An Adaptive Inhibitory WSe 2 Transistor for Retinomorphic In‐Sensor Image Processing

ABSTRACT Retinomorphic vision systems have emerged as a promising approach for low‐power in‐sensor image processing by mimicking the human retina. Retinal contrast enhancement amplifies local visual features at the sensor level, eliminating the need for computationally intensive external processing. However, most conventional retinomorphic devices require illumination‐dependent bias adjustments, which introduce decision‐making overhead and limit scalable array integration. Here, we report an adaptive inhibitory WSe 2 transistor that performs decision‐free, intensity‐adaptive image processing within a single pixel. The current inhibition ratio decreases monotonically from ∼16.2 to ∼1.2 as the back‐gate voltage increases from 3 to 13 V, reflecting regime‐dependent carrier transport. When integrated with photodiodes, the transistor switches between subthreshold and accumulation regimes depending on local illumination. Under a common top‐gate pulse, dim pixels are strongly suppressed while bright pixels largely retain their current. We then processed gradient and noisy “G” patterns on the 5 × 5 photodiode array, achieving contrast enhancement and noise filtering. Furthermore, simulations of license‐plate recognition under rain and fog show that processing restores character morphology, improving classification accuracy from 47.7% to 97.3%. These results show that adaptive inhibition offers a scalable, energy‐efficient route to adaptive in‐sensor processing for edge‐vision hardware in low‐visibility environments.