Cmos Analog Spiking Neurons Demonstrate Excitability Criterion Independent of Input Stimulus

The ability of spiking neurons to rapidly respond to incoming signals is fundamental to how the brain processes information, and researchers continually seek to replicate this efficiency in artificial systems. Léopold Van Brandt, Grégoire Brandsteert, and Denis Flandre, all from the ICTEAM Institute at UCLouvain, investigate the excitability of these artificial neurons, specifically those designed for ultra-low-power consumption. Their work establishes a clear criterion for determining when a neuron will ‘fire’, identifying membrane potential as the key intrinsic factor governing this response, independent of the strength of the input signal. This discovery represents a significant step towards building more biologically realistic and energy-efficient artificial neural networks, paving the way for advanced event-based computing systems.

Quantitative ultra-low-power analog neurons represent a significant area of research within circuit design. Researchers establish an excitation criterion, quantified either in terms of critical supplied charge or membrane potential threshold, utilising conventional SPICE simulations compatible with industrial transistor compact models. The research demonstrates that the membrane potential threshold is intrinsic to the neuron, meaning it operates independently of the input stimulus. Rigorous analysis of the nonlinear neuron dynamics provides valuable insight, and further exploration is needed, particularly regarding the effect of intrinsic noise. Within the neuromorphic paradigm, spiking neurons function as the fundamental building blocks of spiking neural networks and are massively replicated as neuronal cells.

Charge and Threshold Govern Neuron Excitability

Scientists have meticulously characterised the excitability of a novel ultra-low-power analog neuron, establishing a precise criterion for triggering action potentials, or spikes. Results demonstrate that the neuron’s spiking behaviour is determined by either a critical supplied charge or, crucially, by a membrane potential threshold, with the latter proving to be an intrinsic property independent of the input stimulus characteristics. The team assessed excitability by sweeping the amplitude and duration of injected synaptic current pulses, identifying a clear boundary between pulses that successfully triggered spikes and those that did not. This analysis revealed that for a given current amplitude, a minimum pulse duration is required to generate a spike, and conversely, a specific pulse duration imposes a minimum current amplitude.

Observations suggest a threshold criterion that considers both the amplitude and duration of the injected pulse, mirroring similar behaviour observed in state transitions within SRAM bitcells. Measurements confirm that the effective total membrane capacitance plays a key role in determining this threshold, with values extracted from the simulations providing a detailed characterisation of the neuron’s behaviour. The research delivers a fundamental understanding of spiking neuron dynamics and establishes a pathway for designing energy-efficient neural circuits.

Intrinsic Threshold Governs Neuronal Excitability

This research establishes a clear understanding of neuronal excitability using detailed circuit simulations. The team demonstrated that this intrinsic membrane potential threshold is independent of the specific input stimulus, offering a fundamental insight into how these biological systems process information. Through SPICE simulations, they characterised the neuron’s dynamics and determined the excitation threshold required to initiate an action potential, or spike. The work moves beyond theoretical models by utilising industry-standard transistor compact models, ensuring the findings are relevant to practical circuit design.

Researchers assessed excitability by sweeping the amplitudes of input current pulses and precisely defining the conditions necessary for spiking. While the analysis focuses on noiseless simulations, the authors acknowledge the importance of investigating the effects of intrinsic noise on neuronal behaviour, suggesting a direction for future work. This research provides a solid foundation for developing more efficient and biologically realistic neuromorphic computing systems.