A Hybrid Leaky Integrate-and-Fire Neuron with Tunable Reset Behavior in 7 nm Fin Field-Effect Transistor Technology

Abstract

This thesis presents the design and simulation study of a compact hybrid leaky integrate-and-fire (LIF) CMOS neuron derived from a Besrour-style baseline and augmented with a minimal reset/timing branch. The proposed neuron introduces an auxiliary branch comprising Ctref, Npchg, Nbleed, and Nreset to improve control over post-spike reset while preserving the original neuron's compact structure. The central goal of this work is to determine whether a small reset-aware modification can provide a more useful tradeoff among reset behavior, firing frequency, and energy than the baseline Besrour neuron. The circuit was implemented in Xschem and evaluated in ngspice using ASAP7 predictive FinFET device models. Reference neurons and the proposed hybrid neuron were analyzed within a common simulation workflow to enable a normalized comparison. The study focused on three primary evaluation metrics: reset behavior, firing frequency, and energy per spike, with additional consideration of static power and operating robustness. Transient simulations and parameter sweeps were used to examine how the added reset/timing branch modifies the membrane trajectory, internal spike formation, output response, and recovery behavior after firing. The results show that the proposed hybrid neuron exhibits stable integrate-and-fire operation and that the added branch introduces a tunable reset mechanism strongly influenced by the bleed bias. Two practically important operating regions were identified. Near Vbleed ≈ 0.22 V, the hybrid neuron achieved its best performance-oriented operating point, providing the most favorable nominal frequency/energy tradeoff among the tested hybrid points. Near Vbleed ≈ 0.24 V, the neuron exhibited its best robustness-oriented operating point, with stronger reset control and broader clean-operation coverage across the tested sweeps. Compared with the Besrour baseline, the hybrid neuron provides an additional degree of control over reset and timing, enabling a more explicit study of how reset quality influences firing frequency and energy, while incurring a higher static power cost. Overall, this work demonstrates that a compact Besrour-derived neuron can be extended with a minimal reset/timing branch to create a reset-aware hybrid LIF architecture with tunable operating regimes. The results support the view that reset is not merely a secondary waveform detail, but a central design variable that directly shapes frequency, energy, and robustness in compact neuromorphic neurons.