Researchers build brain-like memory device for AI sensors that may improve energy efficiency

Researchers at Oregon State University have developed a light-sensitive digital memory device that integrates sensing, memory, and signal processing within a single phototransistor, potentially reduci... Read original →

Oregon State’s optoelectronic neuromorphic transistor disrupts the AI hardware stack by collapsing sensing, memory, and processing into a monolithic phototransistor—rendering conventional CMOS image sensors with off-chip memory increasingly obsolete. This forces EUV-centric scaling to pivot toward heterogeneous integration, benefiting oxide/organic material suppliers. Regulatory risks loom: while the device may skirt U.S. logic-chip export controls, its organic layer could face RoHS-style scrutiny over long-term stability, inflating early yield costs. Competitively, Sony and Samsung will likely accelerate stacked CIS-DRAM roadmaps, while NVIDIA may leverage this trend to embed tighter sensor-AI co-design in edge SoCs. Within 18 months, if IMEC adopts this approach into its neuromorphic roadmap, 'in-sensor computing' could become a de facto requirement for high-end AI vision systems, squeezing out smaller players lacking material or integration partnerships.