(Source: zapp2photo – stock.adobe.com)
The architectures of embedded systems continue to evolve as newer technologies become available to design engineers. What used to be a single central processing unit surrounded by interface and logic circuitry is now a multiprocessor multi-core design with advanced integrated peripheral functionality—many with their own dedicated microcontrollers and resources to contend with.
The once clearly delineated computer edge is now a much vaster and more distributed architecture. Computers that were islands to themselves are now a brick in the wall as they integrate into larger global high-speed connectivity. The hierarchal nature of globally and locally distributed microcontrollers means dedicated processing can occur closer to where data is being generated. The network is the computer.
As higher-resolution data is sensed and processed, chunks of information extraction at each level often result in huge data sets that require a massive pool of immediate and virtual low-latency storage. Smart houses integrate into a higher level of the distributed processing hierarchy to behave in a city management process domain. Eventually, the need for our power grid to become a dynamically intelligent entity will be more apparent as electric vehicles and charges become the norm.
In all cases, the Internet of Things (IoT) is central. Everything is connected: from handheld devices to cars and homes, traffic lights to biohazard sensors, factories to cities. This results in each system having multiple edges—some secure, some less so. In this evolving landscape of connectivity, machine learning is crucial as security concerns continually increase with hostile forces trying to wreak havoc.
Designers of modern factories, appliances, and more seek machine learning (ML) solutions to tackle new challenges and allow companies to advance and compete. However, low latencies are required to allow useful data integration from extensive and varied sources. As a result, data must be accessed and transported globally very quickly to make these higher-level "thinking machines" respond in an effective and reasonable amount of time.
This poses a challenge. Process-intensive security schemes and algorithms can add latency, meaning hardware security acceleration is needed and must be robust enough to protect against threats. Often, vital infrastructure is online and requires high-security protection.
To meet these processing and security requirements, this new generation of designs calls for highly integrated MCUs. Communications data rates have grown from 50 bits per second (bps) to 10 gigabits per second (Gbps). Software-controlled bit banging will not work at these higher rates, so very high-speed communications hardware must be incorporated into modern microcontrollers.
Additionally, wireless options like Wi-Fi® and Bluetooth®, as well as wired technologies like Ethernet and USB 3.x communications, must peacefully coexist in one system. Intelligent edge devices like facility routers can implement each of these simultaneously. Modern microcontrollers house several encapsulated protocol standards using dedicated accelerated hardware to further unburden CPUs.
Analog and digital signal processing (DSP) requirements must also be addressed, especially with the many IoT devices that include sensor systems as part of their functional requirements. DSP functionality is critical to meet response times for industrial IoT control loops, whether it be simple thermostats or complex high-speed analog data streams from real-time factories. All this increased performance and functionality must be low power and feature power management techniques, as server farms already use tremendous amounts of energy.
NXP Semiconductors' new MCX N Series advanced processor engines are general-purpose and application-specific microcontrollers based on Arm® Cortex®-M33 CPUs. The MCX N Series devices feature a rich mix of hardware-accelerated peripherals, communications, and signal processing, focusing on scalability and ease of development (Figure 1).
Figure 1: The high-speed Arm processing cores are surrounded by advanced memory interfaces and controllers, analog converters and DSP accelerators, communications interfaces, and security. (Source: NXP Semiconductors)
At the heart of the microcontroller is a pair of 150MHz Cortex-M33 processors with up to 2MB of flash memory (code and data) and 512kB of SRAM. Additional dedicated hardware for DSP and CORDIC (pseudo multiplication and division) acceleration functions allows fast implementation of math-intensive algorithm control loops. One of the key components of this microcontroller is a hardware accelerator, eIQ® Neutron Neural Processing Unit (NPU).
The MCX N Series features a plethora of communications interfaces, including Ethernet, USB FS (12Mbits/sec), USB HS (480Mbits/sec), I³C, CAN FD, and UART, as well as advanced mixed-signal interfaces like four single-ended or two differential 16-bit analog-to-digital converter (ADC) channels, 12- and 14-bit digital-to-analog converters (DACs), op amps, comparators, and stable temperature voltage references. A digital Synchronous Audio Interface (SAI) can generate waveforms and levels, and touch-sensing interfaces reduce form factors for handheld and space-constrained designs.
The microcontrollers also feature NXP’s EdgeLock® Secure Enclave, Core Profile to ensure device-wide security intelligence, run time attestation, silicon root trust, and key management. Other security features include extensive cryptographic services, trust provisioning, and simplified path certifications. A high-performance crypto engine is also built in to perform on-the-fly encryption and decryption that provides a secure boot.
An important feature of the NXP MCX N microcontrollers is the elQ® Neutron NPU (Figure 2). The NPU's scalability to billions or trillions of operations per cycle has the power to implement different learning techniques. One technique, the convolutional neural network (CNN) architecture, is a feed-forward neural network that learns feature engineering through filters. Also supported are recurrent neural networks (RNN), which are well suited for sequential time series data sets as part of the deep learning algorithms. Temporal convolution networks (TCN), which employ casual convolutions and dilations, provide adaptive learning and are also supported by the NXP elQ Neutron NPU.
Figure 2: The MCX N microcontrollers’ scalable eIQ NPU is a dedicated controller core with in-line dequantization, activation, and pooling. It boasts learning times that are 30 times faster than conventional computational cores. (Source: NXP Semiconductors)
Unified software tools are essential for design engineers and software developers. The NXP MCUXpresso suite of development software and tools can help designers with a variety of tasks, including RTOS development.
A choice of integrated development environments (IDEs), including MCUXpresso for VS Code, MCUXpresso IDE, IAR Embedded Workbench and Arm Keil MDK, work with a configuration tool to configure peripherals and multifunction pins. Bootables and utilities work with debug tools like Segger, NXP MCU-Link, FreeMASTER, and PEmicro to provide breakpoint, trace, and control/monitor functions that aid in software and firmware development. GUI Guider helps design HMI graphics libraries and displays.
The elQ machine learning software development environment is essential, leveraging inference engines, neural network compilers, optimized libraries, and deep learning toolkits. The hardware abstraction layers support importing Pytorch, TensorFlow and ONNX models and running inference with TensorFlow Lite for Microcontrollers. These neural nets help learn and output useful information for sensor data in anomaly detection and predictive maintenance, voice and keyword recognition response, as well as image and video processing for object detection and recognition.
The eIQ Toolkit provides a machine learning workflow and tools that enable graphical or command line interface to build, profile, and deploy your machine learning models with runtime insights. The eIQ portal is a graphic tool that lets designers create, optimize, debug, and export machine learning models. It can also import models from TensorFlow.
NXP offers a wide range of application software, examples, reference designs, application notes, articles, and blogs on their website for applications such as audio, cloud connectivity, machine learning, motor control, RTOS, security, voice processing, touch sensing, and both wired and wireless connectivity.
Finding the right processor solution can take time and research. Not every design needs high-end supercomputing horsepower, but many applications in the Industrial and IoT edge require a significant amount of processing power and high-end peripherals to perform.
The NXP MCX N Series is designed to meet these needs and more as it addresses the challenges of modern intelligent edge applications, such as IoT, security, power constraints, machine learning, and cost. With a mature and comprehensive development environment, the MCX N Series is ready to help engineers develop their designs, including newer technologies like neural networking and machine learning.
After completing his studies in electrical engineering, Jon Gabay has worked with defense, commercial, industrial, consumer, energy, and medical companies as a design engineer, firmware coder, system designer, research scientist, and product developer. As an alternative energy researcher and inventor, he has been involved with automation technology since he founded and ran Dedicated Devices Corp. up until 2004. Since then, he has been doing research and development, writing articles, and developing technologies for next-generation engineers and students.
NXP Semiconductors enables secure connections and infrastructure for a smarter world, advancing solutions that make lives easier, better and safer. As the world leader in secure connectivity solutions for embedded applications, NXP is driving innovation in the secure connected vehicle, end-to-end security and privacy and smart connected solutions markets.