Energy-efficient neuromorphic system using novel tunnel FET based LIF neuron design for adaptable threshold logic and image analysis applications
Abstract In this study, a novel tunable dopingless band-to-band tunneling mechanism based Leaky Integrate and Fire (LIF) neuron is proposed with a notable improvement in integration density and energy consumption. The forward transfer characteristics of Tunnel FET with sharp sub-threshold swing have...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-04-01
|
| Series: | Scientific Reports |
| Online Access: | https://doi.org/10.1038/s41598-025-93727-6 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract In this study, a novel tunable dopingless band-to-band tunneling mechanism based Leaky Integrate and Fire (LIF) neuron is proposed with a notable improvement in integration density and energy consumption. The forward transfer characteristics of Tunnel FET with sharp sub-threshold swing have been utilised to simulate the neural activity. The simulations performed using Atlas 2D software confirm that the proposed TFET can effectively replicate the spiking behavior of a biological neuron, eliminating the need for additional circuitry, in addition to offering tunable features. The proposed LIF neuron demonstrates significantly lower energy consumption, operating at just 144 aJ per spike. This energy efficiency is at least $$10^6$$ times lower than the single MOSFET-based neuron and $$10^3$$ times lower than TFET-based 1-transistor neurons reported in prior literature. This remarkable improvement is attributed to the underlying mechanism, which leverages tunneling and material engineering techniques. The proposed neuron has also been successfully investigated for the implementation of adaptable threshold logic functions (NOT, OR and AND). This offers a solution for the design of highly scalable and energy efficient threshold logic circuits for future neuromorphic computing systems. Lastly, we implement a multilayer SNN that confirms the image recognition ability of the proposed neuron with 92.1 $$\%$$ accuracy. |
|---|---|
| ISSN: | 2045-2322 |