural networks (ANNs) have shown superior performance in several fields, such as pattern recognition or object detection (Gokmen and Vlasov, 2016; Ambrogio et al, 2018; Kim C.-H. et al, 2018; Kim J. et al, 2018; Kim et al, 2019)
The on-chip training spiking neural networks (SNNs) systems train a weight by applying an update pulse to a synaptic device representing a weight, which leads to low power consumption for training a weight (Hasan et al, 2017)
We assume that synaptic devices have a linear conductance response and no variation, and the baseline accuracy in Figure 6A is evaluated in artificial neural networks (ANNs) that have the same network size
Summary
Artificial neural networks (ANNs) have shown superior performance in several fields, such as pattern recognition or object detection (Gokmen and Vlasov, 2016; Ambrogio et al, 2018; Kim C.-H. et al, 2018; Kim J. et al, 2018; Kim et al, 2019). When the membrane potential exceeds the threshold voltage, the neuron generates a spike to the deeper layer This biological behavior of the neuron in SNNs can be matched to the behavior of the rectified linear unit (ReLU) activation function in ANNs (Diehl et al, 2015; Rueckauer et al, 2017). SNNs adopting the ANN-toSNN conversion cannot update themselves depending on various system situations and only perform the inference process for a given task. For this reason, the performance of SNNs that adopt conversion is sensitive to unexpected variations of hardware and cannot save the power consumption required for training a weight (Kim H. et al, 2018; Yu, 2018). The on-chip training SNN systems train a weight by applying an update pulse to a synaptic device representing a weight, which leads to low power consumption for training a weight (Hasan et al, 2017)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
