A Ferroelectric Tunnel Junction-based Integrate-and-Fire Neuron

TL;DR

This work addresses the need for energy-efficient neuromorphic components by introducing a hybrid FTJ-CMOS integrate-and-fire neuron that leverages the gradual switching of a Hf0.5Zr0.5O2-based bilayer FTJ to accumulate input spikes into a firing event. A compact Preisach-based FTJ model, including dynamic polarization and leakage currents, is developed and calibrated to match device behavior, enabling realistic circuit simulations. The neuron circuit employs a 2T1C readout with tunable inverters to convert FTJ leakage into spikes, enabling electrically tunable neural dynamics controlled by set pulse amplitude and width, and inverter thresholds. Demonstrations show that firing can be modulated by pulse parameters, and the architecture promises a scalable, low-power building block for edge neuromorphic networks, though external bias/pulse generation may be shared across multiple neurons.

Abstract

Event-based neuromorphic systems provide a low-power solution by using artificial neurons and synapses to process data asynchronously in the form of spikes. Ferroelectric Tunnel Junctions (FTJs) are ultra low-power memory devices and are well-suited to be integrated in these systems. Here, we present a hybrid FTJ-CMOS Integrate-and-Fire neuron which constitutes a fundamental building block for new-generation neuromorphic networks for edge computing. We demonstrate electrically tunable neural dynamics achievable by tuning the switching of the FTJ device.

Figures (7)

• Figure 1: \ref{['fig:schematic']} Schematic of the FTJ stack. \ref{['fig:onoff']} DC-IV measurements of the FTJ in ON and OFF state, measured up to 1.5 V. Schematics of the corresponding band structures are shown in the inset.
• Figure 2: Measured FTJ reset currents after an exponentially increasing amount of pulses with amplitude 4 V and length 10 µ s. The shaded area $A_P$ is integrated to obtain the polarization in Fig. \ref{['fig:switched_polarization']}. The dashed line represents the complete switching process with a single -5 V, 500 µ s triangular voltage pulse.
• Figure 3: Measured switched polarization as a function of the number of applied pulses, for (a) exponentially increasing pulse widths at 3.5 V and (b) different voltages in steps of 0.5 V at a pulse width of 10 µ s. The polarization is normalized to a full switching measurement.
• Figure 4: (a) Switching I-V characteristic obtained with 100 µ s wide trapezoidal excitation (shown inset) and the equivalent simulated I-V obtained using the compact model. (b) Equivalent circuit.
• Figure 5: Integrate-and-Fire neuron schematic. Transistors $T_1$ to $T_4$ implement a variation of the 2T1C cell slesazeck2019uniting. Transistors $T_4$ to $T_7$ implement two inverters with electrically tunable switching threshold.