Researchers at Oregon State University have developed a novel imaging sensor capable of programmable memory retention and on-demand information decay, mirroring the synaptic forgetting mechanisms of the human brain. Led by Assistant Professor Li-Jing Larry Cheng, the team published their findings in Advanced Functional Materials, introducing a hybrid phototransistor architecture designed to drastically reduce energy consumption in edge artificial intelligence applications. Traditional digital imaging systems rely on complementary metal-oxide-semiconductor or charge-coupled device sensors that capture photons but lack inherent storage. This separation of sensing and memory creates a data transfer bottleneck, forcing frequent communication between the sensor, memory, and processor. The new device circumvents this von Neumann bottleneck by integrating optical absorption and charge transport within a single pixel. A transparent organic semiconductor layer captures incident light, generating electron-hole pairs. While electrons move into an indium gallium zinc oxide transistor channel for signal transmission, holes are trapped within the organic layer, serving as the physical basis for short-term memory. This functional separation eliminates parasitic light absorption by the transistor and leverages the transistor material high carrier mobility and low leakage current. The breakthrough lies in the sensor ability to modulate memory duration via electrical bias. Applying a positive voltage pushes trapped holes away from the transistor channel, accelerating charge decay and enabling rapid information forgetting. Conversely, a negative voltage retains holes near the channel, extending memory retention to several hours. This tunable behavior allows a single hardware platform to adapt to diverse visual tasks. High-speed drone navigation requires brief temporal awareness, while security cameras monitoring slow-moving subjects benefit from prolonged historical context. By processing recent illumination changes directly at the source, the sensor reduces the computational load and power demands typically associated with frame-by-frame AI analysis. The research team fabricated a four-by-four pixel prototype demonstrating exceptional low-light performance, operating effectively below five microwatts per square centimeter and detecting light intensities as low as 0.5 microwatts per square centimeter. When integrated into artificial neural network simulations using the MNIST dataset, the device maintained high classification accuracy above ninety percent, exhibiting strong fault tolerance against hardware parameter variations. The linear synaptic weight update characteristics confirm its viability for neuromorphic computing applications. While currently limited to a small-scale prototype, the technology addresses a critical efficiency gap in edge AI and autonomous systems. By shifting foundational image processing from centralized processors to intelligent sensors, the design promises substantial reductions in data bandwidth and operational power. Future development will focus on scaling the pixel arrays, refining integration pathways, and validating real-time imaging capabilities in operational environments. This architecture marks a significant step toward hardware that learns, remembers, and forgets on command, aligning physical computing closer to biological neural efficiency.