A neuroscience-inspired photosensitive component, developed through a joint effort between IIT Guwahati and IIT Kharagpur, draws on roughly one five-hundredth of the power that conventional visual processing demands, opening a path toward considerably lower energy consumption in image-handling systems.
Built to take in visual data, hold it in memory, and work through what it has captured, the device operates along principles that parallel how the brain handles sight and recall. According to a media release from IIT Guwahati, the technology holds potential for deployment across facial recognition systems, robotic and artificial vision platforms, and a range of other smart computing contexts.
Researchers worldwide are engineering neuromorphic architectures — computing systems modelled on the structure and function of the human brain. At the heart of these systems sits an artificial synapse — a structure designed to replicate how neurons link and communicate with one another.
In standard electronic imaging architectures, the functions of capturing, storing, and processing visual data are handled by discrete hardware components, a division that drives up both the volume of data moved between parts and the overall power drawn by the system. Trillions of synaptic junctions linking individual nerve cells give the human brain its capacity to carry out those same operations.
Heading up the physics department's research effort at IIT Guwahati was Prof PK Giri, whose team drove the study forward.
Researchers drew a comparison between the technology and how the brain holds onto visual impressions once an object has left the field of view, noting that electronic vision systems work by having a camera gather reflected light from a subject — an apple, for instance — convert that light into digital data, and pass it along to a microprocessor, which then breaks the image down at the pixel level to determine what the object is. Visual input of an apple triggers rapid pattern-matching in the brain, drawing on stored data about colour, shape, and contextual associations to identify the object.
Giri, the lead author of the study, noted that the process reduces the need to consciously work through an image step by step in order to identify objects within it.
When a short flash of light cuts out, the electrical activity it triggered carries on regardless. When the prototype device delivered a second pulse of light, the measured response came back 180% stronger than what the initial pulse had produced. The process bears a resemblance to how neural pathways strengthen through repeated activation.
Giri noted that the device brings together sensing, memory, and information-processing capabilities within a unified platform. A paper detailing the results has appeared in the peer-reviewed journal Carbon.
Listed as contributors to the paper alongside Giri were Sanjoy Sur Roy, Sanju Nandi, Debabrata Sahu, and Garima Choudhary, together with Prof M Meyyappan — all affiliated with IIT Guwahati — while Saswata Goswami and Prof Chandra Sekhar Tiwary brought their involvement through IIT Kharagpur.
