Embedded World Key Takeaways

Introduction: The world of embedded systems is multi-faceted – from hardware and software to services and tools. ABI Research’s Senior Director Andrew Zignani and Senior Analyst Reece Hayden attended the Embedded World Exhibition & Conference in Germany last week. Here they present their key takeaways from the event.

Takeaway #1: New Hardware Was Unveiled, but Conversations Revolved around Software

Numerous announcements across the board introduced more performant Artificial Intelligence (AI) hardware, often including Neural Processing Units (NPUs). However, most conversations did not focus on the performance gains or power efficiency of this hardware, instead focusing on the software stack that is becoming increasingly important for chipset differentiation.

Synaptics introduced Astra, an AI developer platform to support smart Internet of Things (IoT) devices. Other vendors spoke about the vital importance of developer experimentation and the role tools like developer clouds play in bringing down barriers to development. Increasingly, companies like NXP and STMicroelectronics are providing “free” access to resources for developers to run AI model benchmarking tests on “real” hardware. The next step will leverage digital twins to simulate AI models in “real-world” situations to assess performance, power efficiency, and other Key Performance Indicators (KPIs) more realistically.

Takeaway #2: Some Business Models Pivot to Accelerate Embedded AI at Scale

The embedded AI market is a low-margin, high-volume one with vendor profits determined by scaled deployments in the tens of thousands of units. Several vendors have recognized the vital importance of accelerating testing and Proofs of Concept (PoCs) as they target profitable growth.

 Increasingly, this is inducing chip vendors to reduce the cost of using developer platforms and cloud resources to support experimentation and testing. Although this brings risk, companies like Imagimob have decided that developers are unwilling to take on risk themselves without understanding how models perform on specific hardware.

Another aspect is the development of “ready-made” models. While these are a step beyond basic model zoos, they can be tested and deployed without training or fine-tuning and will also help vendors accelerate developer time-to-scale. Hardware and software vendors, especially in the highly fragmented embedded space, will increasingly have to take on commercial risk to access profits determined by deployments at scale.

Takeaway #3: Hardware Vendors Look to Build Developer Value Proposition by Emphasizing Independent Software Vendor (ISV) Partnership Ecosystem and Platform Integrations

Market activity and partnerships from leading chipset vendors highlight the growing importance of software for chipset differentiation. NXP announced a partnership and integration with NVIDIA TAO, aiming to shorten developer time-to-value by providing the vital pre-trained model catalog available through the NGC Catalog, as well as the accompanying edge AI-specific tool chain.

In addition, STMicroelectronics was showcasing its edge AI partnership with Amazon Web Services (AWS). This integration enables developers to leverage AWS SageMaker and IoT Core to support development, deployment, and management of edge AI models. These announcements were not alone, as all chip vendors look to build developer accessibility and provide tools enabling experimentation and accelerated AI deployments.

Next year’s Embedded World event will include numerous announcements from embedded chip vendors looking to widen their software ecosystem through technology partnerships. Edge and device-specific optimization vendors will certainly be interesting to watch, as developers aim to cut costs, improve performance, and implement generative AI across the distributed compute continuum.

Takeaway #4: RISC-V’s Huge Presence Signals Growing Acceptance and Commercial Deployment

RISC-V is now widely accepted by the embedded industry with increasing numbers of commercial partners and “real-world” deployments. Many vendors are even saying that it is just a matter of time before RISC-V dominates chip production.

One of the key drivers cited by numerous companies for interest in RISC-V is the increasing price of Arm Intellectual Property (IP). News reports in early 2023 highlighted that Arm increased chip design prices in a reshaping of its business models; vendors indicated that this has continued with incremental price increases over the last year. This was emphasized by Imagination’s announcement of the APXM-6200 Central Processing Unit (CPU), which directly targets a market segment dominated by Arm-embedded cores.

Another factor is the development of the RISC-V ecosystem beyond just chips. One example is MIPS Technologies, which expanded its collaboration with Synopsys to provide customers access to a processor development kit to support developer accessibility.

Takeaway #5: Generative AI Hype Slightly Subsides in the Embedded Market

Although some generative AI-based applications were on display (AMD running Llama 2-7B on its Ryzen Embedded 8000 processor), commercial offerings were few and far between. Many vendors see customer interest, but significant commercial and technical challenges are hindering deployments at scale. Increasingly, embedded chip vendors are focusing on AI workloads that are more appropriate to the IoT, the edge, and other domains such as Machine Learning (ML) (rather than Deep Learning (DL)). Bringing generative AI to these deployment locations will require more investment in the software stack to support and ensure commercial and technical alignment.

Takeaway #6: Altera Senses Opportunity to Define Spin-off with Edge AI Solution-Focused Vertical Strategy

Altera announced new edge-optimized processors for retail, healthcare, industrial, automotive, defense, and aerospace. The announcement focused on the FPGAi strategy that aims to tightly integrate hardware with the Field Programmable Gate Array (FPGA) AI software suite. This closely reflects wider market dynamics as chip vendors look to lower barriers for developers and drive experimentation.

In addition, this approach will help Altera define its new direction outside of Intel, while still leveraging IP assets that are developed in Intel’s open-source community (OpenVINO). Altera’s highly-verticalized announcement seems shrewd and could help build commercial differentiation from its closest competitor, AMD Xilinx. The challenges remain Intel’s software complexity and the barriers to AI experimentation.

Takeaway #7: Qualcomm Targets Crowded Ultra-Low-Power Wi-Fi Space with QCC730 Wi-Fi 4 Chip

One of the biggest connectivity-related announcements at the Embedded World 2024 show was the unveiling of Qualcomm’s QCC730 Wi-Fi chip. A dual-band 1x1 Wi-Fi 4 solution, the solution claims to offer “micro-power” and is Qualcomm’s lowest-power Wi-Fi solution for IoT connectivity, to date. Qualcomm claims the chip offers an 88% reduction in power consumption compared to its previous generation products, and the goal is to target a variety of applications, including smart door locks, video doorbells, wireless security cameras, sensor devices, and asset tracking tags, as well as other battery-sensitive applications across consumer, healthcare, and industrial segments.

The solution is also being positioned as a Bluetooth® alternative, with the key differentiator being direct to cloud connectivity and enhanced performance via lower latency and higher throughput. The solution also complements its wider IoT connectivity offerings, including the QCC711, an ultra-low power Bluetooth® Low Energy (LE) System-on-Chip (SoC) and the QCC740, a multi-protocol solution supporting Thread, Zigbee, Wi-Fi, and Bluetooth® connectivity. However, there are a number of potential obstacles for Qualcomm in this space.

  • First, it represents somewhat of a strategic shift away from its core Wi-Fi expertise and traditional channels toward more cost- and battery-sensitive IoT-centric applications.
  • Second, Qualcomm is entering a very crowded space with alternative solution providers such as Silicon Labs, Infineon, Nordic Semiconductor, Renesas, Texas Instruments (TI), and InnoPhase, among others, all offering unique solutions that are designed to target similar ultra-low-power applications.
  • Third, Qualcomm’s solution is based on Wi-Fi 4, and much of the competition has shifted its attention and volumes toward Wi-Fi 6 connectivity, which also brings its own unique benefits to reducing power consumption. Nonetheless, it shows the growing demand for ultra-low-power Wi-Fi connectivity as a simple to deploy, globally available, secure, and trusted technology for IoT connectivity across a growing number of battery-sensitive applications.

Takeaway #8: Multi-Protocol and Matter Are at the Forefront of Wireless Connectivity Innovations for the IoT

While Qualcomm’s chip announcement was a standalone Wi-Fi chip, the company was keen to reaffirm its wider portfolio of multi-protocol IoT solutions, and the prevalence of multi-protocol solutions was a clear theme at the show. NXP expanded its MCX series of microcontrollers with the MCX W, a range of wireless Microcontroller Units (MCUs) based on the Arm Cortex®-M33 core. Both the MCX 71 and MCX 72 support Matter, Thread, Zigbee, and Bluetooth® LE, with target applications ranging from wireless sensors to smart appliances, thermostats, and other consumer, commercial, and industrial devices.

In a similar vein, Silicon Labs announced the xG26 portfolio of wireless SoCs and MCUs, key among them the MG26 2.4 Gigahertz (GHz) multi-protocol SoC, claimed to be the most advanced SoC for Matter over Thread. The solution supports Matter, OpenThread, and Zigbee and Bluetooth® LE. On the IP side, Ceva, Inc. launched its Ceva-Waves Links™ family of multi-protocol solutions, the first of which is its Ceva-Waves Links100, a low-power Wi-Fi 6, Bluetooth® 5.4 and 802.15.4 solution for IoT applications. Module vendor Murata also showcased its Type 2EL module, based on NXP’s IW612 tri-radio solution with support for Wi-Fi, Bluetooth® and 802.15.4.

These announcements at Embedded World build on a number of recent multi-protocol solutions being launched, including Atmosic’s ATM34/e, which combines 802.15.4 and Bluetooth® LE. These announcements point to growing momentum for Matter, particularly over 802.15.4, alongside the need for multiple connectivity solutions to work together and meet the growing demands of emerging IoT use cases. ABI Research expects a growing portion of devices to leverage multiple protocols thanks to their ability to reduce design complexity, offer greater flexibility, and provide unique features.

Takeaway #9: LE Audio beyond Headsets—the Emergence of “Talking Sensors”

While there were a number of demonstrations of LE Audio and Auracast broadcast audio for conventional use cases, such as within headsets and multi-stream scenarios, at the show, there has been little discussion, to date, on how the technology can potentially be leveraged within a much wider range of device categories.

Demonstrating one concept of wider usage, Packetcraft and Ezurio were highlighting the potential of “talking sensors,” LE Audio, and Auracast broadcast audio-enabled devices that can communicate directly with headsets, hearing aids, or industrial hearing protection products to provide announcements, readouts, information, and alerts or alarms. Some obvious potential applications could include devices such as alarm systems, gas and smoke detectors, doorbells, smart appliances, industrial sensors, and healthcare devices, as well as other consumer products, that could send audio information directly to headsets. To achieve this, the two companies have developed a Talking Sensor Rapid Prototyping Kit to help developers add this functionality and explore the potential for emerging direct sensor-to-earbud communications.

As ABI Research highlighted in its Bluetooth LE Audio: Market Opportunities and Challenges report, these solutions could create a new way of interacting with our environments and with other Bluetooth®-enabled sensor devices in consumer, commercial, and industrial settings. The combination of audio and data streams via a single LE chip could enable a whole new category of devices that combine data transfer with audio streaming capabilities. This could open new methods of interacting with smart home and other IoT devices, embedding audio and voice control in more and more devices, such as smart appliances and sensors.

Similar to how the arrival of the LE radio in the Bluetooth® 4.0 specification in 2010 created a whole new range of applications for Bluetooth® technology, LE Audio has the potential to vastly transform the way in which we interact with Bluetooth® audio-connected products, and could spur on an unforeseen wave of new audio devices that will likely scale to significant volumes over the next decade.

However, these devices will likely emerge over time only, as LE Audio adoption in headsets increases, and it will be difficult to predict how companies will innovate and combine data transfer with audio applications. Nonetheless, the potential remains considerable.

For ABI Research's latest insight into the AI and connectivity spaces, check out recent content from our analysts in the AI & Machine Learning and Wi-Fi, Bluetooth & Wireless Connectivity research services.

Related Blog Posts

Related Services