Vertical Bilayer Anti-Ambipolar Transistors for Tunable Electrochemical Neurons

The human brain deploys a complex network of interconnected spiking neurons featuring a variety of electrophysiological characteristics to perform complex classification and recognition tasks^1,2,3. Although machine learning algorithms implemented on traditional von Neumann electronics have emerged for diverse computing applications, they lag behind biological calculators in terms of performance and power consumption¹. This “technological gap” stems from the brain’s massively paralleled event-based processing of information, via interconnected neuronal networks, and the high specificity of its circuital elements, via neuronal diversity, which neuromorphic devices attempt to mimic^1,4. Advancements in organic neuromorphic devices over the last few years has led to several circuits that mimic neuronal spiking behaviors, but these systems are limited by complexity, large device footprints, high power consumption, and show limited customizability^1,4,5,6,7,8.<br/>Replicating the vast diversity of neurons with a scalable fabrication strategy is a critical step towards the development of large scale organic artificial neural networks that can approach the computational capabilities and functionality of the brain¹. Recently, neuromorphic systems have been developed based on the integrate and fire neuron model, which uses conventional organic electrochemical transistors(OECTs), and the hodgkin-huxley (HH) neuron model, which uses anti-ambipolar OECTs (aaOECTs), that can exhibit bio-mimetic spiking behaviors with low power requirements. However, these platforms are generally limited by single device footprint and, for the HH-based systems, by the customizability of the aaOECTs characteristics including peak position, width, and threshold voltage ^4,7,8.<br/>Here, a novel vertical aaOECT that can be used in spiking circuits to reduce size footprint and introduce a full control of neurons characteristics, including spiking threshold, refractory period, and frequency, is presented. The vertical aaOECT is fabricated through a bilayer in-series combination of accumulation and depletion modes p and n-type OMIEC materials. To demonstrate the applicability in spiking circuits, the aaOECT is used within the bio-realistic HH model leading to decreased neuron footprint. Neuronal customization is achieved through material selection, which reliably alters the aaOECTs characteristics, with a full control over the peak position, width, and threshold voltage.<br/>This device formulates a critical tool for the creation of large scale organic artificial neural networks that are not only significantly reduced in size, due to the vertical structure, but also customizable in neuron characteristics, thereby providing a fundamental step towards diverse bio-mimetic and interconnected networks of neurons.<br/>1. van de Burgt, Y. et al. Nat Electron, 1, 386–397. (2018)<br/>2. Subkhankulova, T. et al. Front Mol Neurosci. 3, 10. (2010)<br/>3. Ascoli, G. et al. Nat Rev Neurosci 9, 557–568 (2008)<br/>4. Harikesh, P. C. et al. Nat. Mater. 22, 242–248. (2023)<br/>5. Sebastian, A. et al. Nat. Commun. 10, 4199. (2019)<br/>6. Huang, W. et al. Nature 613, 496–502. (2023)<br/>7. Harikesh, P. C et al. Nat Commun. 13, 901. (2022)<br/>8. Sarkar, T. et al. Nat Electron. 5, 774–783. (2022)