Less latency, cost reduction, higher reliability, increased privacy, and enhanced security - these are the advantages of applying Edge-AI in sectors like industrial automation, healthcare, and smart mobility. Joost Sannen explains: ‘Edge-AI brings enormous opportunities. Sioux Technologies helps realize them for our clients by combining our unique expertise in mathware, software, and electronics.’

Artificial Intelligence (AI) is rapidly evolving. Machines are increasingly capable of performing complex tasks once reserved for humans, such as learning, reasoning, and problem-solving. This is largely made possible through massive datasets and algorithms, often processed in the cloud, a global network of data centers where data is stored, managed, and analyzed.

Embedded devices
‘We talk about AI when a computer behaves in a way that seems intelligent,’ says Sannen, Senior Software Architect at Sioux. ‘A popular way to achieve this is through machine learning (ML), where mathematical models are trained, something we're very good at. With Edge-AI, these trained models are used at the edge of the cloud (in mobile devices and local servers) to enable 'intelligent' behavior. At Sioux, we often support our clients in developing cutting-edge technologies: production machines, analytical systems, medical devices, autonomous vehicles… When we talk about Edge-AI, we’re really talking about making embedded devices smarter.’

Growing threats
Why choose Edge-AI over the cloud? Sannen begins with the importance of privacy - keeping sensitive data out of the hands of third parties. ‘Do you want all raw data about your factory's performance ending up on an American cloud server, or would you rather keep it to yourself? Then there's the increasing threat of cybersecurity. Edge-AI significantly reduces the attack surface - all the possible ways a malicious actor can access a digital system.’

Milliseconds matter
Another benefit of Edge-AI is reduced latency; the time between data becoming available and the outcome of its processing. Edge-AI significantly increases processing speed. Sannen: ‘Take quality control during high-throughput mass production. If processing must occur within milliseconds, it’s no help to first send the data to the cloud. The same applies to noise suppression in hearing aids or automatic braking systems in vehicles—they must function in real time.’

More sustainable and cost-efficient
Currently, data centers account for about 1% of global electricity consumption. This is expected to quadruple over the next decade, partly due to the computational demands of the AI revolution. ‘Edge-AI is inherently more energy-efficient than cloud computing, therefore more sustainable and cost-saving in operation,’ says Sannen. ‘It also reduces dependence on network stability, resulting in greater reliability.’

Model and data
What challenges does Sioux tackle when implementing Edge-AI? One major hurdle is physical limitations, Sannen explains. ‘Embedded devices often have limited computing power and may be battery-powered. So how can you still do something useful with AI? This technology isn’t traditionally optimized for energy efficiency. Also, clients often come to us with their own models, when they design the architecture and handle the training themselves. That model, along with the necessary data, is considered core IP and not meant to be shared. It's an exciting challenge to still bring such a model into production.’

Trade-offs
There are many types of embedded hardware for Edge-AI, such as neural processing units from Nvidia (TensorCore), Hailo, and NXP eIQ. Each has its pros and cons regarding efficiency, speed, and functionality. Making the right choice, and sometimes updating it, is essential for optimal Edge-AI application. Sannen: ‘That’s what makes our work so interesting. Plus, there are always trade-offs. We want a device to be as small and affordable as possible. How do you optimize without compromising on the required accuracy? That’s a technical balancing act every time.’

Training and deployment
Sioux possesses multiple in-house competencies that make the company a specialist in Edge-AI. The mathware team focuses primarily on training models and ML Ops: training and deploying efficient, reliable machine learning models. They also conduct technical research, for example, on hardware accelerators using field-programmable gate arrays (FPGAs). Sioux’s embedded software experts handle the interface between hardware and model, and the hardware selection that processes the models. Model optimization for correct speed is typically a shared task.

Preventing overheating
Sannen: ‘Our electronics experts ensure efficient execution units - the hardware needed for optimal Edge-AI integration. A key challenge here is to guarantee enough computing power while keeping energy consumption and temperature under control. The fact that we house all these specialties - mathware, software, and electronics - and work in an interdisciplinary way is what sets us apart. It enables us to get the most out of Edge-AI for our clients, like Nemo Healthcare.’

100x speedup
Nemo Healthcare’s NFMS monitors fetal well-being using electrophysiological signals. This can help prevent delayed interventions when complications arise during pregnancy. This functionality is made possible by Edge-AI. Sioux managed to accelerate the client’s initial model by a factor of 100. Another Sioux project, one of many, involves defining an embedded architecture for the development of a cooperative, connected, automated mobility system for self-driving vehicles.

Tremendous opportunities
Sannen: ‘AI and machine learning aren’t new. These developments have been evolving for decades, long dominated by a few pioneers. But over the past five years, this technology has truly taken off. Nearly every Sioux client is involved, either actively using it or seriously considering it. Edge-AI enables real-time applications not only in healthcare and automotive, but also in production process monitoring, predictive maintenance, analysis of materials and biological samples in analytical instruments, crop health monitoring, and optimization of decentralized energy distribution. The possibilities are virtually limitless. Edge-AI is a revolutionary technology, opening up new opportunities across industries, and Sioux helps bring them to life.’

 

Model.Name