break silos to deliver next gen computing, AI, security

HiPEAC, an organization representing the European advanced computing research community, has launched its latest roadmap document analyzing the trends in computing systems and recommendations for next generation technology development.

The HiPEAC Vision 2023 in its ninth bi-annual report impresses a sense of urgency in addressing intensifying geopolitical and environmental pressures, identifying six major races for technology leadership, and in addition highlighting the need to break the silos between technology and applications development to gain a holistic view necessary for global optimizations. The six major races are described as races for the ‘next web’, artificial intelligence (AI), new hardware, cybersecurity, sovereignty and sustainability.

The editor-in-chief of the 234-page HiPEAC Vision 2023, Marc Duranton (from CEA), told that when analyzing the pace of change of technology and the global challenges both in industry and society, HiPEAC identified the six major races. Overarching this is the need for better collaboration between different disciplines, in other words breaking down the silos that exist between teams developing different parts of the technology stacks and the applications. He said, “In the the technical domain, the speed of change is astonishing: new breakthroughs in artificial intelligence nearly every week, new technologies such as quantum computing driving large investments, and combinations of technologies leading to new paradigms like the industrial metaverse are just some examples. The challenge now is to get a holistic view, efficiently combining technologies while taming complexity, for example with the help of AI and trusted orchestrators, to guide us into an evolution of computing that is more efficient, sustainable and trustable.’

Adding to this, HiPEAC coordinator Professor Koen De Bosschere (Ghent University), said, “From the geopolitical point of view, technology is increasingly seen as a strategic asset, with different world regions competing for leadership. With global supply chains under serious pressure, more world regions are aiming for technological sovereignty, while cyberattacks are becoming more frequent and intense. Climate change is also becoming increasingly evident, meaning that sustainability is finally being taken seriously, so that optimizing processes to use fewer raw materials and less energy is now a major objective.’

Drawing on the expertise of the HiPEAC community, each chapter of the HIPEAC Vision is dedicated to a different leadership race and features a series of standalone articles laying out the key issues, complete with recommendations. In addition, they also developed three global recommendations which cut across the different ‘races’, as follows:

  • Break silos to gain a holistic view, which is necessary for global optimizations. Promote collaboration between teams, launch project calls to create synergies between domains, establish cross-disciplinary European competence centres and promote open source.
  • Develop tooling for cross-disciplinary and multi-dimensional challenges. A global view is necessary for enhanced optimization, but it is more complex. Technologies such as AI should be used as a ‘helper’ to propose solutions – which should always be validated before implementation.
  • Develop trustable runtime orchestrators able to manage complex systems. In some cases, systems will have to select a large number of options ‘in real time’. This will require the development of trustable orchestrators which are loyal to their users.

The six races

The race for the “next web” – the continuum of computing

More and more data generated by interconnected machines will provide new services. This web linked with the physical world means constraints like latency, time, processing time and localization will need to be promoted like other non-functional requirements (such as cost, energy). These interactions between complex systems involving humans, machines and the physical world form a new dimension of cyber-physical systems (CPS).

Another important element is a swarm of “digital twins” that allows modelling both of parts of the real world or of new artefacts. Digitals twins allow manipulation of variables not easily accessible in the real world (e.g. places where it is impossible to place a sensor) or that cannot be handled (like time: in a digital twin, a simulation, time can be accelerated, stopped, or go backwards, none of which is possible in reality). Digital twins allow prediction of what could happen under certain hypotheses, as well as risk assessment, and therefore enable optimal decisions.

HiPEAC 2023 next web race
The race for the next web (Image: HiPEAC)

This major leap in evolution is creating the premise of a seamless continuum extending from handheld devices, veritable bridges to the digital world in personal, social, and professional spheres, to increasingly resource-rich service infrastructures that host pervasive command-and-control applications with feedback loops from digital and physical sensors and actuators, which span from the so-called edge to the center of the cloud. The convergence between the internet of things (edge devices) and cyber-physical systems (with their non-functional properties) is a central part of this transformation. The “continuum of computing” denotes the fact that the automated computation that allows digital services to assist human needs will be a space where applications will move seamlessly across, towards and along with data, in manners that will depend on a variety of functional and non-functional requirements including latency, privacy, security and energy.

What will emerge from this evolution will be a “next web” where programmable links will connect applications, services, digital twins of things and of humans, into a maze of interactions, which will give rise to an immense variety of higher-level services and applications resulting from personalized orchestration of existing resources. This new web will need open interoperable standards, which will allow digital resources to be spatially localized, and to embed collaborative AI engines that can operate also at the edge and not solely in the deepest recesses of corporate clouds.

HiPEAC makes five recommendations for Europe to be at the forefront in the race for the new web:

  • Launch research and development actions to explore the realization of web-inspired re-interpretation of future computing systems.
  • Initiate proof-of-concept level efforts to develop agreed definitions of functional and non-functional specifications and accompanying programming support for execution- and communication-level interoperability across the continuum of computing. Processing should be done where it is the most efficient according to functional and non-functional criteria.
  • Promote demonstrative testbeds of “trusted orchestrators”, running on any mobile device and appliance to serve the needs of safe, private, personalized and effective service composition on the computing continuum.
  • Provide active support to standardization efforts in this field.

The race for artificial intelligence

The “next web” will require data to be processed intelligently and near to where it was created for efficiency reasons, therefore it will be of paramount importance to have efficient AI systems at the edge or on premises. We are moving towards an era of collaborating AIs rather than “centralized” AI running on (private) clouds. However, large language models (LLMs) have shown very surprising results, and yesterday’s large neural networks are now running on smartphones (e.g. for voice recognition, translation). Therefore, adapting the technology of these LLMs to embedded systems will be of high importance.

HiPEAC 2023 AI race
The race for AI (Image: HiPEAC)

HiPEAC Vision 2023 suggests that Europe should focus on developing embedded AI solutions for better integration/understanding of the physical world, e.g. for near-the-edge and edge devices, (like automotive, smart factories), leveraging its knowhow in cyber-physical systems and embedded systems. As new learning techniques that need fewer data or that do not require a lot of labelled data for learning develop, these algorithms will also be interesting in embedded systems or in on-premises systems. Research in Europe could focus on new approaches that do not require a lot of data for learning and could be used at the edge, and on federated learning approaches that allow privacy to be preserved.

It adds that Europe should not drop out of the race for the “megamodels”. Sufficient computing resources and databases need to be available to European researchers and industry to develop European megamodels (following European precepts and ethics) and experiment with them. Europe should be also at the forefront of techniques to move “large” models into edge devices, with tools (pruning networks, using the right data representation (quantization), or other optimization techniques), efficient algorithms and new edge hardware with high performance capabilities and low energy requirements.

It is also essential to support research and development of tools helping to identify bias and misbehavior manifested by AI. The large models (generally transformer based) show interesting results mixed with nonsense. Solutions to separate the good from the bad should be enforced and learning databases curated. Hybrid techniques (using explicit knowledge to check the results of data driven deep neural networks) are a path to explore.

The race for innovative and new hardware (including quantum)

There is always a race to be more efficient in energy, both for large systems (high-performance computing (HPC), cloud) due to the cost of ownership and for embedded systems for autonomy (longer battery life). In this race for the reduction of energy consumption by computing devices, several new computing paradigms are being researched – such as quantum computing, neuromorphic computing, spintronics and photonic devices.

HiPEAC 2023 hardware race
The race for hardware (Image: HiPEAC)

Of these new paradigms, quantum computing is attracting most of the attention today. Quantum computing requires the development of new devices, as there are several ways to implement devices that exhibit quantum phenomena fit for quantum computing. At the heart of these efforts is materials research. The ongoing research in new materials will have a profound impact on the development of new computing technology in the near future.

Quantum computing has also renewed research in information theory, complexity theory, algorithm development and computer architecture. Both quantum computing and neuromorphic computing have shed new light on information representation and information processing. As the successful von Neumann architecture is here to stay (mainly as orchestrator of various accelerators, potentially using different technologies and information coding), embedding architectures based on new computing paradigms in more traditional architectures is becoming a challenging research field, together with the development of programming models for these architectures. As well as connecting old to new, research is also needed to determine how to leverage and embed devices based on new materials in computing devices.

Future computing devices will be hybrid systems, where each device will process its specific part of the computation. IC fabrication processes require large amounts of energy and fresh water, and due to their toxic waste can put high pressure on the environment. Not all application areas of ICs require state-of-the-art devices; in fact, they can be manufactured with relatively slow devices fabricated with less demanding processes, mostly because of high volumes and low device cost.

HiPEAC makes five recommendations for Europe to be at the forefront in the race for innovative and new hardware:

  • Establish and stimulate competence centers that combine the strength of Europe’s position in the fundamental fields of computing (information theory, complexity theory, computer architecture, algorithm design, programming paradigms) with European industrial initiatives. This will both attract top talent and develop a strong European industrial presence in the field of new computing devices. In this context, the role of pilot-lines will be key.
  • Research in system architecture and software stack for emerging computing paradigms needs to be pushed. The potential success of a novel computing scheme is not only about its core technology but rather in how it fits into a flow that allows useful problems to be solved.
  • Heterogeneous integration will play a key role in developing and maturing new design methods, fabrication techniques and novel packaging for the new computing hardware. A European ecosystem needs to be established, combining research, research transfer organizations (RTOs), industry and application prescribers.
  • Design support and investment programs in the knowledge that the race for new technologies is not only about mastering the technologies (e.g. qubits, or neuromorphic devices). It is also about their integration with infrastructures and industry environments, and with more traditional ICT that will help make it real.
  • Invest in designing and fabricating novel IC technologies that have a significantly lower environmental impact, and at the same time apply the lessons learned from those new fabrication processes to existing ones.

The race for cybersecurity

All the systems mentioned above will only be practically usable if they are secure and safe to use. The arms race between hackers and security professionals is as old as the computer. Some countries are creating cyber armies to protect their infrastructure and to set up attacks against other countries to be able to retaliate if attacked, leading to a cyberwar.

HiPEAC 2023 cybersecurity race
The race for cybersecurity (Image: HiPEAC)

The pervasiveness of ICT provides a broad attack surface, and attacks can be economically devastating, but they can also have tangible or even lethal repercussions on the physical world. Despite several highly acclaimed advancements (e.g. the General Data Protection Regulation-GDPR), the EU still has a great deal of work to do in this regard, particularly to maintain its sovereignty and become a leader in the global competition.

HiPEAC makes four recommendations in the race for cybersecurity:

  • Critical infrastructure and supply chains (utilities, transports, health) should have both their hardware and software components hardened against cyberattacks, with reinforced cyber defenses, monitoring and resilience capabilities (including redundancy). In addition, edge and IoT devices should be hardened against side-channel attacks and reverse engineering.
  • The EU should support and invest in research on methods and tools to make security, including privacy, a first-class citizen during the development of new ICT systems, find vulnerabilities in existing ICT systems (e.g. with static analyses on source code and behavioral analyses at runtime), and automatically prevent or mitigate them (e.g. with automated refactoring tools and blocking systems, possibly involving AI). This would provide the technical bricks to widely improve the cybersecurity of the EU industry.
  • The EU should broaden mandatory security and privacy, EU-based audit and certification of ICT systems, with several levels depending on the criticality of the application and regulate to make ICT systems providers and resellers liable in case of subpar cybersecurity. This would incite the EU industrial ecosystems to advance quickly toward better cybersecurity, and give it an edge in the global competition. The expanded Network and Information Security Directive (NIS2) is a first step in this direction but work in this area must be widely enforced and strengthened.
  • Quantum computing has the potential to be a disruptive advance that would invalidate past and current cryptographic schemes, leaving most systems wide open and exposed. In order to avoid being in such a situation and be ahead of the tide, the EU should invest in post-quantum cryptography (PQC) so as to have EU-designed and EU-validated quantum-resistant encryption schemes, hardware and software.

The race for sovereignty

The globalization of previous decades is falling apart due to protectionism, unstable supply chains, and even war. Covid-19, the Ukrainian war, climate change and the energy crisis have led to unreliable supply chains, high production and transportation costs, all of which has encouraged countries to try to become more self-sufficient. This is especially true for the semiconductor supply chain which basically spans the whole world, with each country owning part of the process.

HiPEAC 2023 sovereignty race
The race for sovereignty (Image: HiPEAC)

Sovereignty is, however, much more than building a factory in Europe: it also requires research, innovation, an entrepreneurial ecosystem, (venture) capital, raw materials, energy, and a well-trained workforce to be successful. For digital sovereignty, it is necessary to not only aim for hardware sovereignty, but also for software sovereignty, i.e. a software stack that is owned by Europe and that is not controlled by foreign companies and agencies. Sovereignty does not necessarily mean having everything done locally; it is about having a (good) bargaining position.

HiPEAC makes four recommendations in the race for sovereignty:

  • Europe should strive for digital sovereignty, both in hardware and in software. Hardware sovereignty is covered by the European Chips Act (although creating new designs with innovative architectures should be further promoted), but software sovereignty is as important. Europe should also work on “digital essentials” that can guarantee that Europe’s essential software infrastructure keeps working uninterruptedly no matter what happens outside Europe. The goal is not necessarily to strive for full digital independence, but for at least digital interdependence with the other global regions to have bargaining power.
  • Europe should promote open source, and create an infrastructure to support the open source software and hardware supply chain in Europe. Open source not only helps to collaborate and to innovate inside Europe, but it also helps to build soft power outside Europe. It can also help to create Europe’s digital essentials.
  • Europe should keep investing in talent. Europe’s higher education system is excellent, but it should pick up two additional important roles: lifelong learning to upskill and reskill the existing (and shrinking) European work force and supporting regional entrepreneurial ecosystems to bring more research results to the market. The creation of well-funded European competence centers will help to retain and attract top talent, and to stay at the forefront of new digital technologies.
  • Europe should keep investing in research and innovation, and in a more entrepreneurial ecosystem that generates lots of startups, scale-ups and global companies. Europe needs more venture capital to support the growth of scaleups into global companies. Key areas to invest in are the silver economy, health, mobility, energy, automation and sustainability.

The race for sustainability

HiPEAC Vision 2023 states that each year humanity consumes 1.75 times the resources Planet Earth can generate in one year. Sustainability means that we should reduce our resource consumption to at most one planet. The production of ICT has a large ecological footprint and should be reduced a far as possible. At the same time, ICT could also help to make the world more sustainable by optimizing processes with a large ecological footprint. Sustainability implies that we will have to find a trade-off between the environmental costs and the benefits of ICT.

HiPEAC 2023 sustainability race
The race for sustainability (Image: HiPEAC)

Sustainability is by far the most important grand challenge of the 21st century, and it won’t be possible without extensive use of digital technology to make efficient use of resources.

HiPEAC makes four recommendations in the race for sustainability:

  • System designers should use a full lifecycle assessment when designing new computing systems: from raw material to device, and 100% back to raw material. Based on such life cycle assessments, Europe should develop sustainability labels for digital devices (e.g. TCO certified). These labels should not only focus on repairability, “refurbishability” and recyclability, but also on socially acceptable mining and production operations.
  • Since operational energy is dwarfed by embodied energy for modern low-power devices, computer architects should first and foremost focus on embodied energy, e.g. by avoiding dark silicon, by limiting the number of cores to the number actually needed, by avoiding accelerators that are only used sporadically), and by maximizing the use and the lifetime of a device and its software.
  • Europe should develop new business models and economic models that also take into account the lifecycle environmental costs in order to make sustainable solutions more competitive with non-sustainable ones.
  • Europe should continue searching for ways to dematerialize goods with a substantial environmental footprint by environmentally friendlier (digital) goods and services.

Breaking the silos

HiPEAC observes a tendency to “closing in” on all levels, from countries (with more emphasis on sovereignty), to the personal level, to our own “tribe” (as “defined” by social media). Tension is becoming exacerbated at all levels between these “tribes”, as evidenced by trade (or real) wars between countries, more extreme political parties, and social media “wars”.

This tendency also exists in technology, where there are application silos and technology silos

Application silos

Often, application domains develop their own stack dedicated to their needs, with little or no interoperability with similar stacks developed by other application domains. HiPEAC sees a multiplicity of solutions, which are more or less interoperable, designed to solve similar problems.

Interoperability of solutions that straddle application domains would increase efficiency and create new synergies and new solutions. For example, synergies between electric-vehicle, home and smart-grid stacks would allow vehicle batteries to be used to power homes during peak hours, therefore reducing the pressure on the electric grid. This trivial example (already operational in some locations) shows the added value of communication and interoperability between these three domains.

Some open-source solutions are also available to offer interoperability between various standards, even for the same application domain. It is therefore one of HiPEAC’s recommendations to have exchanges between applications domains and develop interoperability solutions.

Technology silos

It is a natural tendency to group technologies into structures that are easy to understand and to manage. Each structure can be more or less independent and autonomous and only communicate with its adjacent structures in the global flow. Another similar way to manage complexity is to have various levels of abstraction (as in software or communication layers). Research and development activities have a similar structure. Generally, each structure working on a separate topic is making quite good progress and the field has already been optimized.

However, the sum of local optimizations is less efficient than global optimization. Taking this as a starting point, HiPEAC proposes breaking the silos that exist for:

  • Applications
  • Algorithms that are executed in the application
  • The programming language in which the algorithms are implemented
  • The software stack (compiler, operating system, middleware, libraries) that result from the mapping of the algorithm into a language that is compiled or executed.
  • Data coding (floating point, integers, tensors, “spikes”, etc.)
  • The architecture executing the code
  • The technology in which the architecture is realized.

There are many other steps, these are just a sample selected for simplicity of explanation.

The full 234-page HiPEAC Vision 2023 document can be downloaded here.

Related Contents:

Leave a Comment