The run-up to the electronica 2020 virtual conference has been buzzing. As we prepare for the various virtual events — including the embedded forum...
The run-up to the electronica 2020 virtual conference has been buzzing. As we prepare for the various virtual events — including the embedded forum — scheduled to coincide with the trade fair, at EE Times Europe we’re also pondering the implications of proposed mega-mergers like Nvidia-Arm and AMD-Xilinx, and staying on top of the announcements for new microcontrollers, development tools, and software that daily cross our desks.
A pervasive theme in embedded for the past couple of years has been the focus on artificial intelligence at the edge. The shift to edge processing is a result of the demands of almost every industry vertical, including automotive, industrial, and medical, to respond more nimbly to locally gathered intelligence, enhancing security and privacy in the process.
In fact, quoting IDC Futurescape predictions, Intel recently said that up to 70% of all enterprises will be processing data at the edge by 2023, while the edge silicon sector will be a US$65 billion market opportunity by 2024. In its September announcement of new processors for the internet of things, the company said that the new 11th Gen Intel Core processors, Intel Atom x6000E series, and Intel Pentium and Celeron N and J series bring new AI, security, functional safety, and real-time capabilities to edge customers.
Intel’s 11th Gen Core is enhanced specifically for essential IoT applications that require high-speed processing, computer vision, and low-latency deterministic computing. Compared with the previous generation, it delivers up to a 23% performance gain in single-thread performance, a 19% gain in multithread performance, and up to a 2.95× performance gain in graphics. New dual-video decode boxes allow the processor to ingest up to 40 simultaneous video streams at 1080p (30 frames per second) and output up to four channels of 4K or two channels of 8K video. AI-inferencing algorithms can run on up to 96 graphics execution units (INT8) or run on the CPU with vector neural network instructions (VNNIs) built in.
Meanwhile, the Intel Atom x6000E series and Intel Pentium and Celeron N and J series processors represent Intel’s first processor platforms enhanced for IoT. They deliver enhanced real-time performance and efficiency; up to 2× better 3D graphics; a dedicated real-time offload engine; the Intel Programmable Services Engine, which supports out-of-band and in-band remote device management; enhanced I/O and storage options; and integrated 2.5 Gigabit Ethernet (GbE) time-sensitive networking. They can support 4Kp60 resolution on up to three simultaneous displays, meet strict functional safety requirements with the Intel Safety Island, and include built-in hardware-based security.
Plug-in AI acceleration modules
Other recent edge announcements include AI chipmaker Hailo’s launch of its M.2 and Mini PCIe high-performance AI acceleration modules for empowering edge devices. Based on the company’s Hailo-8 processor, the modules can be plugged into a variety of edge devices, enabling AI capabilities to be deployed in smart cities, smart retail, Industry 4.0, and smart homes. The modules enable customers to integrate high-performance AI capabilities into edge devices, providing an optimized solution for accelerating a range of deep-learning–based applications with high efficiency. The standard form factor speeds time to market.
The AI acceleration modules seamlessly integrate into standard frameworks, such as TensorFlow and ONNX, which are both supported by Hailo’s dataflow compiler. Customers can quickly port their neural networks into the Hailo-8 processor. The M.2 module is already integrated into the next generation of Foxconn’s BOXiedge — a 24-core mini-server that only needs 30 W to run — with no redesign required for the PCB. The solution provides good energy efficiency for standalone AI inference nodes.
While intelligence can be placed in any part of edge electronics systems, it often makes the most sense to do so in the sensors. For example, STMicroelectronics released a high-accuracy, low-power, two-axis digital inclinometer that embeds a programmable machine learning (ML) core to integrate AI algorithms in the sensor itself, thereby reducing power consumption and minimizing data transfer to the cloud.
The company’s IIS2ICLX two-axis accelerometer can sense the tilt with respect to a horizontal plane along two axes (pitch and roll) or, by combining the two axes, can measure the tilt with high and repeatable accuracy and resolution with respect to a single direction of the horizontal plane over a range of ±180°. The digital output simplifies system design and reduces bill-of-materials (BOM) cost by saving external digital-to-analog conversion or filtering.
The accelerometer’s dedicated core for ML processing provides system flexibility, allowing some algorithms run in the application processor to be moved to the microelectromechanical system (MEMS) sensor and thereby enabling a reduction in power consumption. The ML core logic provides the ability to determine whether a data pattern (for example, motion, pressure, temperature, magnetic data) matches a user-defined set of classes. Typical applications include anomalous vibration recognition, complex movement or condition identification, and activity detection.
Connectivity solutions are also an important aspect of connecting embedded systems. Silicon Labs recently expanded its Bluetooth Low Energy portfolio with the BGM220S, which at 6 × 6 mm is claimed to be one of the world’s smallest Bluetooth systems-in-package (SiPs). The ultra-compact, low-cost, long-battery–life SiP module adds turnkey Bluetooth connectivity to extremely small products. A slightly larger PCB variant, the BGM220P, is optimized for wireless performance along with a better link budget for greater range. Both support Bluetooth Direction Finding and delivery up to a 10-year battery life from a single coin cell.
IoT security throughout the life cycle
Security is a critical part of designing embedded systems that are connected to the IoT, and especially for management of the secure device throughout its life cycle. To this end, Infineon Technologies has announced a highly integrated IoT life-cycle management solution to help IoT device makers reduce firmware development risks. Infineon’s solution combines the PSoC 64 secure microcontrollers with Trusted Firmware-M embedded security, the Arm Mbed IoT operating system, and the Arm Pelion IoT platform to design, manage, and update IoT products securely, all without the need for custom security firmware. The Pelion-ready and Mbed OS-enabled solution demonstrates industry best practices for security by reaching PSA Certified Level 1.
EDA tool for heterogeneous systems
On the tools and software front, a new EDA design tool is claimed to be the first commercially available platform enabling fully integrated, design environment-agnostic IC and packaging co-design of 2D/2.5D/3D heterogenous systems. The GENIO tool, from Rome-based startup Monozukuri Technologies, integrates existing silicon and package EDA flows to create full co-design and I/O optimization of complex multichip designs that comprise advanced heterogeneous microelectronic systems. GENIO is the first tool to seamlessly work across all existing EDA flows to maximize design efficiency and system optimization, according to the company. The cross-hierarchical, 3D-aware design methodologies streamline the IC ecosystem, integrating IC and advanced packaging design to ensure full system-level optimization, shortening the design cycles, accelerating time to manufacturing, and improving yields.
The industry is moving to heterogeneous integration to overcome the limitations of Moore’s Law in enhancing performance. But the diverse nature of silicon and package technologies and the use of high-speed interconnections call for enhancements in co-design capabilities to enable planning, implementation, and analysis across different engineering domains as well as to provide a complete, consistent model across tool platforms.
This article was first published on EE Times Europe