Lightweight Machine Learning Estimation Solution for Embedded Microcontrollers
Lightweight AI/Machine Learning Implementation Service
No need for an NPU! Machine learning estimation with existing microcontrollers. Achieving low latency, low power consumption, and offline edge decision-making with a lightweight implementation starting from 32KB.
This is a development support solution that implements machine learning algorithms built on a PC into a bare-metal/RTOS environment, taking into account constraints such as CPU, memory, and real-time performance, enabling estimation on microcontrollers. It allows for the addition of lightweight machine learning recognition and judgment functions to existing embedded systems. **Strengths of this solution:** - **No need for dedicated AI hardware:** By using lightweight classifiers such as SVM (SVC), it can be implemented on general existing microcontrollers (MCUs) that do not have NPU or GPU, thereby reducing additional hardware costs. - **High real-time performance and reliability through edge processing:** It enables on-site judgment (in a few milliseconds to tens of milliseconds) without communication, achieving "low latency" without network delays, "offline operation" independent of communication environments, "privacy protection" by not sending sensor information externally, and "low power consumption" by not constantly running high-spec CPUs. - **Diverse recognition and judgment tailored to the field:** It selects and learns optimal features from data such as images, vibrations, sounds, currents, and temperatures, realizing edge judgment functions for various industrial applications.
Inquire About This Product
basic information
【Main Specifications and Features】 ■ Required Memory Capacity: 32KB or more (can operate with limited RAM/ROM resources) ■ Judgment Speed: Several ms to several tens of ms (depends on the performance of the microcontroller used) ■ Calculation Method and Supported Functions: ・Supports quantization (can be processed even on microcontrollers without FPU) ・Supports processing in static memory areas ・Designed with linear kernel adoption for easier estimation of computational load and execution time ■ Implementation Forms: ・Bare Metal Environment: Can be implemented as firmware without assuming an OS or Linux ・RTOS Environment: Can be integrated into periodic processing as a task of a real-time OS ■ Classifier (Algorithm): ・Adopts lightweight classifiers such as SVM (SVC) (no dependency on NPU/GPU) ■ Supported Inputs (Features): ・Various sensor data such as images, vibrations, sounds, currents, temperatures, etc.
Price information
We will provide a quote based on your requirements, target microcontroller, algorithm configuration, and development scope. Please feel free to contact us.
Delivery Time
※It varies depending on the scale of development and your requirements, so please feel free to contact us.
Applications/Examples of results
It can be utilized in industries and embedded fields that require "low latency, offline capability, low power consumption, and privacy protection," where sending data to cloud AI is difficult. 【Main Expected Uses and Applications】 ■ Anomaly Detection and Predictive Maintenance for Equipment - Detection of failure precursors for industrial motors and manufacturing equipment through vibration and abnormal noise data analysis. - Early determination of equipment anomalies by monitoring current and temperature changes. ■ State Assessment and Identification at the Edge - Appearance inspection and simple classification of products using camera images (for sites without network connectivity). - State identification and sensor fusion using a combination of various sensors (sound, current, temperature). ■ Adding AI Functions to Existing Products - Adding judgment functions to existing microcontroller devices that are already installed and operational without significant hardware changes.
catalog（1）
Download All Catalogs
Company information
Since our establishment in 1984, we have contributed to the information society by leveraging electronics technology, computer technology, and their combined technologies. To meet our customers' expectations, we continuously improve our quality management system and strive daily to acquire new technologies as a "research and development-oriented company," developing products and systems with higher added value. Additionally, we are committed to utilizing human resources with a global perspective and focusing on nurturing individuals with rich imagination and a spirit of adventure. As we enter the 21st century, the core of the multimedia society we aim for is the limitless imagination of people. Computers that weave together and shape human thoughts are themselves "crystals of knowledge," intricately designed and developed to surpass human imagination. NDR will continue to possess the energy to organically transform and respond to the demands of the times by enhancing our technological and informational capabilities.
