The State of Silicon and Devices – Q1 2025 Roundup

Welcome to the first quarterly roundup on the State of Silicon and Devices by Canonical. 

Q1 has seen lots of announcements in the areas of Edge AI and cybersecurity. Companies across the semiconductor and software ecosystem are pushing toward more powerful, energy-efficient AI models at the edge, while simultaneously strengthening their security posture to meet compliance requirements like the Cyber Resilience Act (CRA) in Europe. Similarly, many governments and industry bodies now mandate cybersecurity measures in automotive systems, and at Canonical, we recently announced ISO 21434 certification.

At Canonical, we see first-hand how secure, embedded and AI-driven systems are having a noticeable impact on our customers. The pace of innovation is rapid, so to help you keep an eye on the latest trends, we’ve compiled a roundup of the latest advancements in silicon and devices.

Arm announces the first Armv9 Edge AI Platform

Let’s begin our roundup with a focus on Edge AI.

Arm recently introduced the first Armv9 edge AI platform, integrating the Cortex-A320 CPU and Ethos-U85 NPU, optimized for on-device AI with models exceeding one billion parameters. This platform delivers an 8× improvement in “machine learning performance” over Cortex-M85-based solutions and has gained industry support from AWS, Siemens, and Renesas.

As a member of Arm’s partner ecosystem, we’re thrilled to learn about Arm’s latest push towards edge AI, and can’t wait to see the benefits this brings to the IoT ecosystem. Amongst the most exciting features of the new Armv9 platform are improvements for Cortex-A320, which is optimized for industrial automation, smart cameras, and HMI applications. Arm also highlighted 10× “machine learning performance” and 30% scalar performance uplift over Cortex-A35.

The Ethos-U85 NPU (the second component of the new platform), on the other hand, aims to facilitate AI acceleration with transformer support. It features high-efficiency neural processing optimized for low-power, AI-driven edge devices, and it supports transformer-based AI models, improving NLP and vision applications in IoT and embedded systems. This development is part of a wider trend of manufacturers seeking to elevate the performance of their platforms in order to cater for increasingly specialized workloads, in commercialized environments where efficiency and energy usage are key considerations. 

Credit: https://newsroom.arm.com/blog/introducing-arm-cortex-a320-cpu

Qualcomm to acquire Edge Impulse

Staying with the theme of Edge AI and acquisitions, Qualcomm announced it will acquire Edge Impulse, which provides an edge AI platform supporting developers in designing and deploying AI models across a range of embedded and industrial IoT devices. 

Edge Impulse is a SAAS (software-as-a-service) platform which lets users fetch raw data from devices whilst they are running, as well as perform data analysis, create and train an AI model, and validate and upload it to the device. This automates the process and makes it easier to develop AI-based applications.  The platform brings together microcontrollers, DSPs, and AI accelerators from multiple semiconductor vendors.

The acquisition aims to advance on-device AI capabilities for IoT applications by integrating Edge Impulse’s AI-driven development platform with Qualcomm’s AI-optimized silicon and software ecosystem. Example production use cases that would benefit from such capabilities include real-time inferencing for applications such as anomaly detection, predictive maintenance, computer vision, and sensor fusion. 

Post-acquisition, Qualcomm intends to integrate Edge Impulse’s development stack with its Dragonwing processors, including the QCS6490 and QCS5430, optimizing AI inferencing performance through the Qualcomm AI Hub. This integration is expected to deliver up to a 4x increase in inference efficiency, improved memory utilization, and reduced power consumption by leveraging Qualcomm’s advanced AI acceleration and heterogeneous compute architecture. Qualcomm’s broader IoT strategy, which encompasses a unified software framework and an extensive chipset portfolio, will be further reinforced by Edge Impulse’s scalable AI model deployment framework. This will be sure to accelerate AI adoption across industrial automation, smart cities, healthcare, and logistics.

When a market leader doubles down on AI inference efficiency, it signals that on-device intelligence is not just a temporary trend, but a real shift in computing strategy. We believe the acquisition of Edge Impulse reinforces the shift toward accelerating AI at the edge.

We expect inference efficiency will remain a priority, as battery life, thermal constraints, and real-time performance are all critical for edge AI workloads. The stated improvements in inference efficiency mean lower energy consumption and faster decision-making, which is relevant in sectors like industrial automation and smart cities where our secure, OTA-updatable Linux solutions play a role.

Having covered some of the major announcements in the area of Edge AI, let us now shift our focus to the broader theme of cybersecurity.

Canonical achieves ISO/SAE 21434 certification

Security has been a crucial theme in this first quarter of the year.

At Canonical, we announced our achievement of the ISO/SAE 21434 certification for our Security Management System, following an extensive assessment by TÜV SÜD. ISO/SAE 21434 provides a detailed framework for managing the risks of unauthorized access, remote attacks and data breaches  in connected vehicles across their entire lifecycle. The certification aligns with our  work on functional safety under ISO 26262 and contributions to initiatives like the Enabling Linux in Safety Applications (ELISA) project. Achieving ISO/SAE 21434 certification reinforces our commitment to delivering secure solutions in an increasingly complex security landscape. As open source continues to play a growing role in automotive and embedded systems, we are committed to ensuring that open source software meets the highest industry standards. As security is a fundamental requirement, we will continue to invest in building reliable solutions for our customers.

Exein and MediaTek partner on embedded cybersecurity 

Building on our security focus, Exein and MediaTek announced a strategic collaboration.  Italy-based Exein offers technology to secure IoT devices and containers from cyber attacks and vulnerabilities. It serves various industries such as automotive, healthcare, and manufacturing, and has offices in Italy, USA, and Germany.

In a recent announcement, Exein and Mediatek shared that Exein’s cybersecurity technology will be integrated into MediaTek’s Genio platform, beginning with the 24.1 release of MediaTek IoT Yocto. This is particularly important in enabling security on a massive scale, given emerging mandatory security regulations around the world, such as the Cyber Resilience Act (CRA) in Europe. 

Manufacturers of embedded devices are particularly affected by this upcoming legislation: the CRA mandates stringent security measures for devices, prompting manufacturers to adopt robust frameworks that ensure security and reliability. When considering CRA compliance, device manufacturers must ensure their operating systems (OS) include robust measures to protect user data and device integrity. For more information on how device manufacturers should prepare, we recommend viewing our 50 minute webinar on this topic.

Exein is using extended Berkeley Packet Filter (eBPF) to meet their compliance requirements. eBPF is a technology that can run sandboxed programs in a kernel without needing to change the kernel source code or load additional kernel modules. By enabling the extension of kernel capabilities by executing custom programs without adding modules, eBPF offers visibility into network activity, resource utilization, and more. 

At Canonical, we recently showcased some of the tools available in Ubuntu that make use of eBPF, and we also closely collaborate with companies offering eBPF-based solutions. For instance, Polar Signals’ eBPF-based Parca Agent is facilitating continuous profiling. In collaboration with them, we’ve made a commitment that, beginning with Ubuntu 24.04 LTS, our GNU Compiler Collection (GCC) package will enable frame pointers by default for 64-bit platforms. 

Apple introduces its first cellular modem

Moving away from the topics of cybersecurity compliance and Edge AI, another noteworthy development for the quarter came from Apple. In particular, Apple’s iPhone 16e debuts the C1 modem, its first in-house cellular chip, with TSMC reportedly powering Apple’s iPhone 16e with an A18 Chip Built on N3E and C1 Modem on 4nm.

The company claims the C1 chip is the “most power-efficient modem ever in an iPhone”. Furthermore, the Apple-designed 5G modem will provide support for 4G LTE and sub-6GHz 5G.  It will also work closely with the A18 chip to optimize data flow and battery life.

Apple’s modem strategy signals a move toward vertical integration, which could limit open-source modem alternatives due to its proprietary ecosystem. As Apple designs its hardware and software to work in a tightly controlled ecosystem, by developing its own modem, they will no longer work with third-party suppliers which provide Linux-compatible modem drivers. At Canonical we will keep tracking how open source alternatives, from Qualcomm’s Linux-based modem drivers, to RISC-V cellular projects, will evolve in response to this trend.

Whilst we understand the industry trends towards seeking tighter hardware-software integration, we believe Apple’s move reinforces the importance of open 5G modem projects and Linux support for cellular connectivity. In particular, alternative 5G modem vendors may emerge, possibly seeking Linux-based integrations for open source cellular stacks.

Ambient IoT Alliance is launched

Building on the importance of efficient, connected hardware, the Ambient IoT Alliance (AIoTA) was launched jointly by Intel, Qualcomm, Infineon, Atmosic, PepsiCo, VusionGroup, and Wiliot. The alliance’s goal is to drive the adoption of battery-free, energy-harvesting IoT devices, leveraging Wi-Fi (IEEE 802.11bp), Bluetooth, and 5G Advanced (3GPP Release 18+) for large-scale, real-time tracking and logistics automation.

Canonical has long been pushing for secure, containerized solutions in the low-end embedded compute spectrum. We welcome AIoTA’s focus on battery-free, energy-harvesting IoT devices, as those devices will require lightweight, efficient operating systems that support OTA updates and secure communication.

Indeed, Ambient IoT extends traditional RFID and BLE technologies by enabling low-power, energy-harvesting sensors to operate without dedicated power sources. These devices utilize RF, photovoltaic, kinetic, and thermal energy harvesting to support continuous data transmission. They communicate telemetry such as geolocation, temperature, and humidity via existing wireless infrastructure, including smartphones, routers, and IoT gateways.

AIoTA’s emphasis on device AI inference and machine-learning-driven cloud optimizations is exciting for us at Canonical. As we have been working on Canonical Kubernetes , and AI/ML model deployment on the edge, we are excited about the opportunity to provide lightweight, AI-optimized Linux runtimes for edge devices operating in ultra-low-power environments.

Concluding remarks

This quarter has seen the convergence of AI, security, and efficient computing at the edge. From Arm’s Edge AI platform to NXP’s AI expansion and Qualcomm’s Edge Impulse acquisition, the industry is moving toward on-device intelligence and resilient, energy-efficient systems. For any readers who are looking to step up their AI capabilities, this is an exciting moment to explore the hardware options that are out there. 

 But what lies on the horizon? Looking ahead, we expect regulatory pressures to introduce short-term challenges as companies adapt to new compliance requirements. Readers should ensure that they are prepared for the disruption this may bring, although ultimately these regulations will ultimately drive stronger security practices across the industry, particularly for embedded and IoT devices. At the same time, we anticipate continued acquisitions as companies race to secure AI and security expertise from others with experience in this field.  

We are committed to supporting this transformation with trusted open source that fuels AI adoption. We will continue to enable AI workloads, optimize embedded computing, and strengthen open source security frameworks.