A Neuromorphic Device Implemented on a Salmon‐DNA Electrolyte and its Application to Artificial Neural Networks

Citations Over TimeTop 10% of 2019 papers

Abstract

A bioinspired neuromorphic device operating as synapse and neuron simultaneously, which is fabricated on an electrolyte based on Cu2+-doped salmon deoxyribonucleic acid (S-DNA) is reported. Owing to the slow Cu2+ diffusion through the base pairing sites in the S-DNA electrolyte, the synaptic operation of the S-DNA device features special long-term plasticity with negative and positive nonlinearity values for potentiation and depression (α_p and α_d), respectively, which consequently improves the learning/recognition efficiency of S-DNA-based neural networks. Furthermore, the representative neuronal operation, "integrate-and-fire," is successfully emulated in this device by adjusting the duration time of the input voltage stimulus. In particular, by applying a Cu2+ doping technique to the S-DNA neuromorphic device, the characteristics for synaptic weight updating are enhanced (|α_p|: 31→20, |α_d|: 11→18, weight update margin: 33→287 nS) and also the threshold conditions for neuronal firing (amplitude and number of stimulus pulses) are modulated. The improved synaptic characteristics consequently increase the Modified National Institute of Standards and Technology (MNIST) pattern recognition rate from 38% to 44% (single-layer perceptron model) and from 89.42% to 91.61% (multilayer perceptron model). This neuromorphic device technology based on S-DNA is expected to contribute to the successful implementation of a future neuromorphic system that simultaneously satisfies high integration density and remarkable recognition accuracy.

Related Papers

• → Is Neuromorphic MNIST Neuromorphic? Analyzing the Discriminative Power of Neuromorphic Datasets in the Time Domain(2021)26 cited
• Training spiking multi-layer networks with surrogate gradients on an analog neuromorphic substrate(2020)
• → ST-MNIST -- The Spiking Tactile MNIST Neuromorphic Dataset(2020)3 cited
• → EqSpike: Spike-driven Equilibrium Propagation for Neuromorphic Implementations(2020)