A guide to ISO 2768, the standard that sets default permissible deviations for dimensions that carry no individual tolerance.

ISO 2768 General Tolerances: Tables, Classes, Current Status

Embedded AI moved from concept to working prototype at the first-ever embedded world North America Hackathon, where student teams were challenged to turn real-world problems into functional Edge AI solutions.The competition brought together 52 students across 12 teams, who had two days to develop solutions across four challenges: industrial automation, sensing, sustainability and robotics. Each team worked with an Arduino UNO Q, while the hackathon was powered by Arduino, Qualcomm, Quilter and ASUS.[1]<h2>Two Days of Learning, Building and Problem-Solving</h2>Rather than simply proposing concepts, participants had to develop working solutions within a tightly constrained timeframe. Teams had to learn unfamiliar tools, build hardware and software systems, train or integrate AI models, troubleshoot problems and demonstrate their results.That intensity was reflected in comments from Giovanni Salinas, Technical Advisor and Judge from Supplyframe, who said the challenges required students to learn new skills before they could even begin developing their solutions. He praised their drive, positive energy and problem-solving approach throughout the two-day event.Ben Jordan, Judge from Quilter, highlighted another dimension of the competition. He noted that although the solutions were creative and technology-driven, the results showed the continuing importance of human judgement in deciding what should be built and how technology should address practical problems.<h1><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791997928144-1791997928143.jpeg" width="602" height="401" class="fr-fic fr-dii" />Team 10. Source: Supplyframe.</h1><h2>Sustainability: ZOT ROT or NOT</h2>Team 10, ZOT ROT or NOT, won the Sustainability Challenge with a camera-powered conveyor designed to distinguish recyclable material, organic waste, and trash. [1]The team's Hackaday.io project describes a system in which a camera captures objects moving along a conveyor while the Arduino UNO Q processes the images locally. The system classifies items into the three categories and communicates the result to an OLED display, LEDs, and the sorting mechanism. [2]The team also documented its Edge Impulse object-detection model. Its initial unoptimised model achieved 89.6% validation accuracy, with an estimated on-device latency of 5 milliseconds. [3]<a href="https://hackaday.io/project/206767-zot-rot-or-not" target="_blank" rel="noopener noreferrer">Explore ZOT ROT or NOT on Hackaday.io</a>The project demonstrated a practical application of local computer vision: instead of relying on remote processing, the classification pipeline operates directly on the embedded system.<h1><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791997928150-1791997928150.jpeg" width="602" height="401" class="fr-fic fr-dii" />Team 5. Source: Supplyframe.</h1><h2>Robotics: EmbedX</h2>Team 5, EmbedX, won the Robotics Challenge with an on-device AI system for robotic motion and navigation.[1]The project combined machine vision with robotic control, demonstrating how an embedded platform can connect perception with physical movement. The challenge highlighted an important characteristic of Edge AI: intelligence becomes more valuable when it can respond directly to information collected from the surrounding environment.Rather than treating AI as a separate software function, the project connected sensing, computation, and physical action into a single system.<a href="https://hackaday.io/project/206753-team-embedx-robotics" target="_blank" rel="noopener noreferrer">Explore the EmbedX Robotics project on Hackaday.io</a><h1><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791997928156-1791997928156.jpeg" width="602" height="401" class="fr-fic fr-dii" />Team 3. Source: Supplyframe.</h1><h2>Industrial Automation: CRATE</h2>Team 3's CRATE — Chironomy Recognition As A Touchless Experience won the Industrial Automation Challenge.[1]The project uses an Arduino UNO Q, a camera, and a gesture-recognition model to identify hand gestures and translate them into commands. The team's Hackaday documentation states that it used a YOLOv5-based classifier and a dataset containing more than 450 images captured during the hackathon. [4]The final system reportedly achieved an average inference time of about 175 milliseconds and a 0.94 mAP@50 score, while displaying recognised gestures through the Arduino's LED matrix. [5]CRATE demonstrated how computer vision can be used to create alternative human-machine interfaces, particularly in environments where touch-based interaction may not always be convenient.<a href="https://hackaday.io/project/206741-crate-ewna-hackathon-team-3" target="_blank" rel="noopener noreferrer">Explore CRATE on Hackaday.io</a><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791997928162-1791997928162.jpeg" width="602" height="401" class="fr-fic fr-dii" />Team 8. Source: Supplyframe.<h2>Sensing: Zotduino</h2>Team 8, Zotduino, won the Sensing Challenge with a sound-event classification system designed around the needs of an elderly person living alone. [1]The concept focused on recognising potentially important household sounds, including events such as falls, broken glass and fire alarms, while maintaining privacy. According to the team's project documentation, sound classification was performed locally on the Arduino UNO Q, meaning audio did not need to be sent away from the device for processing. [6]The team explained that its design deliberately avoided cameras and cloud-based audio processing. A local classifier analyses the sound, while the system can provide a visual indication through an LED matrix.[6]This approach demonstrated one of the practical benefits of Edge AI: intelligent sensing can happen close to the data source while reducing the need to transmit sensitive information elsewhere.<a href="https://hackaday.io/project/206738-ewna-hackathon-team-8" target="_blank" rel="noopener noreferrer">Explore Zotduino on Hackaday.io</a><h2>Making Embedded AI More Accessible</h2>Across the four winning projects, a common theme emerged: AI was not treated as an isolated software capability. Each team connected machine learning with physical hardware, sensors and real-world actions.Rajan Mistry, a judge from Quilter, described the experience as an opportunity for students to learn new technologies, overcome engineering challenges and turn ideas into reality. He also pointed to platforms such as the Arduino UNO Q as making innovation more accessible.The projects reflected that accessibility in different ways. ZOT ROT or NOT applied computer vision to waste sorting; CRATE used vision for touchless interaction; Zotduino explored privacy-conscious sound classification; and EmbedX connected AI with robotics.<h1><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791997928168-1791997928168.jpeg" width="602" height="401" class="fr-fic fr-dii" />2026 embedded world North America Hackathon. Source: Supplyframe.</h1><h2>From Ideas to Working Systems</h2>The 2026 hackathon demonstrated the growing practicality of Edge AI in embedded development. Students took a problem, selected appropriate sensing and computing tools, integrated AI, and built a working prototype in just two days. [1]Perhaps the most important takeaway was not any single project, but the workflow shared by all four teams: understand a practical problem, experiment with available technology, iterate quickly, and connect intelligence to the physical world.As embedded AI continues to develop, that ability to move quickly from an idea to a functioning system could become increasingly important. The hackathon clearly demonstrated that the next generation of embedded engineers is already experimenting with that approach.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791997928173-1791997928173.jpeg" width="602" height="401" class="fr-fic fr-dii" />2026 embedded world North America Hackathon Sponsors. Source: Supplyframe.<hr /><h3>References</h3><ol><li>Hackaday. Post-event announcement for the first-ever embedded world North America Hackathon. LinkedIn, 2026.<a href="https://www.linkedin.com/company/hackaday/"> https://www.linkedin.com/company/hackaday/</a></li><li>Hackaday.io, “ZOT ROT or NOT.” Project description and system overview for the camera-powered waste-sorting conveyor using Arduino UNO Q. https://hackaday.io/project/206767-zot-rot-or-not </li><li>Hackaday.io, “Object Detection Model ZOT ROT or NOT.” Model accuracy, dataset and on-device latency details. https://hackaday.io/project/206767-zot-rot-or-not/log/251074-object-detection-model </li><li>Hackaday.io, “CRATE: EWNA Hackathon Team 3 — Files.” Dataset and YOLOv5 model information. https://hackaday.io/project/206741-crate-ewna-hackathon-team-3</li><li>Hackaday.io, “CRATE: EWNA Hackathon Team 3  Final log.” Inference performance, model score and system details. https://hackaday.io/project/206741-crate-ewna-hackathon-team-3/log/251101-final-log</li><li>Hackaday.io, “EWNA Hackathon Team 8 — How we came up with the idea.” Privacy-focused sound classification concept and local processing approach. https://hackaday.io/project/206738-ewna-hackathon-team-8/log/251059-how-we-came-up-with-the-idea</li></ol>

Highlights from the 2026 embedded world North America Hackathon

Edge AI in Action

The energy sector is at a critical point, where providers are at once facing concerns related to climate change, resource supply, and geopolitical insecurities while also having to meet growing demand for energy around the globe. All these intersecting factors have been challenging to balance but have been top of mind within the energy segment. As McKinsey &amp; Company stated in its 2025 Global Energy Perspective report, “Energy affordability, reliability (including energy security at the national or regional level), and emission reduction continue to form a trio of priorities that drive energy decision-making.”[1]The challenge remains clear: how can energy suppliers make energy generation more efficient and sustainable—in line with climate and emissions goals—while keeping operations economically competitive? In recent years, additive manufacturing has presented itself as a potential solution to this hurdle. That is, the technology can enable greater sustainability (even within non-renewable energy sectors), boost supply chain resilience, and support an economic shift towards greater energy resource efficiency. At <a href="https://www.wevolver.com/article/automotive-and-additive-manufacturing-converge-at-formnext-2026">Formnext 2026, the leading global additive manufacturing</a> trade show, AM’s growing role in the energy and oil &amp; gas sectors will be spotlighted, with various real-world use cases on display, exhibitors presenting energy AM solutions from across the value chain, and discussions on the latest trends and innovations in the field. <h2>Why Energy is Embracing AM</h2>Organizations across the energy sector are realizing the benefits of <a href="https://www.wevolver.com/article/the-additive-manufacturing-process-chain-and-its-industry-interfaces">additive manufacturing</a> and are increasingly turning to the technology to solve issues related to supply chains, productivity, and overall efficiency. At present, there are specific areas where AM is having the biggest impact; they include:<ul><li>On-demand manufacturing of spare parts, including obsolete or hard-to-source components that keep energy generation systems up and running and minimize system downtimes. </li><li>Repairing worn or damaged structures using large-scale metal deposition processes to extend the lifetimes of legacy systems and minimize system downtimes. </li><li>Designing and producing higher-performing, more efficient, and more lightweight systems like heat exchangers, turbines, energy storage devices, fuel cells, and more. </li></ul><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791995531805-1791995531805.jpeg" width="518" height="345" class="fr-fic fr-dii" />3D printer filament. Source: AdobeStock.Across these three applications, AM is unlocking greater sustainability for energy-related manufacturing by minimizing material waste, enabling distributed manufacturing, and reducing transport emissions. Moreover, the digital, on-demand nature of additive solutions reduces the reliance on large-scale inventories, enabling energy companies to start swapping physical warehouses for digital inventories. Overall, the ability to produce spare parts on demand can mitigate the serious revenue hits and power supply issues caused by downtimes to oil rigs and power plants. This is particularly true in sectors like oil and gas, which typically operate in remote areas and are subject to many supply chain delays. <img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791995531811-1791995531811.jpeg" width="624" height="351" class="fr-fic fr-dii" />3D printer creating prototypes using additive manufacturing. Source: AdobeStock.<h2>AM in Energy Use Cases</h2>The AM industry is seeing a steady adoption increase from the energy sector and a push for more application development and qualification. In a new whitepaper published by Formnext, “<a href="https://formnext.mesago.com/frankfurt/en/industry-insights/whitepaper/whitepaper-energy-oil-gas-industry.html">Additive Manufacturing in the Energy, Oil and Gas Industries: How 3D Printing Improves Energy Supply</a>”, a number of relevant use cases are highlighted. The first is from Norwegian oil &amp; gas company Vår Energi, which uses AM for spare parts production and repairs through its suppliers. With a focus on LPBF, WAAM, DED, and FDM, Vår Energi sees huge potential for AM to reduce costs and lead times for spare parts. Not only that, but the company has also worked with its suppliers to develop new components, like a 3D-printed add-on valve that streamlines operations and fits seamlessly into existing infrastructure.Equinor, a Norway-based multinational energy company, is also among those readily adopting AM. In fact, the company first adopted AM in 2016 and has used the technology for on-site repairs and spare parts production. For example, Equinor replaced an obsolete cooling fan for an electrical motor at its Tjeldbergodden industrial facility with a 3D-printed component. By replacing the part rather than replacing the entire motor and by sourcing it from a local manufacturer, Equinor brought its CO2 emissions down from  4.4 tons to 3.8 kg.[2] Notably, Vår Energi, Equinor, and other oil &amp; gas firms are part of a cross-industry effort to accelerate the development of a vast digital inventory of qualified components for energy systems. This digital inventory is managed by Fieldnode, a digital marketplace for industrial spare parts. <img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791995531815-1791995531815.jpeg" width="624" height="263" class="fr-fic fr-dii" />Wind power and energy farm with wind turbines. Source: AdobeStock.In the wind energy sector, Danish turbine manufacturer Vestas is also embracing AM. To date, the company has built its own digital inventory comprising over 5,600 designs for tooling and spare parts. Not only does the company leverage AM to produce tooling and spare parts on-demand, but it also uses the technology to develop more optimized products, like turbines with optimized power curves and lighter-weight structures. The last case study is from Austria’s main public infrastructure and service provider Wiener Stadtwerke Group, which is leading the adoption of AM for spare parts through Wiener Netze, its central competence center for 3D printing. 3D printing ensures that the company can access “orphaned” components that are critical to the country’s power infrastructure, resulting in cost savings, fewer plant downtimes, and extending the life of legacy systems. <img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791995531848-1791995531848.png" width="624" height="351" class="fr-fic fr-dii" />Source / Copyright: Formnext Mesago Messe Frankfurt. (Bottom left &amp; bottom right: Marc Jacquemin. Bottom center: Mathias Kutt)<h2>Formnext 2026 Industry Highlights</h2>From <a href="https://www.wevolver.com/article/automotive-and-additive-manufacturing-converge-at-formnext-2026">November 17 to 20, Formnext 2026</a> will be highlighting the growing relevance of AM in the energy sector. There, various AM technology companies that work with energy providers will be exhibiting their solutions and highlighting use cases from the field. In addition, the topic will be addressed in the<a href="https://formnext.mesago.com/frankfurt/en/expo/program/event-calendar.html#/?day=2026-11-17&amp;location=industry-application-stage"> stage program</a> through presentations and discussions on current industry developments and applications.<a href="https://formnext.mesago.com/frankfurt/en/expo/visitor-information.html??wt_mc=Display-Banner.ahgz.mf-mesago.formnext.cross-country.vt.cv.wevolver_energy_vp">Full program and tickets can be found here.</a>Outside of the event itself, Formnext organizers have put together resources focused on additive manufacturing in energy, oil &amp; gas. These include the aforementioned whitepaper and <a href="https://formnext.mesago.com/frankfurt/en/industry-insights/energy.html?wt_mc=Display-Banner.ahgz.mf-mesago.formnext.cross-country.vt.cv.wevolver_energy_tp#technology-talk">Formnext Technology Talk</a>, a web special in which industry experts discuss AM’s impact in energy, the state of the segment, and developments on the horizon. Those seeking to go deeper still can also find more use cases and insights on <a href="https://formnext.mesago.com/frankfurt/en/industry-insights.html">Formnext’s Industry Insights page</a>. <hr /><h3>References</h3><ol><li>Global Energy Perspective 2025 [Internet]. McKinsey &amp; Company, October 13, 2025. <a href="https://www.mckinsey.com/industries/energy-and-materials/our-insights/global-energy-perspective">https://www.mckinsey.com/industries/energy-and-materials/our-insights/global-energy-perspective</a> </li><li>3D printing – from science fiction to industrial reality [Internet]. Equinor, May 16, 2023. <a href="https://www.equinor.com/magazine/3d-printing">https://www.equinor.com/magazine/3d-printing</a> </li></ol>

The energy and oil & gas sectors are undergoing transformation and additive manufacturing has a big role to play in this shift.

Formnext 2026 Highlights Additive Manufacturing Growing Impact on Energy Sector

In aerospace and defense, an airframe can stay in service for decades, but what it can do depends on the software built on top of it. That software is now where comparative advantage comes from. Sensing, autonomy, survivability, interoperability, and decision support change at software speed, long after the airframe itself is fielded. Defense programs face pressure to deliver new capabilities faster, while autonomy, sensor processing, and connected mission systems drive adoption of multicore processors and more complex software architectures. Still, the assurance bar does not fall simply because delivery expectations rise.For assurance and verification, <a href="https://www.do178.org/" target="_blank" rel="noopener noreferrer">DO-178C</a>  remains one of the primary standards used by the Federal Aviation Administration (FAA), the European Union Aviation Safety Agency (EASA), and other aviation authorities to approve software-based aerospace systems for flight. Military programs apply the same discipline through MIL-HDBK-516C in the U.S. and EMACC in Europe for airworthiness, with MIL-STD-882E setting how much rigor each software function needs based on its safety criticality. It requires a disciplined lifecycle built around planning, development, verification, configuration management, quality assurance, and certification evidence. Every claim of correctness has to be backed by evidence. When teams move fast, the evidence can fall behind. Legacy development models cannot reconcile speed and assurance by treating verification as a late-stage activity. Instead, verification, traceability, and evidence generation must be integrated into the development lifecycle from the start, making these practices repeatable and sustainable despite rapid changes in mission software.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791986167514-1791986167514.png" width="624" height="416" class="fr-fic fr-dii" />Source: TASKING<h2>Certification-Afterthought Is Now a Clear Program Risk</h2>The airframe is becoming a platform, and capabilities are becoming apps. Autonomy, AI-enabled functions, sensor fusion, electronic warfare, and secure communications are now delivered and updated in software, much like apps on a phone. <a href="https://www.cto.mil/sea/mosa/" target="_blank" rel="noopener noreferrer">Modular Open Systems Approach (MOSA)</a> and <a href="https://aerospacetechreview.com/the-face-approach-a-shining-example-of-mosa/" target="_blank" rel="noopener noreferrer">Future Airborne Capability Environment (FACE)</a>-style approaches aim to improve portability, interoperability, and reuse. However, modularity does not eliminate assurance work. It makes interfaces, configurations, dependencies, and traceability more important, since a reusable component must remain verifiable in the context where it's integrated. Many programs fail because they treat certification as a final gate: capabilities accumulate, integration exposes defects, and teams discover incomplete requirements, insufficient coverage, or missing evidence. At that point, defects become schedule, certification, and mission risk.The remedy is continuous evidence generation. Requirements, source code, test cases, coverage results, and executable behavior should stay connected throughout development. DO-178C requires bidirectional traceability, demonstrating that every requirement is implemented and verified and that no unintended functionality exists. This reshapes/alters the economics of change: when a requirement evolves, teams can quickly identify its impact, rerun the relevant regression tests, and update the associated evidence. That is far more efficient than manually reconstructing traceability and verification evidence from spreadsheets, disconnected tools, and end-of-program documentation.<h2>Multicore Complexity and the Case for an Integrated Toolchain</h2>Multicore processors also introduce assurance problems that cannot be solved by analyzing applications in isolation, because tasks on different cores contend for shared caches, memory, I/O, and interconnects. This contention causes timing interference, as software that behaves correctly alone may run slower or less predictably under concurrent load, which undermines deterministic behavior in real-time, safety-critical systems.<a href="https://www.cast32a.com/" target="_blank" rel="noopener noreferrer">CAST-32A</a>, from the Certification Authorities Software Team, addresses this directly. Shared resources create contention and delay, and therefore applicants must identify interference channels, mitigate their effects, and verify that resource demand does not exceed availability in the intended final configuration. At the same time, they must ensure that applications still execute correctly within their time budgets.Hardware modularity and verification complexity are one continuous problem. The same heterogeneous, reusable multicore platforms enabling flexible mission architectures also demand more rigorous analysis of timing, configuration, data coupling, and execution behavior. The automotive sector has already addressed these challenges. Safety-oriented architectures built around <a href="https://www.tasking.com/products/arm/" target="_blank" rel="noopener noreferrer">Cortex-R</a> and <a href="https://www.infineon.com/products/microcontroller/32-bit-tricore" target="_blank" rel="noopener noreferrer">AURIX</a> families demonstrate the importance of bounded timing, partitioning, traceability, and deterministic behavior under real performance demands. Years of automotive safety-critical multicore engineering have established practices directly applicable to aerospace and defense.The practical solution lies in the toolchain. A disconnected collection of compilers, debuggers, test tools, and requirements repositories creates manual handoffs and evidence gaps. An integrated workflow connects compiler behavior, source code, target execution, test results, structural coverage, timing analysis, and requirements traceability. TASKING addresses this with a <a href="https://www.tasking.com/compile-debug-test/" target="_blank" rel="noopener noreferrer">unified toolchain</a> that keeps compile, debug, test &amp; verification connected in one workflow, rather than scattered across disconnected point tools. That connectivity is what lets teams assess worst-case execution time, identify timing coupling, observe on-target behavior, and repeat analyses as mitigations are applied. In this way, TASKING supports <a href="https://www.tasking.com/do-178/">DO-178C</a> work through a single path from source and build integrity to target execution, coverage, traceability, and certification evidence.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791986167529-1791986167529.png" width="624" height="416" class="fr-fic fr-dii" />Source: TASKING<h2>TASKING's Perspective: Bringing Safety Expertise to Verification at Scale</h2>'Faster development' is an easy promise to make. Aerospace and defense programs need more than that. They need a credible way to scale verification without losing control of determinism, evidence, and certification. TASKING's experience in automotive functional safety and real-time multicore systems is directly relevant here, since automotive programs have long operated under pressure to deliver high-performance software while providing predictable behavior on complex hardware. This experience empowers TASKING's approach to compiler reliability, target-level debug and trace, timing analysis, structural coverage, requirements traceability, and certifiable development workflows.The differentiator is not any single tool, but consistency across the chain of evidence. Specifically, <a href="https://www.tasking.com/content/how-much-can-you-trust-your-compiler-the-case-for-object-code-verification-in-do-178c-level-a-development/" target="_blank" rel="noopener noreferrer">compiler behavior </a>must stay controlled, the relationship between source and object code must be understood, test and coverage results must remain connected to requirements, and timing evidence must account for the final multicore configuration. TASKING's products (i.e., source-to-object-code verification, winIDEA target-level debugging, BlueBox non-intrusive trace, and LDRA testing, coverage, static analysis, and traceability), support these objectives. As systems distribute workloads across cores and reusable modules, preserving that alignment becomes essential to maintaining assurance at scale.<h2>Looking Ahead: Making Continuous Verification Work in Practice </h2>The future of aerospace and defense software assurance lies in continuous verification. Safety and compliance evidence is updated with every controlled software change instead of being rebuilt at the end. Automated workflows can handle regression execution, static-analysis checks, coverage measurement, and traceability updates, while engineers remain responsible for defining correct requirements, interpreting findings, and making safety-critical decisions.As more code is generated by machine, verification matters more, not less. Artificial Intelligence (AI) will play a growing role as verification workloads outpace engineering capacity. It can flag and identify anomalies, prioritize investigations, and automate repetitive analysis. Nevertheless, in mission- and safety-critical development, AI must operate under disciplined human oversight within an independently verifiable process. Tool capability matters only when program teams can implement it effectively. This is where forward-deployed engineering support, working closely with development and certification teams, can establish automated workflows and connect process expectations with day-to-day practice.Aerospace and defense programs need new capability certified and deployed faster, and integrated, continuous verification is what makes that possible. <a href="https://www.tasking.com/products/" target="_blank" rel="noopener noreferrer">TASKING's integrated toolchain</a> offers today a practical path for evolving mission software into defensible evidence.

Bridging the Gap Between Rapid Software Delivery and Rigorous Certification

The Software-Defined Reality of Aerospace & Defense

3D printing can produce un-machinable geometries like lattices

A practical comparison of CNC machining and 3D printing for engineers, covering tolerances, surface finish, materials, geometry, and cost.

CNC vs 3D Printing: How to Choose the Right Process

<h3 id="isPasted">What if the compute your team already paid for is sitting idle while engineers wait in line for access?</h3>That’s happening across engineering organizations. Simulation, AI-assisted engineering, and electronic design automation (EDA) rely on significant computing power, pushing teams to depend on HPC and cloud infrastructure more than ever. As demand grows, queues get longer, and resources remain underused in one environment while engineers wait in another. Hybrid infrastructure only makes this harder to manage.The instinct is to add more compute. But more capacity rarely solves the underlying problem. Throughput depends on matching the right resources to the right workloads, rather than simply adding more capacity. Organizations don’t necessarily need more infrastructure; they need a smarter way to use what they already have.This article examines how intelligent orchestration can improve workload coordination, resource utilization, and efficiency across HPC and cloud environments.<h2>The Real Cost of Reactive HPC and Cloud Operations</h2>Reactive HPC operations carry a cost that rarely appears directly on an infrastructure invoice. Static scheduling rules and manual resource allocation can leave engineers waiting to compute while available capacity remains underused elsewhere. Simulation, AI training, and EDA workloads also place different demands on CPUs, GPUs, memory, storage, and specialized accelerators, making efficient resource allocation more difficult.That mismatch can lead to significant idle capacity. <a href="https://www.jsspp.org/papers21/souza.pdf?" target="_blank" rel="noopener noreferrer">Research on HPC co-scheduling</a> has found substantial gaps between allocated and actual resource use, with some workloads leaving computational resources idle for around 40% of the allocated time. This does not mean every HPC cluster operates at 60% utilization. It shows how resource allocation can differ from actual application demand.Scheduling is not the only source of delay:<ul><li>Jobs can wait for software licenses, compatible hardware, or available resources that meet their requirements.</li><li>When problems occur, administrators may need to manually investigate job status, resource availability, scheduler behavior, or licensing constraints before resolving them.</li></ul>Hybrid environments add another layer of complexity. When on-premises clusters work alongside cloud resources, workload placement must also consider factors such as data locality, network capacity, software licensing, security policies, and cost. The challenge is therefore not only having enough compute, but coordinating available resources effectively across the environment.<h2>How Intelligent Orchestration Solves This: The Core Mechanism</h2>Workload orchestration replaces manual, after-the-fact decisions with continuous coordination across the computing environment. Instead of engineers or administrators manually deciding where a workload should run, an orchestration layer can monitor resource availability, queue conditions, workload requirements, and infrastructure status in real time. It can then help determine where workloads should run based on current conditions.Workload management provides the mechanisms for job submission, scheduling, prioritization, and resource allocation. Intelligent orchestration extends those capabilities by coordinating workloads in real time as environmental conditions – such as fluctuating cloud cost and hardware bottlenecks – change, rather than relying on static submission rules. This distinction is important because the workload manager provides the scheduling foundation, while orchestration uses information about the broader environment to ensure every job lands on the most cost-effective and performant resource available in real time.<h3>The Intelligent Orchestration Workflow</h3>Intelligent orchestration follows a continuous loop and not a one-time setup.<ul><li>Monitor: Keep constant tabs on workloads, resource availability, utilization, and system health across every environment, on-premises and cloud alike.</li><li>Analyze: Determine what each workload needs, from compute type to memory, licenses, and runtime, and compare that to what’s available right now.</li><li>Match: Align workload requirements, such as specific GPU architectures or high-memory nodes, with the most suitable infrastructure, avoiding the "one-size-fits-all" approach of static resource pools. </li><li>Route: Send the job to the right resource and adjust its placement or reroute queued workloads if conditions shift or a better option becomes available.</li><li>Optimize: Feed what happened back into the system so every future decision gets more precise over time.</li></ul>Research demonstrates why this type of adaptive scheduling can be valuable. By understanding job requirements and combining application profiling with real-time cluster monitoring, AI-driven systems are able to learn the most efficient way to schedule workloads. When tested in the real world, researchers reported improvements in cluster utilization of <a href="https://arxiv.org/abs/2401.09706?" target="_blank" rel="noopener noreferrer">up to 51%</a> and reductions in average queue makespan of up to 55% under low-load conditions.These results do not mean every HPC environment will achieve the same gains. They do, however, demonstrate the potential of using workload characteristics and real-time resource information to make more informed scheduling decisions.<h3>The Role of Agentic HPC in Intelligent Orchestration</h3>Agentic HPC represents a further shift from reactive orchestration toward more autonomous decision-making and execution. Rather than relying only on predefined rules or waiting for an administrator to respond to changing conditions, agentic systems can continuously evaluate workload and infrastructure information and recommend or perform actions within defined policies.In an HPC environment, this could include:<ul><li>Identifying idle capacity</li><li>Estimating resource requirements before a job runs</li><li>Detecting potential resource conflicts</li><li>Identifying bottlenecks</li><li>Automatically routing and rebalancing workloads</li></ul>The distinction between automation and autonomy is important. Agentic systems do not necessarily remove human oversight. Engineering and IT teams can continue to define priorities, policies, resource constraints, and approval boundaries, while automated systems handle more of the operational decision-making within those guardrails.Siemens reports that its AI assistant uses historical job data to estimate resource requirements, helping reduce wait times and improve job throughput by <a href="https://www.siemens.com/en-us/products/hpcworks/?" target="_blank" rel="noopener noreferrer">10-15%</a>. Historical job data can therefore contribute to more accurate resource estimates over time. <a href="https://www.siemens.com" target="_blank" rel="noopener noreferrer">Siemens</a> describes its AI assistant as regularly retraining on new data, allowing resource predictions to improve as additional workload information becomes available. The broader direction is clear: AI can help HPC operations move from responding to problems after they occur toward anticipating resource requirements and executing policy-based decisions at scale.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791921394339-1791921394338.jpeg" width="624" height="207" class="fr-fic fr-dii" />Intelligent compute operations. Source: AdobeStock.<h2>The Engineering Impact of Intelligent Compute Operations</h2>The value of intelligent compute operations becomes clear when we measure how quickly engineering teams can move from computation to decisions. Matching workloads to appropriate resources can reduce time spent waiting for compute, while better coordination across environments can help teams make more effective use of available capacity.The gains can compound across development cycles. Faster access to suitable compute can support more simulation iterations and design evaluations, while reduced manual intervention can give engineering and IT teams more time to focus on higher-value work rather than resolving routine infrastructure issues.<div><table><tbody><tr><td> </td><td>Reactive HPC Operations</td><td>Intelligent Orchestration</td><td>Engineering Outcome</td></tr><tr><td> </td><td>Static resource allocation</td><td>Dynamic, real-time workload-to-resource matching</td><td>Faster access to compute, less idle waiting</td></tr><tr><td> </td><td>Long, unpredictable job queues</td><td>Real-time visibility and adaptive routing</td><td>Shorter turnaround and more accurate runtime estimates for simulation and design jobs</td></tr><tr><td> </td><td>Underutilized clusters alongside overloaded ones</td><td>Optimized workload distribution across hybrid environments</td><td>Better return on existing infrastructure investment</td></tr><tr><td> </td><td>Manual troubleshooting after jobs stall</td><td>Proactive detection and automated remediation </td><td>Lower operational overhead, fewer stalled workloads</td></tr><tr><td> </td><td>Siloed on-premises and cloud environments</td><td>Unified orchestration of distributed hybrid resources</td><td>Seamless scaling via cloud bursting when demand spikes</td></tr><tr><td> </td><td>Engineers navigate infrastructure complexity directly</td><td>Autonomous infrastructure management</td><td>More engineering time spent on design, not IT</td></tr></tbody></table></div><h2>How Siemens is Approaching the Future of Engineering Compute</h2>Siemens is helping engineering organizations move beyond reactive compute management by making workload coordination more intelligent across HPC, cloud, and hybrid environments. Engineering teams need access to the compute resources required for simulation, EDA, AI, and other workloads without having to manage every underlying cluster, scheduler, license, or cloud resource themselves.This requires an approach that connects workload management to the infrastructure that supports it. Siemens addresses this through <a href="https://www.siemens.com/en-us/products/hpcworks/?" target="_blank" rel="noopener noreferrer">HPCWorks</a><a href="https://www.siemens.com/en-gb/products/hpcworks/monitor/">™</a> software, a solution portfolio designed to manage HPC and cloud resources across on-premises, cloud, and hybrid environments. As a middleware layer, HPCWorks connects engineering workloads with schedulers, compute resources, cloud environments, licenses, and other infrastructure components. This helps abstract infrastructure complexity from engineers while providing the visibility needed to coordinate workloads and resources across the environment.That foundation supports several capabilities:<ul><li>Routing and prioritization: Directs jobs to the right resources and keeps critical work moving first.</li><li>Policy-driven scheduling: Lets IT and engineering leadership set the rules the platform runs on, rather than deciding on a case-by-case basis.</li><li>Mixed workload orchestration: Coordinates AI and HPC workloads together, managing containerized Kubernetes jobs alongside traditional HPC tasks on shared GPU resources.</li><li>Cloud bursting: Extends eligible workloads into cloud resources when on-premises capacity is constrained, subject to workload, licensing, security, and cost policies.</li></ul><a href="https://resources.sw.siemens.com/en-US/case-study-johnson-and-johnson/?" target="_blank" rel="noopener noreferrer">Johnson &amp; Johnson’s Janssen</a> division used HPCWorks to integrate its HPC environment with AWS. This enabled the division to scale its clusters to three times their previous size, supporting critical research including COVID-19 vaccine development.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791921394347-1791921394347.jpeg" width="624" height="273" class="fr-fic fr-dii" />Computer server room in data center. Source: AdobeStock.<h3>The Shift Toward Autonomous HPC Operations </h3>Siemens is advancing toward fully autonomous HPC operations to eliminate the persistent bottlenecks of reactive environments. These include idle resources, inaccurate resource estimates, scheduling delays, and manual troubleshooting. While current HPCWorks capabilities already include AI-assisted resource estimation, the roadmap points toward a system that doesn’t just assist, but actively manages the computing environment. This transition is already underway. Siemens is targeting a phased rollout of these advanced autonomous capabilities, beginning with a pre-release in late 2026 and a general release planned for 2027. This evolution moves HPC from a reactive service to an adaptive one, where workload demands and infrastructure health continuously drive decisions without requiring manual intervention.<h2>Conclusion</h2>The future of engineering compute is defined by smarter, more adaptive, and increasingly autonomous HPC and cloud operations. As workloads grow more complex, the path to faster innovation isn’t simply adding more capacity; it’s optimizing the infrastructure you already have.This shift makes intelligent orchestration a strategic priority. Better orchestration improves utilization, reduces delays and operational overhead, and helps engineering teams move from compute bottlenecks to faster innovation and time-to-market.For engineering teams ready to move from reactive HPC management to intelligent orchestration, explore <a href="https://www.siemens.com/en-us/products/hpcworks/" target="_blank" rel="noopener noreferrer">HPCWorks</a> to find the right solution for your next compute environment.

Why HPC and Cloud Operations Are Becoming a Strategic Priority

The Future of Engineering Compute

<h2 id="isPasted">Software Is Driving The Conversation</h2>Specialized workloads are changing how compute platforms are designed. Systems built for physical AI must process growing volumes of sensor data locally while meeting strict requirements for latency, power, safety, security, and deterministic response. These demands make the workload and its software requirements an essential input to processor, platform, and silicon decisions.Software can no longer enter the process only after the hardware is complete. A software-first approach begins with the application, representative workloads, response requirements, data movement, memory behavior, safety goals, and system constraints. Hardware and software teams use that evidence to shape the compute architecture before costly implementation decisions are fixed.The design question becomes: Which combination of general-purpose compute, specialized acceleration, memory, interfaces, and software best satisfies this workload’s measured requirements?<h2>What Happens Beneath The Code Matters</h2>Software defines what a system needs to compute. While developers can optimize algorithms and applications, the underlying processor architecture and IP ultimately determine how effectively that software executes. For instance, a workload that relies heavily on matrix multiplication can run on general-purpose hardware, but doing so can result in significantly higher latency and power consumption.  A mismatch between a workload and the selected architecture becomes increasingly expensive to correct as the design moves from architecture definition into RTL, verification, and fabrication. Software teams can mitigate some limitations, but they cannot recover capabilities, data movement efficiency, or acceleration resources that the hardware does not provide. A software-first approach brings workload evidence into earlier decisions and gives developers visibility into how the proposed architecture behaves.<ul><li>Where performance bottlenecks occur</li><li>How workloads map to hardware</li><li>Which functions can benefit from specialized acceleration</li><li>How software decisions affect power and latency</li><li>Where hardware changes could deliver meaningful gains</li></ul>These insights allow developers and silicon designers to optimize the path from software requirements to silicon implementation. In the matrix-multiplication example, workload analysis could show whether the best response is a more suitable CPU configuration, a standards-based NPU, a DSP, or a custom programmable accelerator. That evidence supports an architectural decision before implementation cost and schedule are committed.<h2>From Software Running On Hardware To Software Shaping Hardware</h2>Traditionally, hardware and software have followed a largely linear relationship: Hardware is designed first, and software is developed and optimized to get as much performance as possible from an already defined architecture. When a workload runs inefficiently, developers can only optimize the applications and algorithms to fit the available resources. A software-first approach creates a continuous design loop: workload, measurable requirements, platform architecture, processor and accelerator choices, silicon implementation, and software optimization. Each pass through the loop produces evidence that can refine the next architectural decision.Workload requirements shape software, which in turn informs the architecture and IP design before everything is integrated into silicon. Once an accurate model is available, developers can test workloads against the real architecture and optimize software accordingly. Insights from those workloads can feed back into architectural decisions, creating a continuous feedback loop. The result is an ecosystem where software becomes an input into hardware decisions rather than simply an end product of them. <h2>Open Architectures Change Who Gets To Participate</h2>Open architectures separate the instruction set architecture, or ISA, from the processor implementations built on it. RISC-V combines a standard base instruction set, optional standard extensions, and reserved encoding space for custom instructions. For developers and system architects, meaningful visibility comes from the models, tools, documentation, and performance data that show how software behaves on a specific implementation. That evidence can support standards-based software reuse while informing workload-specific extensions and microarchitecture choices.The value comes from connecting the people who develop software with the people who define hardware. When workload analysis exposes a bottleneck, both teams can evaluate whether software optimization, a different processor configuration, a standard accelerator, or a custom extension offers the best response. Open standards provide a common foundation for that collaboration while preserving room for differentiated implementation.<h2>Why This Matters For Physical AI</h2>Physical AI makes this relationship particularly important. A robot, autonomous vehicle, or industrial system has to sense, process, decide, and act, often under strict power, latency, and safety constraints. A single system may combine signal processing, neural network inference, and real-time control loops, with each workload placing different demands on the underlying compute. Industrial and automotive systems also expect to meet functional safety standards that can demand deterministic behavior. Those requirements can't always be addressed effectively through software optimization alone. Developers can only optimize algorithms and applications so far before the underlying architecture limits them. Even the most efficient software can't reduce power consumption below what the hardware allows or add components the silicon doesn't have. The architecture itself needs to reflect the characteristics of the workloads it's designed to execute. Workload-native computing applies this method at the platform level. MIPS begins with the behavior and constraints of a target workload, then combines processor IP, acceleration, virtual models, software tools, reference silicon, and development hardware into a practical evaluation target. Its three workload-native platforms address distinct needs: <a href="https://mips.com/press-releases/mips-launches-workload-native-platforms-for-physical-ai/">Acies for inference-native AI, Actus for event-native real-time control, and Aegis for safety-native application processing</a>. Each gives developers an earlier environment for software bring-up, workload testing, performance analysis, and ecosystem development.<h2>The Developer Becomes Part Of The Hardware Conversation</h2>The larger shift is cultural as much as technical. Developers have traditionally consumed hardware platforms. They typically worked within whatever architecture they were handed, even though they often have the clearest view of how real workloads and systems behave in practice. In a software-first model, developers contribute measurable workload evidence to architecture and IP decisions. Application behavior can reveal instruction mix, data movement, cache and memory pressure, concurrency, latency, and acceleration opportunities. Hardware and software teams can use those findings to select processor classes, configure subsystems, define custom instructions, and identify the work best assigned to CPUs, DSPs, NPUs, or application-specific accelerators. The same evidence then guides software optimization and the next design iteration.That loop can ultimately produce more differentiated and optimized systems. When the hardware is built around specific workloads, it delivers the performance, efficiency, and specialization that Physical AI computing demands. <h2>Software-To-Silicon Becomes The New Design Loop</h2>Software and hardware now operate as parts of one continuous design process. Workload requirements shape the architecture, measured software behavior informs processor and accelerator choices, and performance observations guide the next generation of platform development.Physical AI systems operate within strict limits for power, latency, safety, security, and availability. They often combine perception, inference, control, and communications workloads on one platform, making the relationship between software requirements and silicon implementation especially important. RISC-V processor IP, configurable acceleration, and pre-silicon virtual models give teams practical ways to evaluate those tradeoffs earlier.Companies that connect workload analysis, software development, processor IP, platform engineering, and silicon implementation can make architecture decisions earlier and validate them against real applications. That capability reduces late discovery of performance and integration problems, preserves more design flexibility, and creates a clearer path from workload requirements to production-ready physical AI systems.<h2>Conclusion</h2>Software-first development gives developers greater visibility into the architecture beneath their code and brings workload requirements into hardware decisions while those decisions remain changeable. Developers can evaluate representative workloads on virtual models, identify performance and efficiency constraints, and work with hardware teams to select or shape the right compute resources. The result is a continuous software-to-silicon design loop built on measured behavior.

Why Developers Need Visibility into the Silicon Beneath Their Code

The Software-First Shift

Body-in-white (BIW) development is one of the most consequential phases of vehicle design, carrying an outsized impact on driving range and total material cost. But despite its importance, BIW comes extremely early in the design cycle, often well before designers have meaningful data to work with.Historically, teams have used relatively slow computer-aided engineering (CAE) to confirm BIW layouts and load-path decisions post-design. But as program timelines shorten and margins for error shrink, engineers need CAE so they can quickly rebuild and rerun at any data-fidelity level.That’s why Siemens built its body development workflows in Simcenter™ Hypermesh™: to put CAE guidance in front of the decision, while the design can still adapt. Moving simulation to the front of a program, Simcenter Hypermesh changes what engineers can commit to and when.<h1><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791909759342-1791909759342.jpeg" width="624" height="437" class="fr-fic fr-dii" />Body engineering challenges. Source: Siemens.</h1><h2>Concept-Phase Decisions and CAE</h2>Engineers begin the BIW concept phase with a wide field of choices, each landing somewhere different on the spectrum of mass and cost. At this point, the goal is to narrow the options to the lightest, cheapest structure that still carries the loads, which ultimately necessitates quick structural feedback on every candidate the team wants to compare. For a broader perspective on how body engineering is evolving into a structural intelligence layer, explore this<a href="https://blogs.sw.siemens.com/simcenter/redefining-the-backbone-how-body-engineering-becomes-the-structural-intelligence-layer-for-faster-more-confident-decisions/"> expert discussion on faster, more confident engineering decisions</a>.Although CAE is the historical solution, its workflows are too slow to keep up. With CAE, analysts need detailed geometry to build a finite element model, but at the concept stage, this detail doesn’t yet exist. Even once that geometry is available, engineers using CAE spend so long preparing and meshing the model that they can realistically evaluate only a handful of combinations.Beyond the speed discrepancy, CAE also creates a chicken-or-the-egg situation between model detail and design data maturity. A fully detailed finite element model requires a proven geometry, but engineers only prove that geometry after making the layout decision the model was supposed to inform. A coarse model, on the other hand, might return an answer too unsubstantiated to justify a platform decision. In most cases, designers will protect themselves by over-designing and adding material wherever the analysis feels uncertain. In turn, those margins tend to push the body past its weight target and cost the vehicle range and efficiency.<h2>The Limits of Traditional Simulation Processes  </h2>One challenge in the current CAE workflow is that teams often model parts that don’t exist yet. An analyst modeling the space around an engine, for example, must wait on geometry from suppliers, who are still quoting the work. In response, the designer will use placeholders and rework their models when the real data arrives. That revision is then only the first of many. Every design change sends them back to the CAD tool to reintegrate, remesh, and rerun the part and its analysis. On an intricate body model, that sequence can stretch a single iteration into days or weeks. And beyond time, detailed finite element analysis (FEA) adds its own computational bill to the project budget.As a result, teams ration variants accordingly. Some engineers run fewer optimization studies than the design warrants, and some skip the optimization phase entirely, forfeiting the mass savings the phase is meant to find.<h1><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791909759350-1791909759350.png" width="624" height="336" class="fr-fic fr-dii" />CAD model. Source: Siemens.</h1><h2>Fidelity That Tracks Design Maturity</h2>The way out starts with matching the model to the moment. Engineers need model accuracy that steps up as design data matures: free-form layout studies while the architecture is still open, and manufacturable joint detail once sections and materials are settled. A process built this way depends only on the inputs available at each stage.The model’s fidelity measures also need to be mixable. Programs mature unevenly, with one section being close to production while another is nearly untouched. The best solution will let designers combine models of different measures in a single study, so they can progress on different subsystems in parallel without waiting for other sections to catch up.<h2>The Ideal Model Iteration Process</h2>Matched fidelity only pays off if engineers can change the model as fast as they change their design. That means building and editing design spaces directly in the simulation environment, without a round trip through CAD for every revision. Designers must also keep the setup manageable as the model grows. To do this, they need a design tool that separates the global design space (i.e., overall body structure) from the local design spaces (i.e., individual components). That way, designers can rework one component without impacting others.Finally, the ideal model iteration process keeps per-iteration cost low. Designers should have access to tools like reduced-order models (ROMs) that strip a detailed structure down, keeping parameter variation cheap enough to repeat dozens of times. Simulation teams should also be able to work the other direction and rebuild detailed geometry from simple inputs such as line models and section profiles.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791909759354-1791909759354.jpeg" width="450" height="219" class="fr-fic fr-dii" />Siemens’s Concept 123 (C123). Source: Siemens.<h2>Simcenter Hypermesh and Concept 123</h2>To bring the ideal workflow to life, Siemens offers Simcenter Hypermesh: a CAE pre- and post-processing environment to build, mesh, and optimize structural models without leaving simulation for CAD.Importantly, Siemens built Simcenter Hypermesh around a three-tier approach called Concept 123 (C123).  <ul><li>C1 identifies the optimal structural layout using free-form models.</li><li>C2 sizes and optimizes idealized sections and joints using low- or mixed-fidelity models.</li><li>C3 optimizes and sizes manufacturable sections and joints using high-fidelity models.</li></ul>And, to address mixability, Siemens designed C123 without a fixed sequence. Teams can start where their data starts and blend models from all three tiers anytime a program matures unevenly. With Simcenter Hypermesh, the simulation process follows the design instead of dictating the order in which engineers work. See how an optimization-led, multi-physics approach helps engineering teams evaluate trade-offs earlier in the<a href="https://webinars.sw.siemens.com/en-US/redefine-automotive-body-engineering-excellence/"> on-demand webinar, “Redefine automotive body engineering excellence”</a>.<h2>Direct Model Building in Simcenter Hypermesh</h2>These capabilities are all a part of the tool itself. In the design space, Simcenter Hypermesh gives engineers tools to directly create and edit regions, so they can set up a topology optimization run and revise it later without reverting to CAD. And, because designers can edit those regions in place, they no longer have to rebuild entire models from scratch for every minor iteration. Global design space setups cover the body's overall structure, while local setups cover individual components such as a B-pillar reinforcement, so reworking that one reinforcement leaves the rest of the body alone.To make the process more affordable, Simcenter Hypermesh offers skeleton modeling: a workflow that simplifies a finished model instead of a design region. With it, engineers can simply point the workflow at a donor CAD or FEA model and get back a ROM  made of the structure's joints, members, and panels. As a result, each run costs so little that teams can freely test a multitude of variants.The tool also includes a CAD-like sketcher that engineers can use to reconstruct skeleton geometries. Simply draw sections and profiles, then grow detailed geometry from that line data. When a ROM study points to a promising section, designers can develop it in place, so the result returns to the design without another trip through CAD.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791909759360-1791909759360.jpeg" width="624" height="415" class="fr-fic fr-dii" />Structural modifications. Source: Siemens.<h2>Designing Before Production</h2>Electrification and software content will keep pushing structural commitments even earlier in the vehicle program, and those commitments will keep changing after teams make them. In that environment, engineers will need simulation that answers quickly and at whatever fidelity the data supports.With Siemens’ Simcenter Hypermesh, engineers finally have a tool that matches their needs. With it, the concept phase stops being a gamble and starts being a place where teams design with evidence and confidence.Learn how an integrated body engineering approach can help teams balance weight, performance, safety, cost, and manufacturability from the earliest development stages in the ebook,<a href="https://resources.sw.siemens.com/en-US/e-book-redefining-body-engineering-excellence/"> “Redefining body engineering excellence”</a>.

Starting with CAE guidance provides engineers with early concept confidence and adaptive designs.

Simulation Moves to the Front of Vehicle Body Development

A struggle in most industrial facilities is that legacy location technologies simply can’t provide high precision in complex indoor environments. For example, operators frequently find that GPS signals can’t penetrate dense warehouse structures or multistory hospital wings. While Bluetooth Low Energy (BLE) can deliver basic proximity metrics, it lacks the fine-grained accuracy needed for high-stakes asset management. Using Qorvo Ultra-Wideband (UWB) technology, engineers can overcome these constraints. UWB technology helps create a connectivity layer that handles both high-speed Wi-Fi data transfer and high-precision location awareness.<h2>Pulse-Based Time of Flight Physics </h2>UWB is a time-of-flight technology, meaning that it measures the exact time it takes short radio pulses to travel between a transmitter and a receiver. A key reason UWB achieves such high accuracy is its 500 MHz signal bandwidth. Because a receiver must sample a signal at least twice its bandwidth, wider bandwidth means finer time resolution, and finer time resolution means finer distance resolution. At 500 MHz, UWB resolves to approximately 15 centimeters, compared to roughly 1 meter for 80 MHz Wi-Fi Fine Timing Measurement (FTM) and several meters for RSSI-based systems. This gives UWB a structural accuracy advantage of five to ten times greater than the next best wireless ranging option.Most time-of-flight technologies suffer from multipath reflections, where signals bounce off surfaces and create interference that typically degrades accuracy in other wireless standards. Qorvo solves this issue with UWB transmitters that emit extremely short 500 MHz-wide pulses centered around the 8 GHz frequency band. The 2 ns duration of these pulses allows sensors to distinguish the direct-path signal from reflected signals, effectively mitigating this issue.And, by calculating the time of flight instead of relying on signal strength as in other location technologies, UWB’s measurements are immune to the fading and fluctuations common in industrial zones. Designers can then use these physical layer properties to maintain a stable location lock even when the line of sight is partially obstructed by environmental clutter. <h2>The Synergy of UWB and Wi-Fi 7 </h2>Using Qorvo-integrated SoCs, network architects can deploy UWB anchors as native components in Wi-Fi 7 access points. This integration allows teams to centralize both UWB ranging and enterprise connectivity within a single hardware footprint, regardless of the transport or backhaul used to carry RTLS data upstream. By centralizing both communication and location services this way, organizations can avoid the capital expenditure of a parallel anchor deployment. Leveraging the existing EAP footprint can significantly reduce the need for additional dedicated localization infrastructure, avoiding the hardware, installation, and cabling costs that standalone UWB anchor deployments typically require—where adding a single dedicated anchor or EAP can cost several thousand dollars in installation and cabling alone.Qorvo's coexistence tools let the UWB system share spectrum with the Wi-Fi 7 radio without packet collisions or spectral crowding, and that same high-speed wireless layer can synchronize clocks across multiple UWB anchors for nanosecond-level timing accuracy. The result is a single unified infrastructure (rather than two networks managed separately) that delivers both simpler deployment and stronger real-time performance.Equally impactful is the expertise Qorvo has in RF filter design. As Wi-Fi 6 GHz and UWB 8 GHz channels occupy adjacent spectrum, high-performance filtering lets both radio chains coexist within the same access point enclosure without degrading each other.<h2>The Standards and Ecosystems for Interoperability </h2>UWB ranging between tags and anchors can be implemented in several ways. Some approaches are fully proprietary to a given vendor, and others are built on open standards that ensure interoperability across devices and platforms. Qorvo aligned its hardware with leading standards bodies to give organizations greater long-term flexibility. Specifically, Qorvo hardware<a href="https://www.firaconsortium.org/about/members/qorvo"> adheres to FiRa Consortium standards</a>, a global organization that provides protocols for engineers to easily build interoperable systems. This compliance matters because it means UWB components from different vendors can communicate and rank distances consistently, freeing organizations from single-vendor lock-in and letting them mix hardware as deployments scale.To protect these interactions, Qorvo provides a distance-bounding protocol that meets IEEE 802.15.4z requirements, preventing sophisticated relay attacks and spoofing. By precisely measuring the distance between devices, the bounding protocol unlocks applications in secure authentication and proximity-based automation that were previously hard to achieve. Beyond FiRA, <a href="https://omlox.com/qorvo" target="_blank" rel="noopener noreferrer">Qorvo works closely with omlox</a>. As a primary standard for industrial locating, omlox provides a vendor-neutral interface that allows engineers to integrate various positioning technologies into a single, cohesive network. Enterprise access point vendors increasingly prefer standardized ranging frameworks like FiRa and omlox—with which Qorvo chips fully comply—as demonstrated by real-world deployments such as Cisco Spaces' FiRa-compliant UWB ranging and Juniper Networks' omlox location services. With greater interoperability and design freedom, organizations can reduce per-unit costs as they scale their location services.<h2>Infrastructure-Free Scaling </h2>Beyond ease of deployment, UWB is compelling for the return on investment that follows when 20 cm accuracy replaces coarser alternatives.First, <a href="https://wifinowglobal.com/news-and-blog/enterprise-wi-fi-integrated-ultra-wideband-uwb-poised-to-power-a-raft-of-innovative-use-cases-qorvo-says/" target="_blank" rel="noopener noreferrer">designers can simply add standard Wi-Fi 7 access points</a> to their existing network infrastructure to significantly expand coverage across locations. That means that with integrated UWB, teams can avoid the labor-intensive process of running new power and data cables, which is typically required for standalone anchors. Standard Wi-Fi planning tools such as Ekahau and Hamina are already integrating UWB ranging into their Wi-Fi installation products, enabling designers to use GPS and UWB data together to simplify access point placement and orientation during installation.With automated placement capabilities, teams can create high-performance network designs that adapt dynamically to changes in the physical environment. Once established, these systems can support tens of thousands of individual tags within a single facility without overwhelming the radio frequency environment. Further, low-power UWB tags can operate for years on battery, which significantly reduces the long-term maintenance requirements for large-scale deployments. With inherent power efficiency, these tags help engineers track massive inventories. Meanwhile, the 20 cm accuracy they provide—offering room-level precision that enables systems to distinguish exactly which side of a wall an asset is on—unlocks asset-recovery, geofencing, and workflow automation use cases that coarser location technologies simply can’t support.<h2>Enterprise Use Cases </h2><h3>Optimizing Smart Factory Workflows </h3>Production managers can use UWB and Wi-Fi to track high-value tools and mobile assets with 20 cm accuracy to facilitate assembly line transitions. To understand the practical value, consider a technician who needs a specific tool at an assembly station but doesn't have it. They check where it should be, ask nearby coworkers, and then visually scan the floor. This process can easily consume 10 to 20 minutes per incident. Using UWB RTLS, the technician can instead query a location app and walk directly to the tool, resolving the search in under a minute. By eliminating this unproductive search time for an entire workforce, manufacturing teams can reduce lead times by up to 50%. Other UWB RTLS benefits include:<ul><li>Optimizing workflows by designing systems that automatically log the movement of parts through every stage of the fabrication process.</li><li>Enhancing individual safety with UWB-enabled handheld devices that provide real-time alerts when workers inadvertently approach a hazard.</li><li>Implementing geofencing protocols that automatically disable machinery if a worker enters a restricted safety zone.</li></ul>In the real world, <a href="https://www.qorvo.com/design-hub/blog/uwb-delivers-a-smoother-ride">SEG Automotive used Qorvo technology</a> to implement these tactics and successfully reallocate 15% of its workforce to higher-value tasks. Ultimately, the unified ecosystems created by Qorvo offered improved operational efficiency and personnel safety.<h3>Advanced Healthcare Logistics </h3>Hospital staff can use integrated RTLS to monitor the real-time locations of medical devices and patients. For example, nurses can use UWB tracking to locate the nearest available emergency equipment in seconds rather than minutes.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791902505638-1791902505637.png" width="624" height="463" class="fr-fic fr-dii" />Qorvo's Ultra-Wideband (UWB) Source: QorvoBeyond equipment tracking, facility designers can use UWB to create comprehensive patient safety systems that monitor movements to prevent falls or unauthorized departures from specialized care units. Similarly, technical teams can use location data to analyze asset movement patterns and identify bottlenecks in patient intake and discharge procedures.<h2>Conclusion </h2>The convergence of UWB and Wi-Fi 7 is a major change in how organizations approach indoor spatial awareness. A key element driving this shift is the support for UWB-based indoor navigation directly on standard smartphones. By combining consumer smartphone adoption (allowing workers or patients to locate tools or offices with standard iOS or Android devices instead of relying on expensive third-party hardware) with integrated infrastructure deployment, organizations can deploy highly precise location systems that are both affordable and easy to scale.As Wi-Fi 7 becomes the standard for enterprise connectivity, Qorvo's integrated UWB approach gives organizations a way to add precise location services (asset tracking and indoor navigation) without deploying separate infrastructure.

Optimizing healthcare logistics and smart factory workflows with Qorvo Ultra-Wideband (UWB) technology.

UWB RTLS and Enterprise Wi-Fi Access Points

Silicon wafer and semiconductor technology underpinning modern data processing and electronic systems.

A semiconductor is a material whose electrical conductivity falls between that of a conductor and an insulator and can be controlled by temperature, light, and impurities. This article covers semiconductor materials, doping, diodes, transistors, and integrated circuits.

What Is a Semiconductor? Band Gap, Doping, Materials, and Devices

<h3>Key Takeaways</h3><ul><li>The surface texture symbol has three forms: plain means any process is permitted, a bar across the “V” means material shall be removed, a circle means material removal is not permitted.[1]</li><li>ISO 1302:2002 has been withdrawn, and ISO 21920-1:2021 is its replacement.[1][2]</li><li>The current symbol carries a bar above it. ISO 21920-1 added that bar specifically to distinguish its symbols from the ISO 1302 ones, so the bar tells you which rulebook applies.[1]</li><li>Newer and older drawings place important information in differing locations around the symbol.</li></ul><h2>Introduction</h2>A surface finish symbol or surface roughness symbol tells the machinist which process is allowed, how rough the surface may be, which way the tool marks should run, and how much stock to leave for a later operation.This guide covers the symbol seen on engineering drawings: the current ISO form, the ISO 1302 form that a large share of drawings in service were drawn to, older styles you will find on legacy drawings, and the North American ASME form. It focuses on reading and writing the callout. For the Ra numbers themselves, see the companion <a href="https://www.wevolver.com/article/surface-finish-chart-ra-rq-and-n-grade-conversions" target="_blank">surface roughness chart</a>, and for how the surface roughness values are captured, see the <a href="https://www.wevolver.com/article/measuring-surface-roughness-a-comprehensive-guide" target="_blank">measuring surface roughness guide</a>.<h2>The Surface Texture Symbol and Its Three Forms</h2><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791407006426-Main+image.png" style="width:100%px" class="fr-fic fr-dib" alt="Surface finish symbols old and new" />The surface finish symbol comes in three main variantsThe basic graphical symbol is two straight lines of unequal length, inclined at approximately 60° to the line representing the surface being specified.[1] It looks like an elongated check mark or square-root sign.Adding a bar across the “V” of the symbol means material removal is required; adding a circle means material removal is not permitted.[1] Those two marks change what the shop is allowed to do.<div><table><tbody><tr><td>Form</td><td>Meaning</td><td>Text abbreviation</td><td>What the shop may do</td></tr><tr><td>Plain symbol</td><td>Any manufacturing process permitted</td><td>APA (any process allowed)</td><td>Machining or non-machining, the shop chooses</td></tr><tr><td>Symbol with a bar across the V</td><td>Material shall be removed</td><td>MRR (material removal required)</td><td>Must machine, grind, hone or otherwise cut the surface</td></tr><tr><td>Symbol with a circle in the V</td><td>Material removal is not permitted</td><td>NMR (no material removed)</td><td>Must leave the as-formed surface: cast, forged, rolled, injection molded</td></tr></tbody></table></div>The three-letter abbreviations exist for written text such as reports and contracts, where you cannot draw the symbol.[1]To carry any complementary requirement, a horizontal line is added to the longer arm of the symbol.[1] A circle over the kink of the complete symbol (where the top right of the “V” meets the horizontal line) means the requirement applies to all surfaces around the workpiece outline shown on that view.[2] But this is no longer used in the latest ISO standard.[1]A surface texture requirement is indicated only once for a given surface, and where possible on the same view that carries the size or location dimensions and their tolerances. The symbol must touch the surface or connect to it by a reference line and leader terminating in an arrowhead, pointing at the surface from outside the material.[1]<h2>What Each Value Around the Symbol Means</h2>This is the part engineers get wrong most often, because the values included on the surface finish callout (and their position) depend on the surface finish standard being used.<h3>ISO 21920-1</h3><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791407061637-New+ISO+example.png" style="width:100%px" class="fr-fic fr-dib" alt="An example ISO 2190-1 surface texture callout" />A callout for Ra 2 roughness with perpendicular lay to be achieved by grinding; material removal is requiredThe current ISO standard for surface finish specifications requires, at a minimum, the graphical symbol described above in one of its three forms, a symbol to indicate the parameter being used (Ra, Rz, etc.), and a numerical value.[1] Everything else is optional and defaults if omitted.Information on the callout can include:[1]<ul><li>Manufacturing process (above the horizontal line, to the left)</li><li>Lay (above the horizontal line, to the right)</li><li>The R-parameter and its specified value (below the horizontal line)</li><li>Various optional modifiers relating to the R-parameter (below the horizontal line)</li><li>Profile direction relative to surface lay (to the right of the “V” of the symbol)</li></ul>For an evaluation-length R-parameter such as Ra, the (optional) elements must be placed in a specific syntactical order, namely: tolerance type, parameter symbol, tolerance limit value, tolerance acceptance rule, S-filter type and nesting index, L-filter type and nesting index, evaluation length, F-operator association method and nesting index, method of profile extraction, an other-requirements placeholder, manufacturing process, surface lay and direction of lay, and profile direction relative to the lay.[1]A minimal indication is just the symbol, the parameter and the limit, and ISO notes that the complete indication “might be rarely used in practice.”[1] Parameters with defined defaults include Rz, Ra, Rp, Rv, Rq, Rmax, Rt, and Pt.[1]Recommended reading: <a href="https://www.wevolver.com/article/surface-finish-standards" target="_blank">Surface Finish: Parameters, Standards, and How to Specify It</a><h3>ISO 1302</h3><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791407228917-Old+ISO+examples.png" style="width:100%px" class="fr-fic fr-dib" alt="Old ISO surface texture symbols" />Three no-longer-current callouts: multidirectional lay, all-surface application, and machining allowance of 3 mm Under ISO 1302:2002, the final version of the old standard, some values are found in different places. And since ISO 1302 was only replaced in 2021, drawings with ISO 1302 are still very common.As with the current ISO standard, the R-parameter and its value is found under the horizontal line and the manufacturing process (if specified) is found above the line.[2]But differences include:[2]<ul><li>Lay (to the right of the “V”)</li><li>Machining allowance (to the left of the “V”)</li><li>Circle symbol indicating that the requirement applies to all surfaces around the workpiece outline shown on that view (where the “V” meets the horizontal line)</li></ul><h2>Historical ISO standards</h2><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791407616438-Historical+ISO+examples.png" style="width:100%px" class="fr-fic fr-dib" alt="Historical surface finish symbols " />Older drawings may put the Ra value or an obsolete N-grade directly above the V of the symbolSurface finishing as a serious engineering practice has existed for almost a century, and common surface symbols entered usage in the 1960s.[3]Drawings outlive standards. On older drawings, the roughness is often given as a roughness grade number rather than an explicit Ra value, a convention that goes back to the 1971 ISO recommendation and its table of grade numbers.[4] Those grade numbers are mapped to Ra values in our <a href="https://www.wevolver.com/article/surface-finish-chart-ra-rq-and-n-grade-conversions" target="_blank">surface roughness chart</a>.The 1971 recommendation that preceded ISO 1302 already used the same basic check mark, the same bar for material removal required and the same circle for material removal not permitted.[4]However, between 1971 and 1992, values were found in different locations around the symbol,[2] notably:<ul><li>The R-parameter and its specified value (directly above the “V”, defined as “position x” in old standards)</li></ul>Recommended reading: <a href="https://www.wevolver.com/article/surface-finish-chart-ra-rq-and-n-grade-conversions" target="_blank">Surface Finish Chart: Ra, Rq, and N-Grade Conversions</a><h3>ASME Y14.36</h3>North American engineering drawings follow the American Society of Mechanical Engineers rather than ISO. The relevant standard is ASME Y14.36M, currently in its 2018 edition, reaffirmed in 2024.[5]Differences between the ISO and ASME standards are fairly minor. Callouts following ASME Y14.36 closely resemble those following the withdrawn ISO 1302.Notable placements include:[5]<ul><li>Lay (to the right of the “V”)</li><li>All-around modifier: Circle symbol indicating that the requirement applies to all surfaces around the workpiece outline shown on that view (where the “V” meets the horizontal line)</li><li>All-over modifier: Variant of the above, two concentric circles indicating that the requirement applies to all surfaces of the part (where the “V” meets the horizontal line)</li></ul>American drawings also commonly work in microinches where ISO drawings work in micrometers; ASME allows either.[5]<h2>Lay Direction Symbols in Full</h2>Lay is the predominant direction of the surface pattern, ordinarily determined by the production method used.[6] It is the direction of the tool marks, and it is specified separately from any roughness value because two surfaces with an identical Ra value can behave completely differently depending on which way the grooves run.That matters most on seals, bearings, and sliding contacts, where the grooves either guide fluid along the joint or across it. On a honed cylinder bore, the crosshatch pattern is specified as lay because the direction of the marks does work that no single roughness value can describe.ISO and ASME standards use equivalent lay symbols. However, the current ISO standard defines the first three according to an intersection plane indicator and datum feature indicator in the top right of the callout, whereas older ISO standards and ASME define them by the plane of projection.[1][5]The mapping below is taken directly from the current ISO standard.[1]<div><table><tbody><tr><td>Symbol</td><td>Meaning</td><td>Typical process</td></tr><tr><td>=</td><td>Parallel to direction</td><td>Shaping, planing, some milling</td></tr><tr><td>⊥</td><td>Perpendicular to direction</td><td>Shaping, planing across the view</td></tr><tr><td>X</td><td>Crossed in two oblique directions relative to the direction</td><td>Honing, crosshatched bores</td></tr><tr><td>M</td><td>Multidirectional</td><td>Lapping, blasting, polishing, superfinishing</td></tr><tr><td>C</td><td>Approximately circular relative to the centre of the surface to which the symbol applies</td><td>Facing and turning on a lathe</td></tr><tr><td>R</td><td>Approximately radial relative to the centre of the surface to which the symbol applies</td><td>Some grinding and facing operations</td></tr><tr><td>P</td><td>Lay is particulate, non-directional, or protuberant</td><td>Powder coating, electro-deposited and sprayed finishes</td></tr></tbody></table></div><h2>ISO Standard Changes</h2>ISO 1302:2002 has been withdrawn and replaced by ISO 21920-1, whose first edition is dated 2021-12.[1] The new standard sits in a three-part series: Part 1 covers indication on drawings, Part 2 covers terms, definitions, and parameters, and Part 3 covers specification operators.[1] Note that ISO 21920-1 is already at stage 90.92, “International Standard to be revised.”[1]Some changes from the old to the new ISO standard have already been discussed, such as the removal of the machining allowance value and the all-surfaces application symbol. Below are three other notable changes in the new standard:<h3>1. Bar above symbol</h3>ISO 21920-1’s graphical symbols carry a horizontal bar above the main symbol: “The bar above the symbol is to distinguish this symbol from the ones used in ISO 1302 and those used for indication of areal surface texture specifications.”[1]The bar is a version stamp. So if the symbol has it, you can read the callout under ISO 21920-1 rules; if it does not, it should be read under ISO 1302 rules or whatever other standard has been specified.The three forms themselves are unchanged in meaning: any manufacturing process permitted, material shall be removed, material removal is not permitted.[1]<h3>2. New default acceptance rule </h3>Under ISO 1302, a limit with no rule stated meant the 16% rule, the ISO 4288 default, which allows at most 16% of measured values to violate the limit.[1][2]Under ISO 21920-1, the default is the maximum tolerance acceptance rule, which “does not allow any measured value to exceed the tolerance limit” and which is “the default case and valid with or without indication of the Tmax symbol.”[1]The same drawing value, unchanged, is harder to pass under the new default. A third option now exists too: the median rule, where the median of all measured values must meet the limit, using at least three measurements.[1]<h3>3. Element set syntax expansion</h3>ISO 21920-1 replaced the “a-to-e” scheme (where five letters were used to specify where values should appear on a callout) with a longer, ordered specification string. For an evaluation-length R-parameter such as Ra, the elements run: tolerance type, parameter symbol, tolerance limit value, tolerance acceptance rule, S-filter type and nesting index, L-filter type and nesting index, evaluation length, F-operator association method and nesting index, method of profile extraction, an other-requirements placeholder, manufacturing process, surface lay and direction of lay, and profile direction relative to the lay.[1]<h2>Conclusion</h2>Surface finish symbols are a compact, positional language, and almost all the difficulty is in the positions rather than the values. Learn the three symbol forms, learn what sits at each position, and learn the seven lay symbols, and you can read any surface texture symbol you meet.Remember that ISO 1302 is withdrawn, ISO 21920-1:2021 replaced it, and the replacement changed both the appearance of the symbol, by adding a bar above it, and the meaning of a bare limit, by making the maximum rule the default.[1][2] However, drawings drawn to ISO 1302 will stay in service for years yet, so you need to read both schemes, and the bar is how you tell them apart.When you write a callout, specify the parameter, the limit, the cutoff, and the lay where it matters, and leave everything else to defaults. When you read one, check the bar first.<h2>Frequently Asked Questions</h2>What are surface finish symbols?They are standardized graphical symbols used on technical drawings to specify the required texture of a surface profile.What does a bar or a circle on the symbol mean?A bar added to the basic symbol means material shall be removed, so the surface must be machined. A circle means material removal is not permitted, so the surface must be left as cast, forged, rolled or molded. A plain symbol means any manufacturing process is permitted.What does 32 surface finish mean?On a North American drawing a bare 32 surface geometry is conventionally read as Ra 32 µin, because American drawings commonly work in microinches.How do you specify surface finish on a drawing?Draw the appropriate symbol form, add the horizontal extension line, then place the parameter designation and its limit value below the extension line.Is ISO 1302 still current?No. ISO lists ISO 1302:2002 as withdrawn at stage 95.99 and names ISO 21920-1:2021 as its replacement.[1] You still need to be able to read ISO 1302 callouts, because many drawings in service predate the change and will not be redrawn.What is the difference between ISO and ASME surface finish symbols?The symbol shape is shared, but value positions and units differ, as described above.What do the lay symbols mean?There are seven: = for parallel, ⊥ for perpendicular, X for crossed, M for multi-directional, C for approximately circular, R for approximately radial, and P for particulate, non-directional or protuberant lay.<h2>References</h2>[1] International Organization for Standardization. <a href="https://www.iso.org/standard/72196.html" target="_blank">ISO 21920-1:2021. Geometrical product specifications (GPS) — Surface texture: Profile — Part 1: Indication of surface texture</a>. Geneva: ISO; 2021.[2] International Organization for Standardization. <a href="https://www.iso.org/standard/28089.html" target="_blank">ISO 1302:2002. Geometrical Product Specifications (GPS) — Indication of surface texture in technical product documentation</a>. Geneva: ISO; 2002. Withdrawn 2021.[3] Nugent P. <a href="https://www.qualitymag.com/articles/87804-50-years-of-quality-a-superficial-history-of-surface-finish" target="_blank">50 Years of Quality: A Superficial History of Surface Finish</a> [Internet]. Quality Magazine. 2011 Apr 1 [cited 2026 Oct 7].[4] International Organization for Standardization. ISO/R 1302:1971. Technical drawings — Method of indicating surface texture on drawings. Geneva: ISO; 1971. Withdrawn 1976.[5] American Society of Mechanical Engineers. <a href="https://www.asme.org/codes-standards/find-codes-standards/y14-36-surface-texture-symbols" target="_blank">ASME Y14.36-2018 (R2024). Surface Texture Symbols</a>. New York: ASME; 2018.[6] American Society of Mechanical Engineers. <a href="https://www.asme.org/codes-standards/find-codes-standards/b46-1-surface-texture/2019/pdf" target="_blank">ASME B46.1-2019. Surface Texture (Surface Roughness, Waviness, and Lay)</a>. New York: ASME; 2019.

Precise surface finishing requires precise notation

Surface finish symbols express a parameter, a limit, a process, and a lay direction in a callout the size of a check mark.

Surface Finish Symbols: How to Read a Drawing Callout

In 2023, researchers in the Soft Transducers Laboratory in EPFL’s School of Engineering presented the world’s first pump in the form of a fiber. The pump’s flexible tubing can generate its own fluid pressure and flow rate, and be sewn into smart textiles to circulate fluids for thermoregulation and haptic feedback, or to power wearable assistive devices.<iframe width="640" height="360" src="https://www.youtube.com/embed/lANIMRoOJeo?&amp;t=39s&amp;wmode=opaque&amp;enablejsapi=1&amp;origin=https://www.wevolver.com" frameborder="0" class="fr-draggable"></iframe> The team has taken their textile-compatible engineering a step further with FiberMotor: a thread-like motor that converts electrical energy directly into motion. Ranging from 1-3 mm in diameter, FiberMotor is composed of two concentric hollow fibers. Tightly coiled around each fiber are insulated copper-wire electrodes only slightly thicker than a human hair. When a voltage is applied to these electrodes, electrostatic forces repeatedly pull the fibers into alignment, causing the inner fiber to slide within the outer one, much like a telescope.“Our FiberMotor is the first fiber-format, sliding motor,” says Soft Transducers Lab head Herbert Shea. “It produces enough force to actuate devices for soft robotic applications, while remaining silent, flexible, and bidirectional.” The work has been published in Advanced Materials.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791376498174-Screenshot+2026-10-07+at+13.34.25.png" style="width:100%px" class="fr-fic fr-dib" />FiberMotor. 2026 EPFL LMTS CC BY SA 4.0Robust, natural movementMany everyday movements like lifting, pushing, and pulling rely on linear (as opposed to rotational) motion. Reproducing this kind of motion has been a longstanding challenge in wearable robotics, because motors powerful enough to drive linear movements often require rigid gears, transmissions, and other bulky components. So-called ‘artificial muscles’ have been developed, but their range of motion is limited by the degree to which their materials can stretch, contract, or bend. Unlike artificial muscles, FiberMotor’s movement range is limited only by the length of the fibers themselves.The absence of gears and transmissions also makes the device backdriveable, meaning it can be safely moved by an opposing force, allowing it to function more naturally with human movements than conventional motors. For example, if a wearer moves unexpectedly or opposite the direction in which the FiberMotor is driven, the device’s fibers simply slide past each other rather than locking up.Strength in numbersIn laboratory demonstrations, individual FiberMotors could support stationary loads equivalent to roughly 75 grams. When bundled together, four motors generated enough force to lift a 46-gram chocolate bar, and flex a tendon-driven robotic finger.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791376521843-Screenshot+2026-10-07+at+13.34.35.png" style="width:100%px" class="fr-fic fr-dib" />Bundled FiberMotors lifting a 46g chocolate bar. 2026 EPFL LMTS CC BY SA 4.0The team also integrated the FiberMotor into a prototype garment to fit the shape of a knee. Because the motors are thin and lightweight, the researchers envision distributing them throughout a smart textile, rather than concentrating force in a single rigid motor unit.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791376558028-Screenshot+2026-10-07+at+13.34.43.png" style="width:100%px" class="fr-fic fr-dib" />The FiberMotor integrated into a prototype garment to fit the shape of a knee. 2026 EPFL LMTS CC BY SA 4.0As a next step, the researchers are increasing the FiberMotor’s performance and durability through even thinner electrodes made mostly of insulators, and optimized materials. Ultimately, they plan to exploit the FiberMotor’s unique combination of force and flexibility to enable wearable assistive devices such as soft exosuits to aid with mobility, wearable haptic systems for virtual reality, and for lightweight prosthetics. First author Sylvain Schaller has already founded a start-up, <a href="https://www.elecsyor.com/" target="_blank" rel="noopener">Elecsyor</a>, to commercialize the technology.Research directions include integrating feedback data about a wearer’s position and movements, so the motor quickly adapt its function to its wearer’s intention.

EPFL engineers have developed a flexible motor just a few millimeters thick. Made of two nested fibers wrapped with electrodes, the device produces enough force to drive soft robotic systems.

Researchers design a motor that looks like a thread

Developing reliable multiple-output power systems is a common challenge across medical, industrial, test and measurement, and semiconductor manufacturing equipment. Engineers must deliver multiple voltage rails, meet stringent safety and EMI requirements, maximize reliability, and bring products to market quickly. Traditionally, these requirements often forced designers to choose between complex multi-supply architectures or costly custom power solutions. The TDK-Lambda MU Series changes that equation by providing a configurable multi-output power platform that simplifies design, reduces compliance risk, and accelerates development.Today, engineers have a better option.Through Sager Electronics’ Certified Value-Add Reseller (VAR) Program, TDK-Lambda’s MU Series can now be assembled, tested, and delivered as application-specific power solutions often in as little as 48 hours. This combination of TDK-Lambda’s proven modular power architecture and Sager’s local configuration capabilities enable customers to accelerate development, simplify compliance, and reduce overall design risk. <img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791316096390-1791316096390.jpeg" width="653" height="503" class="fr-fic fr-dii" />MU Series Products. Source: TDK-Lambda.<h2>Why the MU Series?</h2>The MU Series is a highly configurable AC-DC power platform designed for applications that require multiple voltage rails, high reliability, and compact system integration.Available in 800W and 1500W platforms, the MU Series supports up to 13 outputs from a single power supply chassis. Instead of integrating multiple independent power supplies throughout a system, engineers can consolidate their power architecture into a single compact solution.This approach delivers several important advantages:<ul><li>Simplifies system design and wiring</li><li>Reduces overall footprint and improves power density</li><li>Minimizes leakage-current concerns in medical applications</li><li>Improves long-term reliability by reducing system complexity</li><li>Simplifies inventory management by consolidating multiple power products into one configurable platform</li></ul>For manufacturers of medical imaging systems, diagnostic equipment, semiconductor fabrication tools, automated industrial equipment, and test and measurement systems, these benefits can significantly reduce engineering effort while improving system performance.<h2>Designed for Compliance and Faster Certifications</h2>Regulatory compliance is often one of the largest contributors to development delays.The MU Series helps mitigate this challenge through a platform that is already certified to IEC 60601-1, IEC 62368-1, and IEC 61010-1 safety standards, enabling use across medical, industrial, ICT, test and measurement, and laboratory applications. With Class B EMI compliance, 4kVac input-to-output isolation, and full Means of Patient Protection (2 x MOPP) isolation capability, engineers can enter system validation with greater confidence.By leveraging a prequalified power architecture, design teams can reduce compliance-related design iterations, shorten certification cycles, and accelerate time-to-market.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791316096398-1791316096398.jpeg" width="653" height="467" class="fr-fic fr-dii" />MU4F Products. Source: TDK-Lambda.<h2>Reliability Built for Critical Applications</h2>In high-value systems, power supply failures can lead to costly downtime, missed production targets, lost revenue, and damaged customer relationships.The MU Series was engineered to maximize long-term reliability through:<ul><li>Conservative component derating</li><li>High-efficiency operation up to 94%</li><li>Reduced internal thermal stress</li><li>Published conservative operating specifications</li><li>Industry-leading seven-year warranty</li></ul>The result is a power platform designed to support continuous operation in demanding environments where reliability is critical.For enhanced operational visibility, integrated PMBus capability enables monitoring of key system health parameters such as input voltage, temperature, fan status, runtime, power cycle count, and fault conditions, while supporting remote diagnostics and predictive maintenance strategies that help minimize unplanned outages and maximize uptime.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791316096414-1791316096414.jpeg" width="653" height="365" class="fr-fic fr-dii" />Medical research laboratory. Source: AdobeStock.<h2>Quiet Operation for Medical and Laboratory Environments</h2>Many medical, laboratory, and precision measurement systems require low acoustic noise in addition to reliable power delivery.The MU Series incorporates TDK-Lambda’s patented Intelligent Fan Control (IFC) technology, which dynamically adjusts cooling performance based on operating conditions. This helps reduce audible noise, delivering sound levels as low as 36 dBA while maintaining optimal thermal performance and extending component life.The result is a quieter power solution without sacrificing reliability or efficiency.<h2>Faster Custom Configurations with Sager’s VAR Program</h2>While the MU Series provides a flexible power platform, Sager’s VAR Program unlocks its full potential.As a factory-trained and authorized TDK-Lambda VAR partner, Sager can configure, assemble, and test MU Series power supplies to customer-specific voltage requirements. By maintaining local inventory of MU chassis and output modules in its Lewisville, Texas Custom Solutions and Distribution Center, Sager can rapidly provide configured solutions without the delays traditionally associated with custom power designs.In addition to speed and flexibility, Sager’s VAR Program provides a high level of manufacturing traceability and process control. Each MU Series chassis and output module is individually scanned during assembly, helping ensure the correct components are used while creating a complete record throughout the product’s journey. The same assembly processes, controls, and quality standards used by TDK-Lambda are implemented at Sager’s Custom Solutions Center, helping deliver consistent quality and reliability whether the product is assembled in the United Kingdom or Texas. This traceability also provides valuable documentation and accountability for customers operating in highly regulated medical, industrial, test and measurement, and semiconductor manufacturing environments.<h3>Benefits to Customers</h3>Benefits to customers include:<ul><li>Application-specific voltage configurations without the cost and delays of a custom power design</li><li>Fully assembled, validated, and ready-to-install power solutions</li><li>Rapid delivery to support prototyping, design verification, and production ramp-up</li><li>Complete component and assembly traceability for regulated applications</li><li>Simplified power architecture and reduced inventory management complexity</li><li>Flexible reconfiguration to accommodate evolving system requirements</li><li>U.S.-based technical support from experienced power specialists</li></ul>Every configured unit undergoes rigorous validation, including visual inspection, HiPot testing, and automated electrical verification to help ensure quality and performance.<h3>The Combined Advantage</h3>Together, TDK-Lambda and Sager deliver:<ul><li>A proven configurable power platform from a global power supply leader</li><li>Pre-certified designs that help streamline compliance and system validation</li><li>Local assembly using factory-trained personnel and established quality processes</li><li>Consistent performance and quality backed by rigorous testing and inspection</li><li>Rapid configuration and sample availability to accelerate project timelines</li><li>End-to-end traceability and accountability throughout the assembly process</li><li>A scalable path from prototype through to full production</li></ul>Instead of choosing between the speed of an off-the-shelf power supply and the flexibility of a custom design, engineers can now achieve both.<h2 id="isPasted">Bringing Custom Power Within Reach</h2>The addition of the MU Series to Sager’s Certified VAR Program represents a significant opportunity for the North American market. By combining TDK-Lambda’s proven configurable power architecture with Sager’s rapid assembly, testing, and traceability capabilities, customers can simplify complex power designs, accelerate product development, and reduce overall project risk. Equally important, the MU Series expands the Sager VAR portfolio with a highly robust, configurable power platform designed for long service life and mission-critical applications.Whether developing next-generation medical equipment, factory automation systems, semiconductor tools, or precision instrumentation, the MU Series and Sager’s VAR Program deliver a faster, more reliable path from concept to production.Ready to configure your ideal power solution? Use the joint TDK-Lambda and Sager Quick Product Finder (QPF) to build your MU Series configuration, generate a unique part number, and request a sample in just a few clicks. Accelerate evaluation, simplify specification, and get a customized solution into testing faster.<a href="https://urldefense.com/v3/__https:/config.emea.tdk-lambda.com/QPF/s_configureBeginAuto*productfinder__;Iw!!FtrhtPsWDhZ6tw!DeQthZmzHJ3ASJUEljHoEkphtProZZP1k_3HkVBbDybeLKJd9aen_VSAgsl_XG7tYDZPOH9SOlhWsQabjLA$" target="_blank" rel="noopener">Configure your MU now</a>Learn more about <a href="https://www.sager.com/manufacturers/tdk-lambda" target="_blank" rel="noopener">TDK-Lambda products</a>, or contact <a href="https://sager.formstack.com/forms/tdk_var_mu_series" target="_blank" rel="noopener">Sager’s technical team</a> for application-specific support.<a href="http://www.sager.com" target="_blank" rel="noopener">sager.com</a>

The TDK-Lambda MU Series with the Sager Electronics VAR Program is a configurable, certification-ready power platform with rapid U.S.-based assembly, testing, and traceability.

Accelerating Complex Power Designs

<h2 id="isPasted">Why So Many AI Initiatives Stall</h2>Industrial companies are experimenting extensively with artificial intelligence. Teams are testing copilots, developing proofs of concept, and introducing AI into individual workflows. Yet a successful experiment does not automatically lead to AI that works across an entire industrial organization.The gap becomes clearer when AI moves beyond general productivity tasks. A copilot may summarize documents, answer questions, or help employees search for information. Industrial work, however, depends on relationships among products, components, machines, engineering requirements, manufacturing processes, suppliers, quality records, and business systems.According to Siemens, organizations can struggle to move beyond isolated AI pilots when enterprise data remains fragmented, governance varies across the organization, and AI-generated insights are disconnected from operational workflows.[1]This makes scaling industrial AI a broader problem than simply choosing a capable model. Companies also need industrial context, architecture that can support wider adoption, appropriate governance, and tools that domain experts can use. This is the difference between checking the AI box and building an environment in which AI can become part of everyday industrial work.<h2>The Pilot Trap: AI Without Industrial Context</h2><h3>A Copilot Can Answer Questions, But Does It Understand the Business?</h3>General-purpose AI can improve personal productivity without understanding the structure of an industrial enterprise. It can summarize reports, draft responses, explain documents, or retrieve information.Industrial questions often demand more. For example, a component may appear in several enterprise systems. Product lifecycle management (PLM) may contain its engineering definition. Enterprise resource planning (ERP) may connect it with procurement and cost information. A manufacturing execution system (MES)  may contain information about how it is produced.The underlying object may be the same, but each system can describe it differently. Siemens illustrates the problem with a simple example. The same physical component may appear as a part in PLM, a material in ERP, and a component in MES. If those systems remain disconnected, AI can see the records but not the full business context behind them. [2]The issue is therefore not simply how much data an AI system can access. It is whether that data carries enough structure and context for the system to recognize how information across systems is related. Scaling industrial AI requires connected industrial context, not simply more disconnected data.<h2>Context Is the Foundation for Enterprise-Scale AI</h2><h3>Connecting Data Through Relationships</h3>An industrial context layer provides a way to connect information that already exists across enterprise systems. Instead of treating each record as an isolated piece of information, a context layer represents relationships among business entities. A product can be connected to its components. Those components can be connected to suppliers, quality issues, production batches, customer orders, or engineering changes.Knowledge graphs provide one way to represent these relationships. Siemens describes an AI context layer as a semantic layer positioned above existing enterprise data platforms. It connects information from systems such as ERP, PLM, MES, CRM, quality, and supply chain platforms, enabling AI applications to operate across domains rather than treating each system as a separate source. [3]This matters because many industrial questions depend on relationships rather than individual records. A quality team may need to determine which customer orders are linked to a failed production batch. An engineer may need to understand the downstream effect of an engineering change. A supply chain team may want to identify products linked to suppliers with unresolved quality issues.The answer may not exist in one document or database table. It can emerge from relationships among information distributed across several systems. RapidMiner Graph Studio is Siemens' implementation of this approach. Siemens describes it as creating a common ontology across systems such as PLM, ERP, CRM, and MES, providing a semantic layer through which relationships can be queried. [4]Resource Description Framework (RDF) is one of the standards associated with this architecture. The World Wide Web Consortium defines an RDF graph as a set of subject, predicate, and object triples. In practical terms, this provides a standardized way to describe resources and their relationships. [5]The important point is not the terminology itself. AI becomes more useful for industrial work when it can understand not only individual pieces of information, but also how those pieces connect.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1791309695193-1791309695193.jpeg" width="602" height="289" class="fr-fic fr-dii" />Concept for digital data and AI architecture. Source: AdobeStock.<h2>Scaling the Architecture, Not Just the Use Case</h2><h3>From One Successful Model to Continuous Enterprise Workloads</h3>A proof of concept usually has narrow boundaries. It may serve one team, use a limited dataset, or address a single workflow. Enterprise deployment changes those conditions. More people use the system. More data sources become involved. Questions extend across departments, and AI applications may need to follow multiple relationships before producing a useful result.Siemens distinguishes between smaller graph-based implementations and enterprise context graphs that must operate across domains, accommodate changing ontologies, and manage larger volumes of connected information. [3]The question therefore shifts from whether an AI model can produce a useful answer to whether the organization can operate that capability reliably as adoption expands. Thus, architecture becomes part of the AI strategy.As AI adoption expands, the supporting architecture must be able to handle growing data relationships, workloads, and usage without introducing new bottlenecks. The focus therefore shifts from proving that a model works to determine whether the organization can operate, govern, and reuse it reliably at scale.That does not mean every industrial organization needs the same technical design. It does illustrate a broader requirement: infrastructure that supports a pilot may not be sufficient when AI becomes part of recurring enterprise workflows. Organizations need to ask not only, "Can this model work?" but also, "Can we run, maintain, govern, and reuse it across the business?"<h2>Governance Has to Grow With AI</h2><h3>Scale Without Control Creates Another Problem</h3>Governance becomes more important as AI moves from experimentation into operational workflows. A small pilot may involve a limited group of users and a narrow dataset. Enterprise AI can interact with many systems, models, workflows, and employees. Its outputs may also influence decisions affecting production, quality, engineering, suppliers, or customers.Organizations therefore need visibility into how AI and enterprise data are being used. Siemens describes the Intelligence Center X as a governed environment that connects enterprise data, models, workflows, applications, and AI agents. Traceability, auditability, policy controls, and governance are presented as parts of that architecture. [1, 2]These controls support practical questions. Which data did an AI application use? Who had permission to access it? Can an AI-assisted decision be traced? Who remains responsible for the final action?Governance also extends beyond any one technology provider. The National Institute of Standards and Technology developed the Artificial Intelligence Risk Management Framework, or AI RMF, as a voluntary, non-sector-specific resource for organizations that design, develop, deploy, or use AI systems. Its purpose is to help organizations manage AI-related risks while supporting the trustworthy and responsible use of AI. [6]For industrial enterprises, this reinforces an important point: governance should not be added only after AI has already spread across the organization. Risk management, access control, traceability, oversight, and accountability need to develop alongside the technical architecture.<h2>Enterprise AI Cannot Belong Only to IT</h2><h3>Put AI in the Hands of the People Who Understand the Problem</h3>Technology teams play an important role in enterprise AI, but they are not the only people who understand where AI can create value. Engineers, manufacturing specialists, quality teams, supply-chain professionals, and other subject-matter experts often have the deepest knowledge of the operational problems being addressed. They understand how processes work, which information matters, and where an AI-generated answer requires additional judgment.If every AI use case requires a long custom development project, adoption becomes difficult to scale. Low-code tools can reduce part of that barrier by enabling domain specialists to participate more directly in creating applications and workflows without requiring them to become professional software developers.Siemens combines Mendix low-code application development with Graph Studio and RapidMiner AI capabilities within Intelligence Center X. According to Siemens, this brings enterprise context, AI modeling, application development, orchestration, and governance into a broader shared environment. [1, 2]This does not remove the need for developers, data scientists, security teams, or IT governance. Instead, low-code tools can shorten the distance between domain knowledge and application development.That matters because industrial AI becomes harder to scale when technical expertise sits in one group while knowledge of the operational problem is elsewhere. AI becomes more practical when the people closest to the problem can help shape the tools they use.<h2>Free Time for Your Experts</h2><h3>The Goal Is Not More AI, It Is More Valuable Human Work</h3>The purpose of industrial AI should not be measured simply by the number of models, copilots, or agents an organization deploys. Industrial specialists spend part of their working day searching for information, moving between systems, connecting records, repeating analyses, and coordinating routine workflows. Some of those activities require expert judgment. Others consume expert time because information is fragmented or difficult to access.Context-aware AI can assist with some of that supporting work. When information from engineering, manufacturing, quality, supply chain, and enterprise systems is integrated, AI applications have a stronger foundation for retrieving relevant facts, following relationships, and supporting repetitive analytical tasks.Siemens provides examples of customers who use connected applications and AI-supported workflows to reduce manual work and improve issue-resolution processes. [1] These examples should be treated as specific implementations rather than expected results for every organization, but they show the type of work enterprise AI can support.The objective is not to remove industrial expertise from decision-making. Engineers and other subject-matter experts continue to provide technical knowledge, judgment, and accountability.A more useful goal is to reduce the time experts spend navigating fragmented information and routine workflows. That is the practical meaning of "free time for your experts." Enterprise AI becomes valuable when skilled people can spend more time solving engineering and business problems and less time finding and connecting information.<h2>Scaling AI Means Building for Industrial Reality</h2>Enterprise AI does not scale simply by launching more pilots. Industrial environments contain complex relationships among products, engineering information, manufacturing systems, suppliers, quality records, business processes, and people. AI applications that work with disconnected pieces of that environment may still provide value, but their ability to support broader industrial work remains limited.Moving beyond experimentation requires several capabilities to develop together. AI needs context that connects information across systems. Architecture must support expanding workloads and changing relationships. Governance needs to provide traceability, access control, oversight, and accountability. Domain experts need practical ways to participate in the applications and workflows being created.Siemens Intelligence Center X provides one example of how knowledge graphs, industrial ontologies, AI modeling, low-code application development, orchestration, and governance can be brought together. [1, 2, 4] It should be viewed as one implementation of these broader requirements, not as evidence that one platform solves every AI scaling challenge.For industrial companies, the more important question is whether their AI initiatives reflect the reality of industrial work: how data relates, how systems connect, how decisions are governed, and how experts use information.A successful pilot can show that an idea works. Scaling AI across the industrial enterprise requires creating the conditions for it to keep working as part of everyday operations.<hr /><h2>References</h2><ol><li>Siemens Intelligence Center X <a href="https://news.siemens.com/en-us/siemens-intelligence-center-x/" target="_blank" rel="noopener noreferrer">https://news.siemens.com/en-us/siemens-intelligence-center-x/</a></li><li>Siemens Rapidminer The AI Context Layer: Why Your AI Agents Need a Knowledge Graph to Think <a href="https://blogs.sw.siemens.com/rapidminer/ai-context-layer-knowledge-graph/" target="_blank" rel="noopener noreferrer">https://blogs.sw.siemens.com/rapidminer/ai-context-layer-knowledge-graph/</a></li><li>Siemens Rapidminer  The Brain in a Jar Problem: Why AI Without Context Is Failing the Enterprise <a href="https://blogs.sw.siemens.com/rapidminer/the-brain-in-a-jar-problem-why-ai-without-context-is-failing-the-enterprise/" target="_blank" rel="noopener noreferrer">https://blogs.sw.siemens.com/rapidminer/the-brain-in-a-jar-problem-why-ai-without-context-is-failing-the-enterprise/</a></li><li>Siemens Rapidminer Graph Studio <a href="https://www.siemens.com/en-us/products/rapidminer/graph-studio/" target="_blank" rel="noopener noreferrer">https://www.siemens.com/en-us/products/rapidminer/graph-studio/</a></li><li>W3C RDF 1.2 Concepts and Abstract Data Model <a href="https://www.w3.org/TR/rdf12-concepts/" target="_blank" rel="noopener noreferrer">https://www.w3.org/TR/rdf12-concepts/</a></li><li>NIST Artificial Intelligence Risk Management Framework (AI RMF 1.0) <a href="https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10" target="_blank" rel="noopener noreferrer">https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10</a></li></ol>

What it takes to scale AI across the industrial enterprise.

Beyond the AI Pilot

Wearable devices are expected to do more with less. Smaller form factors, longer battery life, and continuous sensing place tight limits on compute, energy, and system design. At the same time, expectations around intelligence are increasing, as devices are expected to track data, interpret signals, detect anomalies, and adapt to individual users in real time.This creates a fundamental engineering constraint. Advanced AI workloads require significant computational resources, which directly translate into power consumption. In compact, battery-powered systems, this trade-off has historically limited the level of intelligence that wearable devices can deliver.BrainChip’s AkidaTag© Reference Platform addresses this constraint directly. It demonstrates how always-on intelligence can run locally, continuously, and efficiently within strict power budgets. The result is a new class of wearable systems that operate for days or weeks while executing real-time AI workloads on-device.<h2>The Power Behind Wearable AI</h2>Wearable systems operate under tight energy limits, where battery capacity is constrained by size, weight, and user comfort. Even small increases in power consumption can reduce usable time from days to hours.Traditional AI architectures are not optimized for this environment. CPUs and GPUs process data continuously, regardless of signal relevance, which leads to unnecessary energy use in applications that require persistent monitoring.To compensate, developers often simplify models or reduce sampling rates. While this lowers power consumption, it also degrades performance. In applications such as health monitoring or anomaly detection, reduced fidelity can limit usefulness.Cloud-based processing introduces an alternative approach, but it adds latency, dependency on connectivity, and privacy concerns. For many wearable use cases that require immediate response or handle sensitive data, relying on remote infrastructure is not practical.Running meaningful AI continuously within the power limits of compact, battery-powered wearables remains a core engineering constraint.<h2>Event-Based Intelligence for Continuous Operation</h2>BrainChip’s Akida architecture approaches this problem differently. It is designed to process temporal data efficiently using state-space models. These models capture how signals evolve over time, allowing the system to focus on meaningful changes while maintaining continuous awareness of the signal context.In wearable applications, this translates into significant efficiency gains. Periods of inactivity consume minimal power, and processing scales with signal relevance rather than time.The AkidaTag© Reference Platform translates this architecture into a complete, deployable system that combines hardware, software, and models into a validated blueprint for evaluation and development.<h2>A Reference Platform Built for the Real-World</h2>The AkidaTag is a compact, battery-powered wearable platform designed to demonstrate always-on AI in practice. It integrates BrainChip’s AKD1500 neuromorphic processor with a Nordic Bluetooth microcontroller, along with sensing components such as a microphone and accelerometer.This configuration supports continuous data acquisition and local inference within a tightly controlled power envelope. Typical power consumption during inference remains below 50 milliwatts, enabling operation over extended periods without frequent recharging.Latency is also optimized for real-time interaction, with keyword spotting tasks completing in under 50 milliseconds and anomaly detection tasks below 100 milliseconds. These response times support immediate feedback in user-facing applications.Rather than serving as a single-purpose device, the platform provides a development foundation that demonstrates how multiple AI workloads can coexist within the same system, all operating locally and continuously.<h2>Anomaly Detection and Adaptive Learning</h2>Anomaly detection is a core capability of the platform, particularly in wearable systems that monitor dynamic signals such as motion, sound, or physiological activity.Instead of relying on predefined thresholds, the system learns baseline patterns and identifies deviations, which help detect subtle changes that may indicate emerging conditions or events.On-device learning allows the model to adapt incrementally as new data is observed without requiring full retraining or cloud-based updates, enabling continuous personalization directly within the device.This becomes especially relevant in health monitoring, where baseline behavior varies significantly between individuals. As the system adapts over time, it improves prediction accuracy and reduces false alerts, while keeping all learning and inference local and independent of external connectivity.<h2>Extending Battery Life Without Reducing Capability</h2>Battery life remains a primary focus in wearable system design, where limited capacity must support continuous sensing and real-time AI processing without frequent recharging. This places sustained pressure on system efficiency.The AkidaTag platform addresses this by reducing the amount of computation required during operation and minimizing energy per processing step, with energy per operation measured in the picojoule range. Combined with its temporal processing approach, this enables multi-day or multi-week operation depending on usage conditions.This level of endurance supports continuous monitoring in real-world scenarios, where devices can operate persistently in the background while delivering timely insights without requiring user intervention.<h2>Applications for Always-On Wearables</h2>The reference platform supports a range of wearable applications that require continuous sensing and real-time, on-device inference within strict power limits, and can be adapted across industrial, consumer, and healthcare environments.In industrial settings, anomaly detection can be applied to vibration data for continuous equipment monitoring, enabling early identification of irregular patterns that indicate wear or failure, while low power consumption supports deployment across distributed sensor networks.In consumer devices, keyword spotting enables always-listening voice interfaces that operate locally, ensuring low-latency response and preserving user privacy, with processing triggered only by relevant audio patterns.In health monitoring applications, the platform supports real-time analysis of physiological and behavioral data, where continuous operation allows detection of patterns not visible in intermittent measurements, and on-device learning enables adaptation to individual users over time.Across these domains, systems must sustain continuous data acquisition, deliver immediate inference, and operate within limited energy budgets, all of which are addressed within a single integrated platform.<h2>Reducing Development Complexity</h2>Developing wearable AI systems requires alignment across integration, optimization, and validation, particularly in areas such as power management, model deployment, and system-level performance.The AkidaTag Reference Platform provides a validated foundation that integrates hardware, software, and AI models into a unified system, allowing developers to evaluate performance under realistic conditions without building from scratch, while supporting adaptation to product-specific requirements through a broader ecosystem.In both consumer and industrial contexts, this approach shortens development cycles, supports faster iteration, and reduces the risks associated with deploying always-on AI in constrained environments.<h2>Explore the Akida Wearables Platform</h2>The AkidaTag Reference Platform provides a practical foundation for evaluating and developing always-on wearable AI under real-world scenarios, combining low-power operation, on-device learning, and integrated system design in a deployable form factor.Developers and system architects can explore the platform to assess performance, accelerate prototyping, and adapt the reference design to specific application requirements across health, industrial, and consumer domains.For more information, visit: <a href="https://brainchip.com/akida-tag-lp/">https://brainchip.com/akida-tag-lp/</a>

Generics

Latest Posts