Designing a simple sensor based Internet of Things (IoT) device is not difficult, but building an IoT device with edge machine learning (ML) processing capabilities is a completely different matter. The dedicated processor series, development boards, and accompanying software launched by NXP Semiconductors aim to solve key challenges in functionality, performance, and development, helping to deploy complex edge artificial intelligence (AI) functions faster in industrial and IoT applications.
Designers have started utilizing edge AI solutions that can perform ML inference on low-power devices without relying on cloud resources. Functions such as wake-up word detection, sensor data pattern analysis, and basic object detection can typically be handled by energy-efficient processors running ML models (built using model optimization tools and frameworks). However, bottlenecks arise when you try to expand processor resources to handle more complex problems, especially those that require real-time or near real-time response.
How multi-core processors significantly accelerate ML inference
NXP Semiconductors, with its i.MX 93 series application processors, can easily address the functional and performance challenges of these emerging real-time edge AI designs (Figure 1).
Figure 1: The i.MX 93 application processor integrates processing resources, security systems, memory, and a complete range of clocks, timers, connection options, and interfaces, laying the foundation for edge AI design. (Image source: NXP Semiconductors)
This series of processors integrates rich features, including multimedia, storage, interfaces, and connection options, and combines eye-catching processing resources:
Up to two high-performance Arm Cortex-A55 application processor cores for Linux based application processing tasks
An ultra-low power Arm Cortex-M33 platform for low latency real-time control processing
An Arm Ethos-U65 microNPU neural processing unit (NPU) for efficient execution of ML inference
NXP's integrated EdgeLock Secure Enclave (ESE) provides a root of trust for secure boot and key management, real-time encryption, and other features required to protect edge applications
By leveraging the capabilities of these processors, large edge AI applications can be broken down into multiple easily manageable parts: NPUs take over the computational tasks of dense neural network algorithms, reducing the load on Cortex-A55 cores and avoiding preemption of their running application code resources. At the same time, the Cortex-M33 core continues to focus on processing low latency tasks such as sensor data acquisition or process control, while the embedded ESE safeguards system security, software code, and critical data throughout the entire process. The following will introduce the ability of NPU to offload machine learning inference from the Cortex-A55 core, which is a key support for achieving near real-time responsive edge AI applications.
How hardware development boards and software accelerate application development
Although the functionality and performance of the processor are crucial, the efficient development of edge AI applications relies more on the ability to quickly grasp the characteristics of the processor and quickly build effective software. The FRDM-IMX93 development board from NXP (Figure 2), combined with accompanying software development resources, can provide everything needed to start creating applications.