Managing the increasing complexity of systems is another crucial aspect. It is driven by the diversity and dependencies of its components as well as by the requested functions and performance, affecting concepts, architectures and operation at all levels of the system hierarchy. Therefore, the reference architectures for future AI-based systems need to provide modular and scalable solutions that support interoperability and interfaces among platforms that can collaborate, exchange information and share computing resources to allow the functional evolution of the silicon-born embedded systems, ready for the computing continuum evolution.

Still, it is observed that the strategical backbone technologies to realize such new architectures are not available. These strategical backbone technologies include smart and scalable electronic components and systems, the AI accelerator and control hardware and software, the security engines, and the connectivity technologies.

To achieve this, balanced mechanisms must be developed to ensure performance while maintaining interoperability between diverse systems. AI techniques play a vital role in complexity management by enabling self-optimizing, self-reconfiguring, and self-managing systems. These self-X capabilities allow systems to adapt dynamically to changing conditions, enhancing their reliability and resilience, but they are not available yet to the required extent.

The current SoCs are reaching a very high complexity by integrating a lot of functions and different accelerators on the same die. A decomposition of the SoCs into chiplets for more flexibility and scalability (e.g., mix and match) may help dealing with the complexity but for that an ecosystem which facilitates the design, and also provides plug and play capable, interoperable chiplets, is required.

A holistic end-to-end approach is necessary to manage the increasing complexity of systems, to remain competitive and to continuously innovate the European electronic components and systems ecosystem. This approach involves technologies described in other chapters such as AI for system conception described in methods and tools chapter. Besides theses, here is a short list of themes that need to be promoted in order to reduce global complexity of systems in general, and systems supporting AI techniques in particular:

• Tools and techniques leveraging AI to help in the management of complexity, e.g. tools to adapt LLMs to edge / embedded targets are required. For example, the development of AI-based HW/SW for multi-tasking and providing techniques to adapt the trained model to produce close or expected outputs when provided with a different but related set of data. The new solutions must provide dynamic transfer learning, by assuring the transfer of training instance, feature representation, parameters, and relational knowledge from the existing trained AI model to a new one that addresses the new target task. They should also ensure hyperparameter tuning for automated machine learning scenarios, etc. Furthermore, environments for modeling how the various parts interact together should be considered since this topic is still a challenge; having accurate digital twins of complex systems with the right level of abstraction for the various uses (conception, runtime optimization, etc.) is still an unsolved challenge.

• New design and architecture concepts including HW-SW codesign with corresponding HW/SW platforms for AI-born embedded systems should be promoted to facilitate trust by providing the dependable design techniques, that enable the end-to-end AI systems to be scalable, make correct decisions in a repetitive manner, provide mechanisms to be transparent, explainable, interpretable, and able to achieve interpretable results and embed features for AI model’s and interfaces' interpretability. Linked to the previous point, infrastructure for the secure and safe execution of AI should be created.

• Enabling factors to manage the complexity of the continuum of computing are the use and development of standardized APIs for hardware and software tool chains, methods and interoperability across different system layers like sensors, gateways, on-premise servers and edge processing units. Additionally, interoperability concepts for AI edge-based platforms for data tagging, training, deployment, and analysis should be supported, allowing the development of distributed edge computing architecture with AI models running on distributed devices, servers, or gateways away from data centers or cloud servers. This implies scalable and hardware agnostic AI models capable of delivering comparable performance on different computing platforms, (e.g. Intel, AMD or ARM, Risc-V architectures) and seamless and secure integration at HW/SW embedded systems with the AI models integrated in the SW/HW and APIs to support configurable data integrated with enterprise authentication technologies through standards-based methods.

• Deterministic behaviors, low latency and reliable communications are also important for other vertical applications, such as connected cars, where edge computing and AI represent “the” enabling technology, independently from the sustainability aspects. The evolution of 5G is strongly dependent on edge computing and multi-access edge computing (MEC) developments.

In this chapter, five Key Focus Areas have been identified to tackle this complexity challenge:

• Complexity Management Utilizing AI

• Decomposition of Complex SoCs into Chiplets and Interposers

• Ensuring Programmability and Interoperability

• Combining Processing Devices to Work Together

• Modeling Interactions Among System Components

With the power of embedded AI, we can now unleash the self-X concept which didn’t took off in the past. We can now design HW/SW techniques and architectures that can dynamically monitor their behavior, predict default (predictive maintenance), allowing devices to self-optimize, reconfigure and self-manage the resource demands (e.g. memory management, power consumption, AI model selection, etc…). This will have a direct effect in increasing the lifetime of devices and increasing their up-time.

Supporting the increasing lifespan of devices and systems can mainly be based on three main principles:

• Upgradability

• Modularity

• Reuse

By extension, this means that devices and systems have to increase their scalability and interoperability.

Upgradability

This requires hardware that can accommodate for example software upgrades, ensuring that devices remain functional and up-to-date over extended periods. This is the drive for approaches such as Software Defined Vehicle.

The first level of lifetime extension is clearly the upgrade to avoid replacing an object but instead improving its features and performance through either hardware or software update. This concept is not new as it has already been applied in several industrial domains for dozens of years.

Hardware requirements for edge processing devices that support software upgrades must address several critical factors to ensure that these devices remain functional and up-to-date over extended periods. The hardware must have sufficient computational power and memory to handle the evolving complexity of software applications. Storage capacity must be flexible and expandable, allowing for the installation of future updates and new applications without compromising performance. The devices must also incorporate robust connectivity options, such as advanced wireless protocols and multiple I/O ports, to ensure seamless integration with new peripherals and communication technologies that may emerge over time. This has an initial cost because the hardware is perhaps over-dimensioned for the exact initial requirements but is compensated for the user by its potential increased lifetime.

Secondly, hardware for edge processing devices must be compatible with various operating systems, RTOS and software ecosystems, providing the flexibility needed to adapt to different applications and use cases over time. This compatibility ensures that the devices can support a wide range of software upgrades and maintain interoperability with other systems. On the smartphone level, some companies (Google, Samsung), are guaranteeing that their device will be able to support software upgrade for 7 years. This is an important point that should also happen in the other domains of edge computing. But new disruptive innovations (like generative IA) can invalidate this long lifetime because of new drastic hardware requirements (mainly memory and requirement of a NPU for embedded generative IA).

Modularity

Enhancing interoperability and modularity allows different generations of devices to work together, extending their usability and reducing electronic waste. It would lead to systems which are more eco-conceived and with an augmented sustainability. The hardware should be designed with modularity in mind, enabling easy replacement or upgrading of individual components. This implies a long-lasting chip to chip communication protocol between units (like I₂C, SPI, PCI-e, USB, Ethernet, MIPI, AXI, etc.). Modularity at the level of chiplets is not yet here (adding/removing/upgrading chiplets). This modular approach not only extends the device's operational life by allowing for incremental hardware improvements but also reduces electronic waste and the overall cost of ownership.

Interoperability, modularity, scalability, virtualization, upgradability are well-known in embedded systems and are already widely applied. But they are brand new in AI and nearly non-existent in edge AI. On top, self-x (learning/training, configuration or reconfiguration, adaptation, etc.) are very promising but still under research or low level of development. Federative learning and prediction on the fly will certainly take a large place in future edge AI systems where many similar equipment collect data (Smartphone, electrical vehicles, etc.) and could be improved and refreshed continuously.

Reuse

The last aspect of increasing lifetime is to reuse a system in an application framework less demanding in terms of performance, power consumption, safety, etc. Developing the concept of a second life for components further contributes to sustainability by repurposing older hardware for new applications. Having hardware compatible with standard protocols (such as the one used for the web or cloud) could improve the chances for such a reuse.

Ensuring Long-Term Functionality and Up-to-Date Operation

A fundamental aspect of supporting the increasing lifespan of devices is ensuring that they remain functional and up-to-date over extended periods. This involves not only maintaining their operational capabilities but also providing ongoing support for software updates and new features. By keeping devices updated with the latest advancements and security patches, manufacturers can extend their usability and relevance, thereby maximizing their value to users and reducing electronic waste. This may incur over-specification to be able to be future proof.

• Dynamic reconfiguration and Self-X techs

Intelligent reconfigurable concepts are an essential key technology for increasing the re-use and service life of hardware and software components. Such modular solutions on system level require the consideration of different quality or development stages of sensors, software, or AI solutions. If the resulting uncertainties (measurements, predictions, estimates by virtual sensors, etc.) are considered in networked control concepts, the interoperability of agents/objects of different generations can be designed in an optimal way.

• Dynamic reconfiguration: a critical feature of the AI circuits is to dynamically change their functions in real-time to match the computing needs of the software, AI algorithms and the data available, and create software-defined AI circuits and virtualise AI functions on different computing platforms. The use of reconfigurable computing technology for IoT devices with AI capabilities allows hardware architecture and functions to change with software providing scalability, flexibility, high performance, and low power consumption for the hardware. The reconfigurable computing architectures, integrated into AI-based circuits can support several AI algorithms (e.g. convolutional neural network (CNN), fully connected neural network, recursive neural network (RNN), etc.) and increase the accuracy, performance and energy efficiency of the algorithms that are integrated as part of software-defined functions.

• Realizing self-X (adaptation, reconfiguration, etc.): for embedded systems self-adaptation, self-reconfiguration has an enormous potential in many applications. Usually in self-reorganizing systems the major issue is how to self-reorganise while preserving the key parameters of a system (performance, power consumption, real-time constraints, etc.). For any system, there is an operating area which is defined in the multi-dimensional operating parameter space and coherent with the requirements. Of course, very often the real operating conditions are not covering the whole operating domain for which the system was initially designed. Thanks to AI, when some malfunctioning parts are identified, it could be possible to assess, relying on AI and the data accumulated during system operation, if it affects the behaviors of the system regarding its real operating conditions. If this is not the case, it could be considered that the system can continue to work, with maybe some limitations, but which are not vital regarding normal operation. It would then extend its lifetime “in place”. The second case is to better understand the degraded part of a system and then identify its new operating space. This can be used to decide how it could be integrated in another application making sure that the new operating space of the new part is compatible with the operating requirements of the new hosting system.

• Self-learning techniques are promising. Prediction on Natural Language Understanding (NLU) on the fly or keyboard typing, predictive maintenance on mechanical systems (e.g. motors) are more and more studied. Many domains can benefit of the AI in mobility, smart building, and communication infrastructure. LLMs shows interesting properties in this field, such as few-shot learning.

• Monitoring the devices and systems health

Dynamic reconfiguration and Self-X Techs also need data to be as efficient as possible. It means that monitoring of devices and systems health are necessary to acquire those data. Several techniques exist.

• Distributed monitoring: continuous monitoring and diagnosis also play a crucial role for the optimization of product lifetime. Where a large amount of data is collected during daily life operation (e.g. usage, environment, sensor data), big data analysis techniques can be used to predictively manipulate the operational strategy, e.g. to extend service life. Similarly, an increase in power efficiency can be achieved by adjusting the calibration in individual agents. For example, consider a fuel cell electric vehicle where the operation strategy decisively determines durability and service life. Distributed monitoring collects data from various interconnected agents in real-time (e.g. a truck platoon, an aircraft swarm, a smart electricity distribution network, a fleet of electric vehicles) and uses these data to draw conclusions about the state of the overall system (e.g. the state of health or state of function). This allows to detect shifting behavior or faulty conditions in the systems and to even isolate them by attributing causes to changes in individual agents in the network or even ageing of individual objects and components. Such detection should be accomplished by analyzing the continuous data stream that is available in the network of agents. A statistical or model-based comparison of the individual objects with each other provides additional insights. Thus, for example, early failures of individual systems could be predicted in advance. This monitoring should also cover the performance of the semiconductor devices themselves, especially to characterize and adjust to ageing and environmental effects and adjust operations accordingly.

• Distributed predictive optimization is possible, whenever information about future events in a complex system is available. Examples are load predictions in networked traffic control or demand forecasts in smart energy supply networks. In automation, a concept dual to control is monitoring and state observation, leading to safety-aware and reconfigurable automation systems. Naturally, all these concepts, as they concern complex distributed systems, must rely on the availability of vast data, which is commonly associated with the term big data. Note that in distributed systems the information content of big data is mostly processed, condensed, and evaluated locally thus relieving both communication and computational infrastructure.

• The other area of lifetime extension is how AI could identify very low signals in a noisy data environment. In the case of predictive maintenance, for instance, it is difficult for complex machinery to identify a potential failing part early in advance. The more complex the machinery, the less feasible it is to obtain a complete analytic view of the system, which would allow for simulation and thus identification of potential problems in advance. Thanks to AI and collecting large datasets it is possible to extract some very complex patterns which could allow for very early identification of parts with a potential problem. AI could not only identify these parts but also give some advice regarding when the exchange of a part is needed before failure, and then help in maintenance task planning.

• Artificial Intelligence challenges

Secure software upgradability is necessary in nearly all systems now and hardware should be able to support future updates. AI introduces additional constraints compared to previous systems. Multiplicity of AI approaches (Machine learning, DL, semantic, symbolic, etc.), multiplicity of neural network architectures based on a huge diversity of neuron types (CNN, RNN, LLMs, etc.), potential complete reconfiguration of neural networks for a same system (linked to a same use case) with a retraining phase based on an adapted set of data make upgradability much more complex. This this why HW/SW, related stacks, tools, data sets compatible with the edge AI system must be developed in synergy. HW/SW plasticity is necessary whatever the AI background principle of each system is, to make them as much as possible upgradable and interoperable and to extend the system lifetime. HW virtualization will help to achieve this, as well as standardisation. The key point is that lifespan extensions, like power management, are requirements which must be considered from day one of the design of the system. It is impossible to introduce them near the end without a strong rework.

One challenge for the AI edge model is upgradability of the firmware and new learning/training algorithms for edge devices. This includes the updates over-the-air and the device management of the updating of AI/ML algorithms based on the training and retraining of the networks (e.g. neural networks, etc.) that for IoT devices at the edge is very much distributed and is adapted to the various devices. The challenge of the AI edge inference model is to gather data for training to refine the inference model as there is no continuous feedback loop for providing this data. The related security questions regarding model confidentiality, data privacy etc. need to be addressed specifically for such fleets of devices.

The novelty with AI systems is to upgrade while preserving and guaranteeing the same level of safety and performance. For previous systems based on conventional algorithmic approaches, the behavior of the system could be evaluated offline in validating the upgrade with a predefined data set representative enough for the operating conditions, knowing that, more than the data themselves, the way they are processed is important. In the case of AI, things are completely different, as the way data are processed is not typically immediately understandable, but what is key is the data set itself and the results it produces. In these conditions it is important to have frameworks where people could reasonably validate their modification, whether it is hardware or software, in order to guarantee the adequate level of performance and safety, especially for systems which are human-life-critical. Another upgrade-related challenge is that of designing systems with a sufficient degree of architectural heterogeneity to cope with the performance demands of AI and machine learning algorithms, but at the same time to be flexible enough to adapt to the fast-moving constraints of AI algorithms. Whereas the design of a new embedded architecture or electronic device, even of moderate complexity, takes typically 1-3 years, AI models such as Deep Neural Networks are outdated in just months by new networks. Often, new AI models employ different algorithmic strategies from older ones, outdating fixed-function hardware accelerators and necessitating the design of hardware which functionality can be updated.

Designing with Modularity and Extensibility in Mind

To achieve long-term functionality, devices need to be designed with modularity and extensibility as core principles. Modularity allows to extend the functionality of a chip for example by externally adding another support chip. For this to be practically usable means that a specific chip-to-chip interface should be added and internally connected to ensure this modularity at board/system level. This design philosophy supports easier maintenance and enables incremental improvements, which can significantly extend the device's lifespan. Extensibility ensures that the device can adapt to new technologies and requirements over time at system level, adding extra features with new supporting chips. This seems at the cost of integration and performance, but new techniques could be developed that overcome most of the disadvantages. For example, Compression Attached Memory Module (CAMM)⁵³ still allows to replace the memories in a system like a PC, while the recent trend is to integrate memory directly on a (passive) interposer, thus preventing any memory upgrade.

Advancing Modularity at the Chiplet Level

While the concept of modularity is well-established at board and system level, achieving it at the level of chiplets is an emerging challenge that has not yet been fully realized. Chiplet-based architectures involve breaking down a monolithic System on Chip (SoC) into smaller, interconnected modules (chiplets) that can be individually upgraded or replaced. This approach promises greater flexibility, improved scalability, and easier integration of new technologies. However, the industry is still in the early stages of developing standardized, interoperable chiplet designs. Advancing this modularity at the chiplet level is crucial for enabling devices to support prolonged lifespans.

Virtual Modularity

Virtual Modularity refers to the dynamic integration and seamless interaction between multiple devices within an interconnected ecosystem, enabling them to function as a unified system. This concept allows different hardware components and devices to pool their resources and capabilities, creating a flexible, scalable environment where services and tasks can be effortlessly transferred or shared across the network. As a result, users experience a cohesive, uninterrupted operation, regardless of which device they are using, as the underlying system intelligently manages and allocates resources to optimize performance and deliver a consistent user experience across all connected devices⁵⁴.

Virtual Modularity can increase the lifespan of devices by distributing workloads and optimizing resource usage across a network of connected devices, rather than relying on a single device to handle all tasks. Since different devices in the ecosystem can share and collaborate on tasks, the wear and tear on individual devices is reduced. For example, a smartphone may offload demanding processing tasks to a more powerful device, such as a tablet or computer, thereby reducing its own strain and energy consumption. This collaborative approach helps prevent any one device from being overworked, which can extend its operational life. Additionally, as new devices are added to the system, older devices can continue to contribute and function effectively within the network, further extending their useful life by leveraging the collective power and capabilities of the entire ecosystem. Older devices can have their lifetime increased thanks to the extra computing and storage resources delivered by other devices in the pool.

Reuse of components or systems in a downgraded or requalified use case

Re-use: One concept called the “2^nd life” is actually the re-use of parts of systems. Such re-use could be adapted to edge AI as far as some basic rules are followed. First, it is possible to extract the edge AI HW/SW module which is performing a set of functions. For example, this module performs classification for images, movement detection, sound recognition, etc. Second, the edge AI module can be requalified and recertified downgrading its quality level. A module implemented in aeronautic systems could be reused in automotive or industrial applications. A module used in industrial could be reused in consumer applications. Third, an AI system may be re-trained⁵⁵ to fit the “2nd life” similar use case, going for example from smart manufacturing to smart home. Last, the business model will be affordable only if such “2nd life” use is on a significant volume scale. A specific edge AI embedded module integrated in tens of thousands of cars could be removed and transferred in a new consumer product being sold on the market.

Edge computing and Artificial Intelligence are quickly becoming more and more tied together, converging towards the unified concept of Embedded Artificial Intelligence. To ensure the growth of this emerging technologies it is crucial that the European ecosystem of companies and academia can strategically cover all the steps of the associated value chain. This means covering the entire stack starting from the Embedded Artificial Intelligence hardware, including all the software layers required to exploit the hardware functionalities and to develop the final applications, the engineering of the tools for AI development and the data sets, with a trustable and certifiable environment.

Economic and environmental sustainability represents to key factors when defining actions and investments promoting the coverage of the value chain and of the technology stack. Technology is strongly affected by sustainability that, very often, tips the scale between the ones that are promising, but not practically usable, and the ones making the difference. E.g. cloud computing, based on data centers, plays a fundamental role for the digitalization process. However, data centers consume a lot of resources (energy⁵⁶, water, etc.), they are responsible for significant carbon emissions during their entire lifecycle and generate a lot of electronic and chemical waste.

Today, the percentage of worldwide electricity consumed by datacenters and data transmission networks is 2 to 3%. “Rapid improvements in energy efficiency have, however, helped moderate growth in energy demand. Data centres and data transmission networks are responsible for 1% of energy-related GHG emissions” ⁵⁷. A report from January 2024 of IEA⁵⁸ indicates that “Global electricity demand from data centers could double towards 2026” (see Figure 2.1.15). Interestingly, the electricity demand to process crypto-currencies seems still to surpass the consumption related to AI in 2026 (see Figure 2.1.16).

Figure 2.1.: growth of cryptocurrency and AI centric data centers between 2022 and 2026, from ⁷²

Data centers are progressively becoming more efficient, but shifting the computing to the edge, for example, allows to temporally reduce data traffic, data centers storage and processing. However, only a new computing paradigm could significantly reduce their environmental footprint and ensure sustainability. Edge Computing could contribute to reach this goal by the introduction of ultra-low power and efficient computing solutions, and Embedded Intelligence (and collaborative AI agents approach) can significantly contribute to optimize their operations and, directly and indirectly, their overall energy consumption profile.

Shifting to green energy is certainly a complementary approach to ensure sustainability, but the conjunction of AI and edge computing, the Embedded AI, has the potential to provide intrinsically sustainable solutions with a wider and more consolidated impact. Indeed, a more effective and longer-term approach to sustainable digitalization implies reconsidering the current models adopted for data storage, filtering, analysis, processing (including AI training itself), and communication. By embracing Edge Computing, for example, it is possible to significantly reduce the amount of useless and wasteful data flowing to and from the cloud and data centers, because it means adopting architectural and structural solutions that are more efficient and that permanently reduce the overall power consumption. These solutions also bring other important benefits such as real-time data analysis, reduced memory and storage capacity and better data protection. The Edge Computing paradigm also makes AI more sustainable: it is evident that cloud-based machine learning inference is characterized by a huge network load, with a serious impact on power consumption and huge costs for organizations. Transferring machine learning inference and data pruning to the edge, for example, could exponentially decrease the digitization costs and enable sustainable businesses. To avoid this type of drawbacks, new AI algorithms specifically tailored for embedded systems, in conjunction with new HW components (e.g. based on neuromorphic architectures), should be designed and developed. Considering the application areas, in some cases, this could lead to more specialized and extremely efficient solutions.

Sustainability of Edge Computing and AI is affected by many technological factors, which require investments to consolidate and strengthen EU global positioning, generating a potential positive fallback on the sustainability of future digitalization solutions and related applications. GAMAM already master these technologies and are already controlling the complete value chain and technology stack associated with them: to achieve strategic autonomy in this field, Europe must cover the entire value chain, consolidating the existing segments, identifying/filling the technology gaps, and leveraging on a strong cooperation between the European stakeholders, with a particular attention to SMEs (which generate a large part of European GDP). From this perspective, European coordination to develop AI, edge computing and Embedded AI technologies is fundamental to create a sustainable value chain, based on solid alliances, and capable to support the European key vertical applications.

Ensuring sustainability for Embedded AI in Europe involves several key initiatives: open-source hardware can drive innovation and reduce costs, while energy efficiency improvements are critical for sustainable development. Engineering support represents the third key factor and is necessary to advance sustainable Embedded AI from several perspectives: optimize the energy profile of Embedded AI solutions (technical and environmental levels), tackle the lack of expertise and of human resources (operational/engineering level), and keep the costs of these solutions sustainable (economic). For example, by mastering the adoption of new Embedded AI techniques, such as transformers and large language models (LLMs) with a focus on federation of smaller specialized models developed in Europe⁵⁹, and integrating them with efficient accelerators and algorithms, Europe can maintain a competitive edge. This is crucial for many vertical domains that are key for Europe: for example, in the domain of autonomous systems, Europe has the potential to lead in the development of autonomous vehicles and robots, leveraging advanced AI and embedded intelligence. The necessity to ensure an efficient engineering process, with consequent short time-to-market, is motivated by the speed of global market evolution: for example, the emergence of cheap Chinese RISC-V microcontrollers emphasizes the necessity for fast development, robust, cost-effective computing solutions made in Europe. Another example from a vertical application perspective is the partitioning of complex systems on a chip (SoC) into chiplets and interposers, which requires the development of an ecosystem of interoperable chiplets, comprehensive architecture and engineering solutions to support this approach: the automotive market can significantly benefit from this, maintaining Europe at the global market forefront.

Sustainability can also be measured in terms of democratizing technology use, to make it more accessible to a broader community of users and facilitate market entry to startups and SMEs: natural language interfaces for example could contribute to facilitate the access to digital technologies. This requires that the associated AI models are trained with European data covering European languages, respecting European ethics. Additionally, democratizing technology also means supporting the diversity of European computing systems/chips, simplifying their programmability and consolidating this diversity through interoperability. This is also crucial to support the research into new computing paradigms, like neuromorphic and analog computing, and to ensure they can reach high TRL levels (i.e. 7, 8 and 9).

One of the major challenges that need to be accounted for in the next few years is related to the design of progressively more complex electronic systems to support advanced functionalities such as AI and cognitive functionality, especially on the edge. This is particularly challenging in the European landscape, which is dominated by small and medium enterprises (SMEs) with only some large actors that can invest and support larger-scale projects. To consolidate and strengthen European competitiveness and ensure sustainability in advanced Embedded AI solutions it is therefore crucial to create an ecosystem covering the entire value chain, in which SMEs can cooperate and increase their level of innovation and productivity. The definition of open industrial standards and a market of Intellectual Properties (IPs) is crucial to accelerate the design and development of these solutions, guaranteeing EU competitiveness and increasing the market dimensions. Open-source software, hardware and engineering tools can play an extremely important role in this regard, because they allow to reduce engineering costs for licensing and verification significantly, lowering the entry barrier to design innovative products.

Leveraging Open-source Hardware and Software for Innovation and Cost Reduction

Open-source Hardware and Software are drivers of innovation and cost reduction in the fields of Edge Computing and Embedded AI, allowing democratization of design and creating a collaborative environment where innovation can emerge and making the development of cutting-edge solutions that are both affordable and highly effective possible. Open-source also facilitates transparency and security, which are essential factors for building trust and compliance with European standards.

Developing and Federating Smaller Specialized AI Models

The federation of smaller specialized AI models developed in Europe represents a strategic approach to harnessing localized expertise and enhancing the efficiency of AI applications at the edge. These models, when federated (in “Agentic AI”), can work together to tackle complex tasks as, or even more effectively than monolithic models. Promoting the development of these models in Europe contributes to the alignment of technology with European technologies and applications necessities, leveraging regional strengths in AI research and development. Specialized agents can be derived from few foundation models (made by few European companies or from Open Source) by fine tuning that can be achieved by SMEs or start-ups, as this specialization requires less computing power and data than the creation of a foundation model.

Training Models with European Data and Ethical Compliance

For AI models to be truly sustainable and beneficial to Europe, they must be trained on data that covers European languages and reflects the diverse cultural and social contexts of the continent. This ensures that AI applications are more accurate, inclusive, and useful across different European regions. Additionally, adherence to European ethical standards in AI development is crucial. This can be achieved:

• Finding solutions which ensure privacy, fairness, and transparency in AI systems, thereby aligning technological advancements with the core values and regulations of Europe.

• Allowing those AI to work at the edge, hence promoting the development of specialized AI chips for edge AI, this will facilitate access to digital technologies to a broad audience.

• Supporting natural language interfaces which enable users to interact with technology in intuitive ways, breaking down barriers to adoption and use, in a continent characterized by cultural, societal and language diversity.

• Increasing the re-use and sharing of knowledge and models generated by embedded intelligence, improving the energy- and cost-efficiency of AI training, and adopting a new benchmarking AI approach oriented to sustainability.

Deploying Efficient and Sustainable Embedded AI-oriented ECS

Efficient ECS like new memories, accelerators and algorithms specifically conceived for Embedded AI are fundamental to the performance and sustainability of the domain. Developing and deploying these technologies within Europe ensures that AI applications can operate with high efficiency, reducing energy consumption and enhancing overall system performance. Most current designs involve perception-based AI accelerators, but Europe should be at the forefront to use new disruptive technologies such as generative AI and ensuring its sovereignty by developing the HW and SW that will enable it at the edge. This will require for example:

• New materials and substrates oriented to low power consumption.

• New generations of embedded ultra-low power electronic components.

• 3D-based device scaling for low power consumption and high level of integration.

• Chiplet-based solutions promoting modularity and reuse (sustainability by design).

• Strategies for self-powering nodes/systems on the edge and efficient cooling solutions.

• Policies and operational algorithms for power consumption at edge computing level.

• Efficient and secure code mobility.

• Advanced Neuromorphic components.

• Generative AI for the edge.

• Inclusion of existing embedded systems on the edge (huge market opportunity).

Accelerating Development of Robust, Cost-Effective Solutions

There is a pressing need for the rapid development of robust, cost-effective computing solutions made in Europe. These solutions must be designed to meet the specific demands of European markets and regulatory environments. By focusing on speed of realization, reliability, and cost-effectiveness, European manufacturers can remain competitive and responsive to global and local market needs. This involves developing the full spectrum of technologies, from hardware innovations to sophisticated software solutions, to new tools allowing to increase productivity, ensuring that European products stand out in the global marketplace. For example, this must include:

• Engineering process automation for full lifecycle support of Embedded AI solutions.

• Edge Computing and Embedded AI security and sustainability by design.

• Engineering support for Embedded AI verification and certification, addressing end-to-end edge solutions.

• Support for education and professional training to enable, facilitate, improve and speed-up the design, development and deploy of Embedded AI.

• Engineering process automation based on generative AI.