The increasing demand for efficient and scalable computing has highlighted the limitations of traditional von Neumann architectures, which suffer from significant energy consumption and latency due to the separation of memory and processing units. In-memory computing offers a promising alternative by performing computations directly within memory, reducing data movement and enhancing efficiency. Magnetic topological insulators (MTIs) have emerged as a compelling material platform for this approach, leveraging their unique electronic properties to enable low-power, high-speed operations.

This paper explores the potential of MTIs in cryogenic in-memory computing, where reduced temperatures enhance device performance by minimizing noise and dissipation. Through theoretical modelling, simulations, and experimental verification, we demonstrate how MTIs can implement efficient computational tasks, including artificial intelligence applications such as image classification. The results highlight the transformative potential of MTI-based in-memory computing for future low-power, high-performance computing technologies.

Cryogenic In-Memory Computing with Magnetic Topological Insulators

Magnetic topological insulators (MTIs) have emerged as a promising material class for next-generation computing technologies. Their unique electronic properties, combined with cryogenic operating conditions, offer new opportunities for in-memory computing architectures that are both energy-efficient and scalable. This article presents recent advancements in MTI-based systems, including experimental verification of anomalous Hall currents, equivalent circuit models, and neural network implementations.

Simulations reveal how electrical signals propagate within MTI structures at cryogenic temperatures. These studies highlight the material’s ability to process information efficiently with minimal energy dissipation. The quantum properties of MTIs enable robust signal transmission, making them ideal for high-speed computing applications.

Experimental work with devices such as D7 has confirmed the presence of anomalous Hall voltage and current in MTI systems. These measurements provide critical insights into electron behavior under magnetic fields, confirming the stability of magnetic states required for data retention. Such findings validate the theoretical predictions and pave the way for practical implementations.

Impact of Hall Bar Configurations

Extended Data Figures 8 and 10 demonstrate how serial and parallel configurations of Hall bars affect voltage and current scaling in MTI-based systems. These setups enhance computational capabilities by aggregating contributions from multiple devices, enabling scalability without significant performance degradation. This flexibility is essential for designing efficient and robust computing architectures.

The article also explores the implementation of neural networks on MTI hardware. Using datasets like CIFAR-10, researchers have evaluated the performance metrics of these systems, demonstrating their potential for energy-efficient machine learning tasks. A boost algorithm has been introduced to further improve accuracy and efficiency in neural network implementations, showcasing the versatility of MTIs in advanced computing applications.

While MTIs offer significant advantages, challenges remain particularly the requirement for cryogenic operating temperatures. This necessitates specialized equipment and infrastructure, though the benefits—such as enhanced computational efficiency—may justify these investments. Further research is needed to optimize MTI-based systems for niche applications like quantum computing.

Magnetic topological insulators hold immense potential for revolutionizing in-memory computing at cryogenic temperatures. By combining theoretical insights with experimental and practical implementations, this research underscores the feasibility of MTIs as a transformative technology for future computing architectures.