Ultra-Low Energy LIF Neuron Using Si NIPIN Diode for Spiking Neural Networks

Bhaskar Das, Jörg Schulze, Udayan Ganguly

IEEE Electron Device Letters · 2018 · 53 citations · 18 references

Concepts

Abstract

An energy efficient neuron is essential for spiking neural network (SNN) to operate at low energy to mimic the human brain functionalities in hardware. Several CMOS-based Si transistors, memory devices, spintronic devices have been used as a neuron for SNN. However, the main concern is the energy efficiency for these neurons. In this letter, we experimentally demonstrate a Si-based CMOS compatible asymmetric NIPIN diode as a LIF neuron. First, we demonstrate the LIF neuron characteristics by comparing the spike-frequency (f) versus voltage curve with that of a simple LIF neuron model. This neuron shows a classical ReLU behavior, which is attractive for typical software neuron models. Then, we show an ultra-low energy consumption of ~2 × 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-17</sup> J per spike at 10-nm node of this neuron, as NIPIN diode is highly scalable (4F <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ) due to its capacitorless structure. This is the lowest reported energy/spike for any LIF neuron for SNN application. Thus, the NIPIN is suitable for ultra-low energy LIF neuron application for energy efficient SNN.

References

18