« in Materio Computing » Classification Device

Simple SummaryContent extracted from patent full text and abstract with AI.

This invention presents a novel 'in materio computing' classification device that uses an engineered magnetic medium as a physical analog of a neural network for classification tasks. It encodes input signals into physical excitation (such as a magnetic field or current) to manipulate magnetic domain walls in a specially structured magnetic material. The final magnetic configuration of the medium, resulting from the movement of domain walls, is then read and interpreted to classify the input signal, enabling efficient, hardware-based pattern recognition or sorting without traditional training of the material system itself.

Use CasesContent extracted from patent full text and abstract with AI.

• Speech and voice recognition systems (e.g., digit recognition in spoken commands)
• Real-time pattern recognition in sensor networks or IoT devices
• Sorting and classification in automated quality control for manufacturing
• Implementation of neuromorphic computing hardware for AI and machine learning applications
• Security devices for biometric authentication (voiceprint, signature)
• Biomedical signal classification (e.g., ECG or EEG pattern identification)
• Embedded AI in edge devices with energy or space constraints
• Adaptive recognition systems in robotics or autonomous vehicles

BenefitsContent extracted from patent full text and abstract with AI.

• Rapid and energy-efficient classification using physical processes instead of large-scale digital computation
• Eliminates the need for time-consuming material-system training—only the output neural layer requires minor adjustment per task
• Compatible and integrable with existing CMOS technology, enhancing manufacturability
• Highly reconfigurable: same device can be quickly adapted for different classification tasks
• Scalable state space allows handling complex signal patterns
• Potential for miniaturization and integration into small, portable devices
• Flexibility in signal pre-processing and excitation methods (magnetic, electric, etc.)
• Robust to variations and defects due to the nature of the inhomogeneous magnetic medium
• Experimental evidence shows high accuracy in relevant tasks (e.g., >97% recognition rate in speech digit recognition)

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

This device (1), comprising a pre-processing stage (10), a material system (20) and an identification stage (30), is characterized in that: the pre-processing stage (10) transforms a signal of interest (S0) into an excitation signal (Sext); the material system (20) comprises a magnetic medium (50) subdivided into a plurality of zones (51) either in a first magnetic state or in a second magnetic state, an excitation module (26) generating, from the excitation signal (Sext), a physical excitation quantity adapted to displace the magnetic domain walls (56) in the magnetic medium (50), and a reading module (62) measuring the final magnetic configuration reached by the magnetic medium (50); and the identification stage (30) determines a class (Ci) to which the signal of interest (S0) belongs from the final magnetic configuration.