Traditional computer systems have separate units for processing and storing data, which makes them less efficient when dealing with complex data like AI.

Researchers at KAIST have developed a new computing chip that mimics the brain.

This chip integrates data processing and storage into one device using a memristor, which can change its resistance based on the electrical charge it has experienced. This allows the chip to learn and correct errors on its own, much like human learning.

This could enable important applications. For example, devices like smart security cameras to detect suspicious activity instantly without needing to send data to cloud servers. Medical devices could also analyze health data in real time.

The chip features a 32×32 memristor array. It processes video streams, learning to distinguish moving objects from the background, and it improves over time.

Accurate results in real-time image processing

The chip’s ability to learn independently has shown accurate results in real-time image processing. This self-learning capability makes the system practical and reliable. The innovation lies in the memristor’s ability to mimic synapses in neural networks, allowing simultaneous data storage and computation. This mimics how our brain cells work, making the chip very efficient.

The research team has overcome previous limitations in neuromorphic devices by creating a system that adapts to environmental changes quickly. They achieved this by designing memristors that can control resistance changes precisely, eliminating the need for complex adjustments. This study, published in Nature Electronics, not only proves the technology’s potential but also its readiness for commercialization.

This technology promises to change how AI functions in on-board applications. It allows AI operations to happen locally on the device, enhancing speed, privacy, and energy efficiency. The researchers describe this system as a smart workspace where everything is immediately accessible, much like the brain’s efficient processing. The project received funding from various national research initiatives in Korea, emphasizing its significance and potential impact.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.