Silicon-compatible optoelectronic artificial neuron

Abstract

Neuromorphic visual–perception systems, as mimics of the human visual system, have achieved transcendence in working scope beyond the visible light region. However, the slow response, low output signal contrast and high fabrication and operation costs of existing devices hinder their ability to reproduce human cognitive functions in a highly accurate and energy-efficient manner. Here, we developed a simple two-terminal, low-cost device that can receive and pre-process optical stimuli instantaneously by integrating an amorphous molybdenum silicon thin film onto silicon. By implementing this universal integration approach, a device with negative-to-positive photoconductivity conversion, a fast response (∼10⁻⁴ s), and less-stringent preparation and operation requirements has been built as a promising candidate for future neuromorphic visual–perception systems with improved efficiency and excellent image sharpening functionality. This outcome provides numerous opportunities to simplify the circuit configuration and facilitate the development and application of cutting-edge artificial intelligence.