AI CHIP

Green digital devices for the future

Metaplastic Magnetic Synapses for Neuromorphic Computing

Context

In the last decade, AI algorithms have reached human-level performance across increasingly complex tasks. But this progress comes with a trade-off: bigger neural networks, more computing power, and a huge spike in energy consumption when training deep learning models. As these networks grow and AI applications expand, the demand for more energy-efficient, AI-specific hardware is more urgent than ever.

Traditional computing, powered by CPUs and GPUs, still relies on the Von Neumann architecture, where computation and memory are separate. As a result, about 80% of energy goes into moving data around.

That is where neuromorphic computing comes in. By using non-volatile technologies like Magnetic Random-Access Memory (MRAM), we can drastically cut energy use while mimicking the brain’s structure and function. Neuromorphic chips bring AI directly to embedded devices, enabling a new wave of ultra-efficient edge computing.

Technology

At Spin-Ion Technologies, we have pioneered a cutting-edge solution using ion beam processes to precisely tailor the magnetic properties of spintronic devices at the atomic level. Our breakthrough neuromorphic chips customized by our ion beam process, built on low-power synapses made from non-volatile magnetic devices, are designed to mimic the brain’s function with unmatched efficiency.

These chips tackle critical challenges in neural applications, such as catastrophic forgetting, which we solve with continual learning , and they minimize device variability. The result is highly efficient, robust hardware that is perfectly suited for next-generation edge computing and neuromorphic systems, delivering the performance and reliability needed for intelligent, energy-efficient devices at the edge.

Impact

Our groundbreaking initiative combines advanced hardware and software to showcase the implementation of a metaplastic Artificial Neural Network on a magnetic chip. By seamlessly bridging computational neuroscience with deep learning, this innovation paves the way for the next generation of embedded neuromorphic systems, offering a transformative leap in efficiency and intelligence.