Probabilistic Boolean Operations in Spiking Neurons for Neuromorphic Systems

Spiking Neural Networks (SNN) along with neuromorphic computing has the advantage of low power dissipation compared to traditional Von Neumann architectures. Most of the studies related to SNNs have been influenced by the major advancement of artificial neural network (ANN) and deep learning (DL) over the last two decades. However, the deterministic floating point operations that facilitate the success of deep learning based solutions is vastly different from the information processing mechanism of a biological brain. Intelligence and cognition of biological brain are acquired through probabilistic operations. In this work a novel probabilistic computational approach suitable for SNNs is presented. The Boolean logic realization had been the first major breakthrough in the artificial neuron paradigm. In this work a single spiking neuron is modeled as a probabilistic Boolean operator. A stochastic time to first spike (TTFS) encoding scheme is adopted. A common framework for realizing several Boolean operators is presented. A physical variable <tex-math notation="LaTeX">q</tex-math> , relating to the amplitude of the post synaptic potential is shown to be a key variable to control the probability of a particular Boolean operation. A novel relationship between <tex-math notation="LaTeX">q</tex-math> and the probability of firing a logic HIGH is established. The present framework of implementing Boolean operations with stochastic TTFS encoding is shown to improve power efficiency compared to traditional rate coding based approaches. The framework should be useful in readdressing the problems in SNN by incorporating probabilistic algorithms efficiently in a neuromorphic platform.