A Self-Formed Ag Nanostructure Based Neuromorphic Device Performing Arithmetic Computation and Area Integration: Influence of Presynaptic Pulsing Scheme on Mathematical Precision
Modern computers although capable of performing complex tasks, come with inherent limitations arising from their architecture. Following neuromorphic approaches, researchers are developing brain-inspired devices using diverse materials and geometries to replicate neural processes. This research is crucial for developing low-energy technologies and future computing systems that go beyond the traditional CMOS and Boolean logic.
The presentation is related to the fabrication and performance of a neuromorphic device consisting of self-formed labyrinthine silver nanostructure capable of exhibiting potentiated states upon electrical activation, emulating a bio-synapse with ultralow energy requirement (~ 1fJ/synapse). The conductance and its retention in the potentiated state both vary linearly with positive and negative polarity pulses exhibiting very low nonlinearity factors. We made several such devices with comparable performances which are indefinitely stable. Leveraging on high linearity, we performed arithmetic operations of summation and subtraction by varying the sequence of feeding positive and negative pulses. These operations yielded highly accurate results; lesser the mixing of polarities higher was the accuracy. Notably, increasing the current compliance increased accuracy, which we attribute to the formation of stronger filaments within the percolation network, thereby enhancing conductance and retention. The treatment was extended to estimation of ‘the area under a curve’ matching the expected results. Moreover, the device demonstrated a simulation-based image classification accuracy of 94.95%. These features underscore the potential of our scalable device architecture for precise computational applications away from traditional von Neumann architecture.
Mousona Pal, Manpreet Kaur, Bhupesh Yadav, Arti Bisht, Vidhyadhiraja NS, Giridhar U Kulkarni
