Researchers at Oregon State University have developed a novel brain-inspired phototransistor capable of detecting light and storing visual data simultaneously. By mimicking the synaptic functions of the human brain, this device can retain information about light intensity and perform “gradual forgetting,” a process that could significantly improve the efficiency of robotic vision systems by reducing the need for constant data transfers between sensors and memory components.
Traditional digital imaging, which relies on complementary metal-oxide-semiconductors (CMOS) or charge-coupled devices (CCDs), requires a continuous flow of data from the sensor to a separate memory unit for processing. According to Larry Cheng, a professor of electrical engineering and computer science at Oregon State University, these standard sensors are inherently limited because they “immediately forget” images unless the data is offloaded. The new phototransistor, described in the journal Advanced Functional Materials, addresses this bottleneck by integrating memory directly into the pixel array.
How the Brain-Inspired Sensor Functions
The experimental prototype is a four-by-four-pixel array roughly the size of a USB stick. The device utilizes a transparent, light-absorbing organic layer that converts incoming photons into electrical charges. When light hits this layer, it creates both electrons and “holes.” While the electrons move into an indium gallium zinc oxide (IGZO) transistor channel, the holes remain trapped within the organic semiconductor aggregates.
These trapped holes act as a memory mechanism, electrostatically modulating the transistor channel long after the light source is removed. Cheng notes that this process mirrors the role of dopamine in the human brain, which strengthens connections between synapses to facilitate memory retention. Because the device can be tuned using voltage, researchers can control the duration of this memory. Applying a positive voltage pushes the trapped holes away, accelerating the “forgetting” process, while a negative voltage pulls them closer to the channel, extending the duration of the stored data.
The use of IGZO is critical to the device’s design. Because the material is transparent to visible light, it does not interfere with the light-sensing function of the organic layer. This decoupling of electrical transport from light detection allows for more efficient, large-area fabrication, a technique already common in modern display technologies.
Implications for Robotic Vision and Energy Efficiency
The ability to perform on-sensor processing represents a significant shift in how artificial intelligence algorithms might handle video data. Current AI-driven image recognition requires processors to analyze video feeds frame by frame, which is energy-intensive due to the constant shuffling of data between hardware components. By storing the recent history of light intensity directly on the sensor, this new technology allows the device to flag only the changes or patterns that are relevant.
The researchers emphasize the adaptability of this “tunable memory.” A high-speed application, such as a drone flying at 250 kilometers per hour, requires the sensor to quickly discard old data to track fast-moving changes. Conversely, a security camera monitoring for loitering individuals might require a longer memory sequence. By adjusting the voltage, the same sensor can be calibrated for different tasks, potentially leading to massive reductions in energy demand for autonomous systems.
While the current work is limited to a small-scale prototype, the team is now focused on scaling the technology to larger pixel arrays. The next objective is to develop an integrated imaging prototype that can demonstrate real-time temporal imaging and on-sensor processing in practical environments. As the research moves toward these larger arrays, the goal remains to showcase how this brain-inspired approach can handle complex visual tasks with greater speed and efficiency than traditional electronic sensors.
Please feel free to share your thoughts or questions in the comments section below.