Feature: Embedded
Built-in AI support Many processor vendors are now adding support for AI and ML techniques within their frameworks and integrated development environments, but these tools are relatively new and require signifi cant prototyping and exploration to maximise their use in a given application. T erefore this ‘training’ stage will require a new way of thinking about prototyping,
where data capture using a representative prototype must be performed early in the development cycle. Indeed, timing is of the essence; in one
instance, we saw a 100-fold performance improvement when porting a Python- based image-recognition algorithm to a dedicated embedded platform, but only aſt er ensuring the algorithm matches the underlying platform capability.
Software and test engineers of today have a challenge on their hands when it comes to creating future-proof products for tomorrow
Moving to the embedded target a lot
earlier in the design process revealed these optimisation opportunities, and enabled the team to focus on creating the right algorithm to balance performance and quality. Another shiſt required of embedded
developers lies in fi nding a diff erent way to verify and validate their designs. Some aspects of the system will now be tested in terms of confi dence level rather than on an absolute pass or fail basis – something that conservative embedded developers might fi nd uncomfortable. Yet, an appraisal of how other platforms might off er disruptive new approaches to team work fl ow and product development is a necessary part of embedded technology’s continued evolution.
AI in tools development We’ve discussed how AI and ML will infl uence system architecture and impact hardware design, but how will these technologies be used as tools to optimise the development of embedded technologies? Currently we don’t see much technology
Electronics of today must be future-ready
push for AI code generation within the embedded technology space. T is is possibly a refl ection of the inherently risk-averse nature of embedded technology development. Whilst there is likely to be pressure to leverage these tools more in the future, to optimise an otherwise very complicated and time-intensive process, one limitation is the ease with which developers can troubleshoot hardware- level interactions. Generally this requires a deep
understanding of the implementation detail, something that is potentially compromised with AI-generated codebases. Until this is addressed and AI becomes a truly tried, tested and trusted resource, developers will continue to exercise caution until this technology is proven reliable. We expect an imminent surge in
these AI-enabled embedded processing platforms, certainly within the next two years – although the design patterns for using them have not yet been established. Indeed, some of these will fail in the
24 May 2024
www.electronicsworld.co.uk
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48
