Content Creation

Robotics Software

Explore IoT security solutions with insights into the threat landscape, secure architectures, and best practices for engineers building resilient connected systems.

IoT Security Solutions: Threat Landscape, Architecture, and Engineering Best Practices

Practical and architectural insights into IoT gateways, including edge versus cloud placement, protocol translation, data aggregation, hardware designs, and deployment patterns for industrial IoT.

IoT Gateway Architecture: Edge vs. Cloud, Protocol Translation, and Deployment Patterns

This article explores TPU vs GPU differences in architecture, performance, energy efficiency, cost, and practical implementation, helping engineers and designers choose the right accelerator for AI workloads today!

TPU vs GPU: A Comprehensive Technical Comparison

TPU vs GPU: Comprehensive Technical Comparison

AI Infrastructure Engineering: Building Enterprise-Scale Machine Learning Systems

This article covers advanced architectural patterns, performance optimisation strategies, and critical operational considerations for building production-grade AI infrastructure systems!

Unexpected shutdowns? Costly errors? SCADA systems are the solution industries rely on to stay in control. Let's understand how this system allows owners to run industries through PLCs, RTUs, and other field devices, allowing them to make smarter decisions in real-time

What is SCADA?: The Basics you Need to Know

Specialized hardware accelerators like Neural Processing Units (NPUs) and Tensor Processing Units (TPUs) have revolutionized artificial intelligence (AI) hardware by providing capabilities for accelerating machine learning tasks. Both are designed to optimize performance for tasks like machine learning, deep learning, and neural network computations, but they cater to distinct use cases and requirements.&nbsp;This article discusses the difference in architectures, performance, practical applications, and challenges of these powerful processing units, offering actionable insights for professionals.Recommended reading:&nbsp;<a href="https://www.wevolver.com/article/tpu-vs-gpu-in-ai-a-comprehensive-guide-to-their-roles-and-impact-on-artificial-intelligence" target="_blank">TPU vs GPU in AI: A Comprehensive Guide to Their Roles and Impact on Artificial Intelligence</a><h2 dir="ltr">Introduction</h2>The rapid advancement of AI has led to an increasing demand for specialized hardware capable of efficiently handling complex computations. NPUs and TPUs have emerged as pivotal technologies in this space, each tailored to optimize tasks involving complex machine learning algorithms in unique ways. NPUs, with their low-power design, enhance edge computing performance, making them indispensable for IoT and mobile devices. TPUs, on the other hand, excel in scaling deep learning workloads across massive datasets, driving breakthroughs in cloud-based AI.This article aims to provide professionals with a detailed comparison of these technologies, offering practical insights into their architectures, applications, and limitations. Understanding these distinctions can guide informed decision-making and optimize AI deployments for diverse requirements.<h2 dir="ltr">What Are Neural Processing Units (NPUs)?</h2>NPUs are specialized processors designed to optimize computations for neural networks, enabling efficient processing in real-time. With their ability to efficiently handle workloads like image recognition, voice processing, and natural language tasks, NPUs are an integral part of modern central processing units (CPUs) and System-on-Chip (SoC) designs. Companies like Huawei, Samsung, and Apple are leading the way in integrating NPUs into their devices to enhance AI capabilities without sacrificing power efficiency.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNzM3NTY2MzI0NTQ0LWltYWdlMi5wbmciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 100%px;" class="fr-fic fr-dib" alt="apple-neural-engine-devices">Apple's NPU, branded as the Neural Engine, is a key component of Apple's A-series and M-series chips. It is designed specifically for accelerating on-device machine learning and AI tasksRecommended reading:&nbsp;<a href="https://www.wevolver.com/article/npu-vs-gpu-understanding-the-key-differences-and-use-cases" target="_blank">NPU vs GPU: Understanding the Key Differences and Use Cases</a><h2 dir="ltr">What Are Tensor Processing Units (TPUs)?</h2>TPUs are hardware AI accelerator application-specific integrated circuits (ASICs) developed by Google to accelerate tensor-based computations, which are foundational to deep learning. Their design is optimized for mathematical operations like matrix-multiplication and parallel computing, making them ideal for tasks such as training large-scale neural networks and performing batch processing.Here is a block diagram showcasing the architecture of TPUs:<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNzM3NTY2Mzc3NTczLWltYWdlNS5qcGciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 100%px;" class="fr-fic fr-dib" alt="TPU-architecture-simplified-block-diagram">A simplified block diagram of a TPUThe TPU architecture is designed to optimize AI workloads by integrating specialized components for efficient computation, storage, and communication. At the core is the Matrix Multiply Unit, which performs rapid large-scale matrix operations critical for AI tasks. Data flows through the Unified Buffer, a local storage unit for managing activations, ensuring quick access and reducing dependency on external memory. The Control Unit orchestrates data movement and synchronizes operations between components. The PCIe Interface and Host Interface facilitate communication between the TPU and external systems, enabling smooth data transfer and coordination. The DRAM Ports provide access to external memory for managing large datasets that exceed the buffer's capacity. The Activation Pipeline processes intermediate results, applying transformations essential for refining AI computations. Additionally, the Accumulator stores partial computation results temporarily for aggregation and further processing. Finally, the Miscellaneous I/O manages input/output tasks, ensuring seamless integration with the broader system. This balanced architecture ensures high efficiency and performance for AI tasks in data centers.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNzM3NTY2MzkwMzg5LWltYWdlNC5wbmciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 100%px;" class="fr-fic fr-dib" alt="Google-TPU-v4">Google TPUs are primarily used in servers within data centers to accelerate large-scale AI workloads, such as training and deploying deep learning models.Recommended reading:&nbsp;<a href="https://www.wevolver.com/article/tensor-cores-vs-cuda-cores" target="_blank">Tensor Cores vs CUDA Cores: The Powerhouses of GPU Computing from Nvidia</a><h2 dir="ltr">Comparing the Performance of NPUs vs TPUs</h2>NPUs and TPUs exhibit distinct performance characteristics, making them suitable for different AI workloads. Following section compares both the AI chips based on different parameters.<h3 dir="ltr">Speed and Latency</h3>NPUs&nbsp;are designed for&nbsp;real-time, low-latency&nbsp;tasks, making them ideal for&nbsp;edge computing&nbsp;scenarios where immediate responses are required. For instance,&nbsp;NPUs&nbsp;enable instant image processing in&nbsp;smartphones&nbsp;and&nbsp;IoT devicesTPUs often excel in speed and scalability for large datasets, leveraging their specialized architecture for tensor computations. They are optimized for throughput rather than latency, making them suitable for batch data processing tasks like training massive deep learning models. In contrast, NPUs are designed for real-time, low-latency tasks on the edge.&nbsp;<h3 dir="ltr">Power and Energy Efficiency</h3>Power consumption is a critical factor for both NPUs and TPUs, but their optimization differs. NPUs provide better efficiency for edge computing and low-power applications, optimizing computing tasks with limited computational resources. &nbsp;This ensures extended battery life for small devices while performing complex AI tasks. TPUs, while efficient for their scale, are designed for environments like data centers where power availability is less constrained but efficiency remains essential for handling extensive computations.<h3 dir="ltr">Scalability and Flexibility</h3>NPUs&nbsp;are less scalable compared to&nbsp;TPUs, as they are tailored for specific edge&nbsp;AI&nbsp;applications. However, their&nbsp;flexibility&nbsp;in&nbsp;real-time&nbsp;processing tasks makes them invaluable for&nbsp;localized AI workloads.&nbsp;TPUs, on the other hand, are built for&nbsp;scalability, capable of handling&nbsp;large-scale&nbsp;AI&nbsp;models and&nbsp;datasets&nbsp;in&nbsp;cloud-based&nbsp;or&nbsp;distributed environments.<h2 dir="ltr">Architectural Differences</h2><h3 dir="ltr">Core Design</h3>NPUs&nbsp;are engineered to handle&nbsp;neural network activations&nbsp;and&nbsp;real-time&nbsp;matrix computations with specialized cores that focus on efficient, localized processing. These cores typically feature lightweight architectures optimized for low power consumption and latency-sensitive tasks.&nbsp;TPUs, in contrast, utilize systolic array architectures with a high density of computational cores designed for parallel matrix multiplications.<h3 dir="ltr">Programming Frameworks and Ecosystem</h3>NPUs are compatible with general-purpose AI frameworks such as TensorFlow, PyTorch, and ONNX, allowing developers to integrate AI tasks into various devices. As they are integrated within larger processors, they are more general-purpose and are compatible with a number of different computing tasks. TPUs, in contrast, are tightly integrated with Google’s TensorFlow ecosystem, making them an ideal choice for developers already working within Google’s AI infrastructure. The trade-off lies in flexibility versus specialization.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNzM3NTY2NDEzOTA4LWltYWdlMS5wbmciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 100%px;" class="fr-fic fr-dib" alt="intel-core-ultra-gen-1-series-with-NPUs">Intel Core Ultra processors integrate NPUs to deliver efficient and enhanced performance for modern AI desktop applications.<h2 dir="ltr">Use Cases and&nbsp;Practical Applications</h2>NPUs and TPUs cater to distinct scenarios based on their specialized capabilities.NPUs find their strengths in edge computing, enabling AI tasks directly on devices like smartphones, cameras, and IoT sensors. Examples of&nbsp;NPU Applications&nbsp;include:<ul><li dir="ltr">Real-time object detection in edge devices, such as security cameras.</li><li dir="ltr">Voice recognition in smart assistants operating on low-power devices.</li><li dir="ltr">AI-powered healthcare devices for monitoring patient vitals and diagnosing conditions.</li><li dir="ltr">Industrial <a href="https://www.wevolver.com/article/building-automation-systems">automation systems</a> requiring real-time decision-making capabilities.</li></ul>TPUs are the backbone of AI operations on large amounts of data. Examples of&nbsp;TPU Applications&nbsp;include:<ul><li dir="ltr">Training large-scale machine learning models in cloud environments.</li><li dir="ltr">Batch processing for natural language understanding tasks in virtual assistants.</li><li dir="ltr">Optimizing recommendation systems in large-scale e-commerce platforms.</li><li dir="ltr">Enabling real-time analytics in large datasets for financial and scientific domains.</li></ul><h2 dir="ltr">Challenges and Limitations of NPUs and TPUs</h2><h3 dir="ltr">Engineering Challenges in Implementation</h3>Deploying NPUs and TPUs presents several technical challenges, including:<ul><li dir="ltr">Compatibility Issues: Existing software ecosystems, such as libraries and frameworks, often require extensive adaptation to fully leverage the capabilities of NPUs and TPUs.</li><li dir="ltr">Hardware Integration:&nbsp;Ensuring seamless integration with existing system architectures can be complex, particularly in legacy systems.</li><li dir="ltr">Optimization Challenges:&nbsp;AI models often need to be specifically optimized to achieve peak performance on these specialized processors.</li><li dir="ltr">Cost-Effectiveness:&nbsp;While powerful, the initial investment in NPU or TPU hardware can be prohibitive for smaller organizations.</li></ul><h3 dir="ltr">Limitations in Real-World Applications</h3>Despite their capabilities, NPUs and TPUs face limitations in specific environments:<ul><li dir="ltr">Latency-Sensitive Systems: TPUs, designed for high-throughput batch processing, may not be ideal for real-time applications that demand low latency.</li><li dir="ltr">Hybrid Requirements: Certain applications, such as autonomous vehicles, may require a combination of TPU and graphics processing unit (GPU) capabilities to balance precision and speed effectively.</li><li dir="ltr">Scalability Constraints:&nbsp;While TPUs scale well in cloud environments, NPUs are less adaptable to large-scale deployments due to their edge-specific design.</li></ul>Here is a table summarising the discussion on NPUs and TPUs:<div dir="ltr" align="left"><table><tbody><tr><td>Feature</td><td>NPUs</td><td>TPUs</td></tr><tr><td>Primary Use Case</td><td>Real-time AI tasks on edge devices (e.g., smartphones, laptops, IoT, robotics)</td><td>Large-scale AI workloads in data centers (e.g., training deep learning models)</td></tr><tr><td>Key Applications</td><td>Image recognition, natural language processing, voice recognition, augmented reality</td><td>Language models, natural language processing, large-scale image classification</td></tr><tr><td>Architectural Focus</td><td>Low-latency, power-efficient processing for real-time AI tasks</td><td>High-performance parallel processing and matrix multiplication for deep learning</td></tr><tr><td>Core Design</td><td>Lightweight architectures optimized for low power and real-time processing</td><td>Systolic array architectures optimized for massive parallel tensor operations</td></tr><tr><td>Scalability</td><td>Limited scalability, designed for specific edge AI tasks</td><td>High scalability for training large datasets and deploying AI models</td></tr><tr><td>Power Consumption</td><td>Optimized for energy efficiency in portable devices</td><td>Designed for efficiency in data centers with less strict power constraints</td></tr><tr><td>Programming Ecosystem</td><td>Supports general-purpose AI frameworks (e.g., TensorFlow, PyTorch, ONNX)</td><td>Tightly integrated with TensorFlow and Google Cloud services</td></tr><tr><td>Manufacturers</td><td>Companies like Apple (Neural Engine), Huawei, Samsung, and others</td><td>Google</td></tr><tr><td>Integration</td><td>Embedded in SoC designs for smartphones and IoT devices</td><td>Used as standalone accelerators in server environments</td></tr><tr><td>Key Strengths</td><td>Real-time processing, low latency, energy efficiency</td><td>Exceptional performance for large-scale AI training and inference</td></tr><tr><td>Key Limitations</td><td>Less suited for large-scale AI workloads</td><td>Higher power requirements, less suited for edge applications</td></tr></tbody></table></div><h2 dir="ltr">Conclusion</h2>NPUs and TPUs play complementary roles in the advancement of AI technology, each excelling in specific environments. While TPUs offer unparalleled performance for large-scale machine learning tasks in cloud environments, NPUs are perfect for real-time inference and energy-efficient edge applications.&nbsp;<h2 dir="ltr">Frequently Asked Questions</h2><ol><li dir="ltr">What Is the Primary Difference Between NPUs and TPUs? NPUs are specialized for low-power, real-time tasks, making them ideal for edge devices. TPUs are optimized for large-scale tensor computations, excelling in cloud-based machine learning.</li><li dir="ltr">What Are the Key Roles of NPUs and TPUs in AI? NPUs focus on real-time, low-power AI processing in edge devices like smartphones and IoT systems, enabling efficient inference for tasks such as image recognition and voice commands. TPUs specialize in high-performance, large-scale tensor computations, excelling in data centers for AI training and deployment.</li><li dir="ltr">How Do NPUs and TPUs Compare to GPUs? NPUs are specialized for efficient AI/ML tasks, particularly inference, and are optimized for low-power edge devices like smartphones. TPUs excel in tensor-based operations for deep learning, offering high-speed training and inference, primarily in Google's ecosystem. GPUs, versatile in design, handle a wide range of workloads, including graphics rendering, ML training, and scientific computing, making them ideal for parallel processing. While GPUs are general-purpose, NPUs and TPUs focus on AI-centric optimizations.</li></ol>Recommended reading:&nbsp;<a href="https://www.wevolver.com/article/understanding-nvidia-cuda-cores-a-comprehensive-guide" target="_blank">Understanding Nvidia CUDA Cores: A Comprehensive Guide</a><h2 dir="ltr">References</h2>[1] Google. Cloud TPU documentation. Available from: <a href="https://cloud.google.com/tpu/docs" target="_blank">https://cloud.google.com/tpu/docs</a>[2] Apple. Deploying Transformers on the Apple Neural Engine. Available from: <a href="https://machinelearning.apple.com/research/neural-engine-transformers" target="_blank">https://machinelearning.apple.com/research/neural-engine-transformers</a>[3] Van Belle JL. Why neural processing units (NPUs) are the next Big Thing in AI. 2024. doi:10.13140/RG.2.2.35619.87845. Available from: <a href="https://www.researchgate.net/publication/383184868_Why_neural_processing_units_NPUs_are_the_next_Big_Thing_in_AI" target="_blank">https://www.researchgate.net/publication/383184868_Why_neural_processing_units_NPUs_are_the_next_Big_Thing_in_AI</a>[4] Sanmartín D, Prohaska V. Exploring TPUs for AI Applications. IE Robotics &amp; AI Club, IE University, Madrid, Spain. arXiv preprint arXiv:2309.08918. Available from:<a href="https://arxiv.org/pdf/2309.08918" target="_blank">&nbsp;https://arxiv.org/pdf/2309.08918</a>.

NPUs are integrated units that excel in real-time AI tasks on edge devices like smartphones and IoT systems with low power consumption. TPUs are standalone processors designed for large-scale AI workloads in data centers, delivering exceptional performance in deep learning tasks.

NPU vs TPU: Understanding the Key Differences in AI Hardware Accelerators

GPUs excel in parallel processing for graphics and AI training with scalability, while NPUs focus on low-latency AI inference on edge devices, enhancing privacy by processing data locally. Together, they complement each other in addressing different stages of AI workloads efficiently.

NPU vs GPU: Understanding the Key Differences and Use Cases

Fine-tuning large language models adapts pre-trained models to specific tasks or domains using tailored datasets, while Retrieval-Augmented Generation (RAG) combines retrieval systems with generative models to dynamically incorporate external, up-to-date knowledge into outputs.

RAG vs Fine-Tuning: Differences, Benefits, and Use Cases Explained

Large Language Model (LLM) training involves teaching AI models to understand and generate human-like text by processing vast amounts of data, significantly enhancing their language comprehension and production capabilities.

LLM Training: Mastering the Art of Language Model Development

Industrial process automation is changing manufacturing by using technology to reduce the need for human involvement. From robots assembling products to smart systems managing inventory, factories are being transformed. This article will explore the successful elements behind industrial automation.

Industrial Process Automation: The Future of Manufacturing

A Programmable Logic Circuit (PLC) with analog digital input and output card, with power supply and processor module

A Programmable Logic Controller (PLC) is a specialized computer that operates factory machines, power plants, and even theme parks. Unlike a regular PC, a PLC can handle tough environments to manage production lines or operate elevators. This article will discuss its operations and unique features.

What is a PLC (Programmable Logic Controllers): A Comprehensive Guide

Communication protocols in industrial-based PLC automation system

This article delves into the key difference between PLC and DCS, providing a clear framework for choosing the appropriate technology for specific applications, and finally decoding the automation divide. 

Difference between PLC and DCS: Decoding the Automation Divide

In medical devices, automotive systems, and industrial automation, software is a regulated component of the product itself. That changes what "good enough" means for a build process. A binary you cannot reproduce is a binary you cannot validate, and a build environment you cannot audit is a liability in every certification cycle you run.Despite this, many embedded teams still operate with manual or loosely controlled build processes: toolchains installed directly on developer machines, no enforced versioning, and test and analysis steps that run inconsistently, if at all. The problems this creates are well understood in theory but chronically underestimated in practice. Build reproducibility directly affects validation scope, certification timelines, and release predictability.CI/CD pipelines provide a structured path out of this. Automated builds, integrated analysis, and containerized environments together standardize execution, generate compliance evidence continuously, and reduce the audit burden at release time. Adoption does not require a dedicated DevOps team or a full infrastructure overhaul. Teams can start with automated builds alone and extend the pipeline incrementally from there.<h2>The Real Price of "Good Enough" Build Processes</h2>The most immediate problem with uncontrolled build environments is variability. When each developer maintains their own toolchain configuration, subtle differences accumulate, from compiler versions to library paths, optimization flags, and environment variables. Two developers building the same commit may produce different binaries. In most software contexts, this is inconvenient; in regulated development, it is a compliance failure waiting to materialize.Unqualified or non-certified toolchains compound this. During audits, teams must self-justify tool behavior, documenting that a given compiler version produces correct output under the conditions used. This is significant validation work that certified toolchains eliminate by shifting that burden to a qualified third party.The downstream effects are predictable. When defects are found late, as they inevitably are when analysis and testing run manually or infrequently, debugging requires first reconstructing the environment the defect occurred in. That reconstruction may not be possible if the environment state was never captured. Troubleshooting build and pipeline failures alone can consume up to <a href="https://www.iar.com/blog/enabling-consistent-workflows-with-cloud-enabled-solutions">40% of development time</a>, time that should be going toward innovation, testing, and actual feature work.Audit preparation in these environments is largely manual. Engineers assemble evidence from disparate sources: build logs from one system, test results from another, and analysis outputs from a third. Certification timelines extend unpredictably not because the product is unsafe but because the documentation trail is incomplete or inconsistent. Engineering effort shifts toward rework and environment management rather than toward the product.<h2>Anatomy of an Embedded CI/CD Pipeline</h2>A well-structured embedded CI/CD pipeline follows a 5-stage, fixed sequence: Code → Build → Analyze → Test → DeployCode commit triggers build, which feeds into static analysis, which gates automated testing, which then produces validated artifacts for deployment. Each stage is automated, version-controlled, and produces structured outputs.<a href="https://www.iar.com/embedded-development-tools/embedded-ci-cd?utm_source=web&amp;utm_medium=article&amp;utm_campaign=wevolver" target="_blank" rel="noopener noreferrer"><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1777030324914-image1.png" style="width:100%px" class="fr-fic fr-dib" />A Modern CI/CD Workflow. Credit: IAR</a>The value of this goes beyond speed. The pipeline becomes a system of record. Build logs, analysis reports, and test results are generated automatically on every run, linked to the specific commit that produced them. The path from source code to deployed binary is fully traceable and reproducible. When a regulator asks for evidence that a given artifact was built from a known state and tested against defined criteria, that evidence exists and is organized.Continuous validation through static analysis and runtime checks, integrated directly into the pipeline, shifts defect detection to the earliest possible point. A coding standard violation found at commit time costs minutes to address. The same violation found during integration testing can cost days.<h2>Containerization: The Foundation for Consistency</h2>Reproducibility requires that the build environment itself is controlled and versioned. <a href="https://www.iar.com/blog/a-new-chapter-for-our-docker-support-introducing-the-iar-cloud-ready-containers-repository?utm_source=web&amp;utm_medium=article&amp;utm_campaign=wevolver">Containers</a> solve this directly. Packaging the toolchain, compiler flags, analysis tools, and all dependencies into a versioned container image means every pipeline run, whether on a developer's workstation, a cloud runner, or an on-premises CI server, executes in an identical environment.This has a direct compliance benefit. Qualification effort can be focused on the container image rather than each developer's local setup. Once the toolchain configuration inside a container is qualified, that same configuration runs identically everywhere, eliminating the risk that a local toolchain update silently changes the environment under assessment. When <a href="https://www.iar.com/embedded-development-tools/functional-safety?utm_source=web&amp;utm_medium=article&amp;utm_campaign=wevolver">TÜV-certified</a> toolchains are packaged inside containers, the certified boundary is clear and fixed. The container image itself becomes auditable: its contents are known, its version is tracked, and its behavior is consistent across infrastructure.Isolated container environments also provide a security boundary. Dependencies cannot drift silently, and there is no ambiguity about which version of a tool was used for a given build.<h2>Performance and Quality Gains You Can Measure</h2>The shift to automated, containerized pipelines produces concrete gains.<a href="https://www.iar.com/embedded-development-tools/embedded-ci-cd?utm_source=web&amp;utm_medium=article&amp;utm_campaign=wevolver"> IAR Build Tools</a> on Ubuntu deliver build times roughly twice as fast as equivalent Windows-based workflows, and <a href="https://www.iar.com/embedded-development-tools/code-quality-and-compliance?utm_source=web&amp;utm_medium=article&amp;utm_campaign=wevolver">C-STAT</a> static analysis runs approximately 3.5 times faster. Across build and debugging effort combined, automated analysis with pre-qualified toolchains reduces total time by 50 to 80 percent.C-STAT enforces MISRA C/C++, CERT C/C++, and CWE coding standards across every pipeline execution. At that speed advantage, comprehensive standards coverage stops being a bottleneck and becomes a continuous, low-friction part of the build process. Quality metrics follow a similar pattern. Continuous enforcement of coding standards on every commit, rather than periodic manual review, produces roughly a <a href="https://www.iar.com/embedded-development-tools/code-quality-and-compliance">41% reduction in defect injection rates</a>. Defects caught earlier in the pipeline require significantly less effort to fix, and downstream testing loads decrease as a result.These are not marginal improvements. For teams operating under release pressure in regulated markets, halving build and debug time while reducing late-cycle defects directly affects both certification timelines and product quality.<h2>Starting Small: Incremental Adoption</h2>CI/CD adoption in embedded development does not require a dedicated DevOps team or a large infrastructure investment. The risk of incremental adoption is low: starting with automated builds does not disrupt existing workflows or trigger re-validation requirements.The practical entry sequence is to automate builds first, capturing the toolchain configuration and making builds reproducible. Then, introduce static analysis as a pipeline gate. Afterwards, you add automated testing. At each step, containerized environments simplify setup and make the expanded pipeline portable. Teams retain ownership of their pipelines throughout. The incremental approach also means compliance evidence accumulates progressively rather than requiring a large documentation effort at the end of a release cycle.<h2>Enabling Embedded CI/CD with IAR </h2>IAR's toolchains are built for automation-first workflows. IAR Build Tools are headless, scriptable, and integrate with standard CI platforms including GitHub Actions, GitLab CI, and Jenkins. Static analysis through C-STAT and runtime verification through C-RUN run inside pipelines without manual intervention. Toolchains pre-certified to IEC 61508 (industrial automation), ISO 26262 (automotive), and IEC 62304 (medical devices) reduce the validation work teams must do themselves during certification.Cloud-based licensing removes the constraint of node-locked licenses tied to specific machines, enabling consistent workflows across local development, cloud runners, and on-premises systems. Support for over 20 processor architectures means teams working across device families can standardize on a single toolchain ecosystem without being forced into architecture-specific tool fragmentation.IAR Embedded Trust extends pipeline automation to firmware security. Automated signing and encryption within the pipeline integrates security requirements from CRA and UN R155/R156 directly into the build process rather than treating them as post-build steps.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1777030409990-image2.png" style="width:100%px" class="fr-fic fr-dib" />Credit: IAR<h2 id="isPasted">Conclusion</h2>Build reproducibility in modern embedded development is the foundation that compliance work depends on. Automated workflows, built on certified toolchains, ensure consistent execution, validation, and traceability from commit to deployed artifact. An environment that cannot consistently reproduce a binary cannot validate, certify, or maintain it.Regulatory expectations around software development processes are continuing to formalize across medical, automotive, and industrial sectors. Teams that establish reproducible pipelines now gain direct control over their development timelines and compliance readiness, rather than discovering gaps under deadline pressure during certification. The cost of manual builds is always higher than it appears, and it compounds with each release cycle.

This article explores how automated CI/CD pipelines replace fragile manual build processes with reproducible, auditable workflows that improve compliance, reduce defects, and accelerate development in regulated embedded systems

The Hidden Cost of Manual Builds: Why Teams in Regulated Sectors Must Embrace CI/CD

Work from Carnegie Mellon University may someday help doctors measure skin health as precisely as a blood test. Researchers in the university’s Robotics Institute (RI) are turning that possibility into reality with new software that works with a compact imaging device to bring the skin’s microscopic landscape into clearer focus. “When skin texture becomes a measurable signal rather than a subjective one, dermatology can move toward truly quantitative assessments, helping clinicians to track disease progression and treatment response over time,” said Akhil Padmanabha, a 2025 RI Ph.D. graduate and lead researcher on the project. “With that foundation, skincare efficacy can be evaluated with the same rigor as medical tests.”The team’s software interprets subtle variations in the skin’s surface by locating wrinkles and comparing them to nearby high points to estimate depth. In doing so, it transforms otherwise imperceptible skin texture changes into data that can be both visualized and measured. Paired with a handheld imaging device, the software creates detailed 3D maps of the skin’s surface, revealing fine features with newfound consistency. The work could be used in both clinical and cosmetology settings. It could aid in diagnosing and monitoring textural changes in some skin conditions such as eczema and psoriasis and help evaluate treatment efficacy. It could also give insights into skin aging, sun damage and the effectiveness of skincare products.“From a medical standpoint, these technologies have the potential to fundamentally change how we assess tissue in a noninvasive and objective way,” said <a href="http://dermatology.pitt.edu/people/sonal-choudhary-md" target="_blank" rel="noreferrer noopener">Dr. Sonal Choudhary</a>, a dermatologist at UPMC who worked on the research. “In dermatology and other clinical fields where visual inspection and palpation are still highly subjective, high-resolution tactile sensing could improve diagnostic accuracy; enable earlier detection of disease; and support more personalized, data-driven care. It opens the door to better remote assessment and training.”The compact sensor, 3D-printed probe and newly designed elastic gel work together to measure fine skin surface features.Current portable 3D imaging technologies typically lack the resolution or reliability needed to work consistently on soft surfaces like skin and across different parts of the human body. These gaps in technology make it difficult to accurately measure skin texture.To address this challenge, the RI team adapted a tactile imaging technique known as <a href="https://www.gelsight.com/" target="_blank" rel="noreferrer noopener">GelSight</a> that creates detailed 3D maps by observing how surfaces deform when pressed against an object. The team expanded GelSight technology to measure fine skin surface features using a compact sensor, a 3D-printed probe and a newly designed elastic gel that conforms to the skin. The system improves on the accuracy of currently available commercial devices.The custom elastic gel reduces how much skin deforms when the sensor is pressed against it. It has three layers: a flexible base that supports the structure, a powder coating that helps the sensor capture surface slopes, and a skin-safe layer that covers it all. The layered design allows the device to create precise 3D maps of the skin while maintaining gentle contact.“We attached our custom gels to a GelSight sensor and integrated the sensor with a 3D-printed probe that can be used on skin,” Padmanabha said. “We then trained algorithms to estimate 3D geometries from 2D images and to identify and analyze microstructures on the skin surface rather than relying on the naked eye.”To further validate their probe and skin pattern detection methods, the team conducted a human study with participants who had no diagnosed skin disorders. The goal was to measure skin texture across multiple areas of the body and proceed to evaluate how those measurements changed after the application of an over-the-counter moisturizer. The system detected consistent, measurable reductions in wrinkle depth, demonstrating an accurate sensitivity to everyday skin interventions.“Finding the right material to match the compliance of skin turned out to be an interesting challenge for this project,” said Carmel Majidi, CMU’s Clarence H. Adamson Professor of <a href="https://www.meche.engineering.cmu.edu/index.html" target="_blank" rel="noreferrer noopener">Mechanical Engineering</a> and a part of the research team. “We wanted something gel-like that could conform to the features and texture of the skin but also have enough elastic integrity to convey these features to our optical tracking system in a repeatable and reliable manner. In the end, we came up with a material system that had just the right balance.”

CMU researchers develop software that turns subtle skin changes into trackable data

A New Way to Track Skin Health

Embedded SIM or eSIMs are essential to unlocking next-generation device design and scalability. At the heart of eSIM functionality is the eUICC, the Embedded Universal Integrated Circuit Card that sits inside the device, stores operator profiles, handles authentication and enables updates via Remote SIM Provisioning RSP, removing the need for physical SIM swaps and cumbersome practices that stifled prior Over-the-air update mechanisms.The topic of always-connected isn't just about mobiles anymore. Connectivity is now a core requirement across wearables, asset tracking, EVs, e-bikes, industrial equipment, and a growing range of IoT devices. The global connected device volume is expected to hit 23.9 billion devices in 2026 [1]. This shift has been enabled by miniaturization, which reduces the physical size of components while increasing efficiency.Today's products are expected to always be connected, work seamlessly across the globe, and meet strict security requirements. To make this happen, miniaturiization is no longer optional. It is a crucial requirement to build the next generation of connected products.<h3>Why designers feel the squeeze</h3>As user expectations continue to rise, product designers are under constant pressure to deliver more features, such as always-on connectivity and AI, while keeping devices compact. On a PCB, every millimeter matters [2]. Designers need space for processors, memory, sensors, and antennas. Even with careful layout optimization, traditional connectivity hardware can feel dated in the current scheme of things.Removable SIM cards require trays and mechanical tolerances that use up valuable PCB area. To address this issue, many product teams are moving toward embedded SIM solutions to optimize space and efficiency. According to GSMA specifications, an embedded form factor can save up to 90% of the space compared to traditional SIM housing [3].Embedded form factors help towards this - however, once you solder an eUICC on your device PCB, management and customization to destinations can be hard to manage as a different soldered component is needed for each geography or market. This can be a hindrance to managing product variants or SKU (Stock Keeping Units), and harmonizing reliability testing during manufacturing. This implementation challenge provided impetus to the need for eSIMs in IoT - <a href="https://www.wevolver.com/article/ces-2026-how-physical-ai-is-redefining-mobile-and-why-esim-matters">something we’ve discussed earlier.</a><h3>SIM form factors: what product teams need to know</h3>Not all SIM cards are made the same way, though. To understand why eSIMs stand out, it helps to understand how SIM form factors (FF) have evolved over the years.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1774385483914-1774385483914.png" width="602" height="343" class="fr-fic fr-dii" />Evolution of SIM technology over the years. Source: Kigen<h3>1FF (Full SIM)</h3>These were the first commercially used SIM cards. Introduced in 1991, 1FF cards were roughly the size of a credit card and were primarily used in early car phones. At the time, device size was not a major constraint, so the large format was acceptable.<h3>2FF (Mini SIM)</h3>This became the standard SIM for early GSM mobile phones. Measuring 25 × 15 mm, it significantly reduced size compared to 1FF while remaining removable. For many years, 2FF was the dominant format in feature phones.<h3>3FF (Micro SIM)</h3>As smartphones became slimmer, Micro SIMs reduced the plastic around the chip while keeping the same electrical interface. Popularized in the early 2010s, this format helped free up internal space for other components such as the memory card slot.<h3>4FF (Nano SIM)</h3>The smallest removable SIM format in widespread use today. Nano SIMs strip away the excess plastic, leaving just the chip with a minimal bezel. While they save space compared to earlier formats, 4FF cards still require a tray and connector.<h3>MFF2</h3>MFF2 was the first widely adopted soldered SIM form factor in IoT and is often what people mean when they refer to eSIM hardware. Defined by ETSI, MFF2 measures 5 × 6 × 0.75 mm and uses an 8-pin layout designed to operate across industrial temperature ranges.By soldering the SIM directly onto the PCB, MFF2 removes the need for SIM trays and connectors. This improves mechanical robustness, reduces failure points, and frees up valuable board space. These advantages have made MFF2 a popular choice for industrial, automotive, and long lifecycle IoT products where reliability and longevity are crucial.<h3>MFF4</h3>MFF4 pushes miniaturization even further with its compact design. With a footprint of just 2 × 2 mm, MFF4 significantly reduces the space requirement for cellular connectivity.Despite its small size, MFF4 is a compact 8-pin soldered package that complies with JEDEC MO-229F specifications [4]. It supports UICC signals including VCC, RST, CLK, IO, and interfaces such as I3C or SPI.<h3>How does MFF4 form factor work with eSIM architecture?</h3>By dramatically shrinking the SIM footprint, designers get more freedom at the system level. From a manufacturing perspective, soldered eSIMs lead to more streamlined assembly processes compared to removable SIMs. There are fewer mechanical failure points, making it easier to build rugged products ideal for industrial settings.These form factors often matter in product design as physically containing the capability of an energy efficient, fully functional and remotely provisionable eSIM into the ever compact devices such as wearables, asset trackers, meters, etc, has been nigh impossible in prior years. Often, the limiting factor is the memory size that allows for adequate number of connectivity profiles as well as the critical software stack that can enable security - the primary function of the SIM or eSIM and all the business logic for eSIM functional management.From the view of how these can be remotely provisioned, the eSIM RSP standards are largely form-factor agnostic. The same operating system and subscription management logic can run across MFF2, and MFF4-based eSIM, or iSIM implementations. But this will depend on the provisions available to you to fit the secure OS, associated applets for business logic and profile management and the needed profiles all within the memory size. The secret to achieving this is an OS tuned to the needs of energy-efficient devices, such as those IoT demands. And an added advantage with choosing the right eSIM OS is that the one time investment of the decision and implementation expertise spans across all form factors easily, keeping the toolchain consistent and simplifying scalability across product portfolios. <h3>Putting this into practice for eSIM Operations: Why SGP.32 Matters</h3>The requirement of being able to have multiple profiles either pre-provisioned or switchover can result from a variety of use-case requirements. But, this is where a new standard addresses how to manage complexity not only within one device but to take it to fleet level in various scenarios.Managing connectivity across a fleet of thousands devices can be an operational challenge - something that now is increasingly in the realms of product design from the word go. Different regions, multiple network operators, and various data plans add to the complexity. This is where eSIM lifecycle management standard SGP.32 streamlines the process to manage devices such as robots, smart meters, and connected vehicles. It is compatible with LPWAN (Low Power Wide Area Networks), which may use HTTPS/CoAP or other data protocols and not able to support SMS for provisioning as was the case in smartphone use (as consumers) [5]. Additionally, SGP.32 offers secure remote profile downloads and subscription switching without in-person access to devices. It also allows fallback or bespoke profiles to be deployed remotely in case of network issues or when business conditions demand.<h3>Key Advantages of eSIM’s multi-faceted evolution</h3>Moving to eSIMs deliver several benefits. For starters, soldered and embedded form factors, especially as compact as MFF4, free up valuable PCB area for additional electronics, sensors, batteries, radios, or for other innovations. The soldered approach also reduces connector points prone to mechanical failure. eSIM designs are also less susceptible to wear and tear. Security, the primary function of SIM technology is also stronger. In the physical sense, eSIMs are harder to tamper, which is critical for industrial, automotive, and medical devices. Functionally, eSIM is the implementation of a SE (Secure Element) layer for cryptographic operations and to enforce robust authentication protocols - with the new remote management capabilities introduced in SGP.32 with the eIM, this becomes the digital anchor for trust and end-to-end security.From a business perspective for manufacturers, eSIM enables SKU consolidation. A single hardware variant can be deployed globally, with connectivity profiles provisioned over the air. This streamlines supply chains and enables faster market entry. Operationally, eSIMs adherent to SGP.32 and functionally certified with the GSMA eSA scheme, allow for the secure transfer of profiles through data structure that is standardised for simpler fleet level remote access. When such eSA scheme certified eSIMs are used, this allows efficient integration for the best utility for the eSIM - either through a device-side agent called the IoT Profile Assistant - native to the eUICC, or one that is resident in the device.<h3>Device-Side Integration: Choosing The Right eSIM Enablement Suite</h3>Device-side software integration plays a critical role in time-to-market and long-term success. Two popular integration approaches are IPAe and IPAd.<h3>IPAe</h3>IPAe (IoT Profile Assistant embedded) focuses on faster utilization of eSIM capabilities and simplifies GCF certification when a product uses an eSA certified eSIM. In this approach, the profile assistant resides within the eSIM chip rather than in the device's main operating system. This approach is effective for devices that need to end-to-end security or demand long battery life, or are low-power devices.<h3>IPAd</h3>On the other hand, IPAd (IoT Profile Assistant on the Device) hosts IPA software on the device's operating system. It offers more customization and control potentially desirable in complex use cases. However, it requires a device with sufficient processing power beyond the scope of many IoT devices. Additionally, it takes longer for development and certification work.Moving on to SDKs and software utilities, product teams should look for ones that translate across most devices for maximum compatibility to development toolchains. While the choice of IPAd SDKs is increasing, it is useful to understand if it is suited for IoT implementation, or a derivative of LPA used in smartphones. Further, pay good attention to the API structures, opting for SDKs that allow the scalable, open API3.0 or similar such that upstream integration and handovers are simpler through manufacturing, volume production and ultimately device lifecycle. Choosing the right enablement suite is not just about getting connected. It paves the way for certification, operational efficiency, and scaling.<h3>Case study: A global smartwatch brand switches to modern eSIM for scale and innovation</h3>A global smartwatch heavyweight moved to bring in LTE-connectivity to its premier products and expand into further global markets. With scale came growing complexity around region-specific hardware variants. Traditional SIM-based designs would not ever meet the compact hardware specs, plus, MFF2 soldered form factors required different SKUs for different markets, which slowed regional launches. If based on prior standards, pre-determining every aspect of one-to-one relationship per market would be a severe shackle to meet the market timeline for this innovator.Transitioning to MFF4-based eSIMs with SGP.32 features, the same hardware platform could be shipped worldwide to meet local market requirements, with the right customisation - i.e. with the right connectivity profile securely embedded per market during manufacturing.The brand launched with multiple networks, including non-terrestrial network (NTN) connectivity as a premium feature. It could appeal to trekkers or those who need a fallback in case of network outages or switchover for the profile that was suitable for streaming usecases vs pure messaging. By relying on eSIM standards such as SGP.32, these features can be introduced with minimal changes to the core hardware - and with the confidence that end-to-end security enables new use-cases that were never imagined possible on a smartwatch.Their success with premium product lines integrating eSIMs in volume production with over 500,000 per year should bear evidence that eSIM-first and eSIM-only device experiences are not restricted to premium smartphone brands alone, but within reach of wider IoT experiences. Furthermore, moving away from pre-determined agreements that rely on bootstrap profiles changing over to operational profiles when in field, the use of advanced SGP.32 features (typically deemed/found as ‘optional requirements’) and In-Factory Profile Provisioning from Kigen, allowed them to provision devices within minutes - shaving off multiple months of development delay.<h3>Partner Selection: What to Look For</h3>Choosing the right technology partner for eSIM integration is as critical as choosing the hardware itself. When evaluating vendors, look beyond the mandatory ‘Profile State Management Operations’ on the standards spec sheet. Prioritize vendor solutions that support quality in pragmatic scenarios, interoperability - both technically, and commercially, and troubleshooting capabilities. Look for evidence of successful deployments in similar verticals to see how they address real-world challenges. Effective partners can also prompt teams to address key connectivity questions early, including how devices will connect after production, how provisioning can be validated in real factory conditions, how to maintain a global SKU with local compliance, and how to scale securely across diverse networks<h3>Conclusion</h3>The future of hardware is compact and always connected. The MFF4 form factor is a key driver of this shift. It frees designers from focusing on connectivity logistics and allows them to focus more on building innovative products.If you want to explore how MFF4 and SGP.32-ready IoT eSIMs can support your business, get in touch with Kigen. It was the first company to introduce compact MFF4 eUICC solutions.<a href="https://kigen.com/contact/">https://kigen.com/contact/</a> Resources:1.  <a href="https://iot-analytics.com/number-connected-iot-devices/" target="_blank" rel="noopener noreferrer">https://iot-analytics.com/number-connected-iot-devices/</a>2.  <a href="https://www.proto-electronics.com/blog/techniques-for-pcb-design-under-space-constraints" target="_blank" rel="noopener noreferrer">https://www.proto-electronics.com/blog/techniques-for-pcb-design-under-space-constraints</a>3.  <a href="https://www.scribd.com/document/395847381/e-SIM-Report" target="_blank" rel="noopener noreferrer">https://www.scribd.com/document/395847381/e-SIM-Report</a>4.  <a href="https://kigen.com/resources/blog/the-future-of-esim-hardware/" target="_blank" rel="noopener noreferrer">https://kigen.com/resources/blog/the-future-of-esim-hardware/</a>5.  <a href="https://kigen.com/resources/blog/sgp-32-iot-esim-certification-and-oem-success/" target="_blank" rel="noopener noreferrer">https://kigen.com/resources/blog/sgp-32-iot-esim-certification-and-oem-success/</a>

for the always-connected world

How MFF4 eSIM Enables Miniaturised Device Design

The event takes place from March 10 to March 12, 2026 in Nuremberg, Germany. It brings together companies, engineers, and technology providers from across the embedded systems ecosystem.  <a href="https://www.embedded-world.de/en" target="_blank" rel="noopener noreferrer" class="button-main-action">Learn More</a>embedded world covers the full stack of embedded development, from hardware components and modules to operating systems, software development tools, and system integration. The exhibition presents new products and services while the conference program focuses on technical and industry topics related to embedded systems design.  The event is a meeting point for engineers, developers, system architects, and product teams working with embedded technologies. Attendees can explore solutions related to hardware design, software engineering, machine to machine communication, safety and security, and connected devices.  Alongside the exhibition, the conference includes technical presentations and expert sessions covering current developments in embedded systems. Topics include embedded AI, embedded vision, IoT, system architecture, and secure system design.  The exhibition floor includes companies presenting semiconductors, development tools, software platforms, and embedded computing solutions. Visitors can learn about new technologies, meet suppliers, and explore tools used in embedded system development.  embedded world takes place annually in Nuremberg and serves as a central meeting point for the embedded systems industry. The event provides an overview of current technologies and ongoing developments across the embedded ecosystem.  <h2 id="isPasted">Core topics of the industry</h2>The embedded world offers an unparalleled insight into the whole world of embedded systems:  <ul><li>Hardware </li><li>Systems </li><li>Distribution </li><li>Services </li><li>Tools </li><li>Application software </li><li>Electronic displays </li><li>Embedded vision </li><li>Safety&amp;security </li><li>IC&amp;IP design </li><li>Start-up  </li></ul><h2 id="isPasted">embedded world Conference</h2>The embedded world Conference 2026 offers three days of knowledge transfer at the highest level. From 10 to 12 March 2026, participants and speakers will take the opportunity to exchange ideas and expand their knowledge with top-class experts.In 2026, the embedded world Conference will once again bring together the state of the art with trend topics in a very clearly structured programme. Experts from research and development will exchange ideas with colleagues from practice. The entire embedded industry benefits from this knowledge transfer.<a href="https://www.embedded-world.de/en/conferences-programme/embedded-world-conference" target="_blank" rel="noopener noreferrer" class="button-main-action">Learn More</a>

An international trade fair and conference focused on embedded systems and the technologies used to design and build them.

embedded world Exhibition & Conference 2026

As robotics rapidly shifts toward systems that can safely assist, collaborate, and adapt in human environments, the ability to perform coordinated, human-like actions has become a defining frontier. From precision assembly to remote intervention, bimanual manipulation and intuitive teleoperation are unlocking new levels of dexterity and autonomy in modern robots.In this webinar, hosted by Franka Robotics, Prof. Fei Chen, one of the leading voices in collaborative and humanoid robotics—shares how the CURI human-like robot is advancing this frontier through innovative control, interaction, and teleoperation strategies. Join us to explore the technologies shaping the next generation of intelligent, versatile robotic systems. <a href="https://wevlv.co/frank_webinar_lp_direct" target="_blank" rel="noopener noreferrer" class="button-main-action">Register now</a><h2>Event Details</h2><h3 id="isPasted">📅 Date: December 10, 2025</h3><table><tbody><tr><td>Location</td><td>Virtual (Join from anywhere)</td></tr><tr><td>Time</td><td>10 AM (CET)</td></tr></tbody></table><h2> </h2><h2 id="isPasted">Who should attend</h2><h2> </h2><ul><li>Industrial automation leaders exploring next-generation collaborative robotic systems</li><li>Robotics engineers and researchers working on manipulation, teleoperation, or human–robot interaction</li><li>Developers and integrators interested in dexterous, human-like robotic capabilities</li><li>Academics and students studying advanced control, bimanual manipulation, or humanoid robotics</li><li>Innovation teams assessing emerging robotic technologies for real-world applications</li></ul><h2>About the Speaker</h2><table><tbody><tr><td><img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNzYzNjM4MzUwODkxLUZlaV9fLnBuZyIsImVkaXRzIjp7InJlc2l6ZSI6eyJ3aWR0aCI6OTUwLCJmaXQiOiJjb3ZlciJ9fX0=" style="width:100%px" class="fr-fic fr-dib" /></td><td>Prof. Fei Chen, Head of the Collaborative and Versatile Robots Laboratory at The Chinese University of Hong Kong, focuses his research on humanoid bimanual manipulation and human-robot collaboration. Prior to joining CUHK, he led research efforts at the Italian Institute of Technology and contributed to numerous European, Italian, and Hong Kong projects. With more than 100 publications, he is also deeply engaged in the IEEE community, serving in key leadership roles across major international robotics conferences and technical committees. In his talk on Bimanual Manipulation and Teleoperation of the CURI Human-like Robot, he highlights how advanced control and interaction strategies are shaping the next generation of intelligent, versatile robotic systems. </td></tr></tbody></table><h2 id="isPasted">Why you should attend</h2><ul><li>Learn how bimanual manipulation enables more complex, human-level tasks in collaborative robotics</li><li>Understand cutting-edge control and interaction strategies used in the CURI human-like robot</li><li>Discover how researchers working on advanced manipulation can benefit from using a highly sensitive <a href="https://franka.de/franka-research-3" target="_blank" id="isPasted">cobot arm</a> that complements bimanual teleoperation studies. <a href="https://wevlv.co/frank_webinar_lp_direct" target="_blank" rel="noopener noreferrer"><img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNzYzNjM4Mjg2MjE2LVdlYmluYXIgcG9zdCAxMCBEZWMgKDEpLmpwZyIsImVkaXRzIjp7InJlc2l6ZSI6eyJ3aWR0aCI6OTUwLCJmaXQiOiJjb3ZlciJ9fX0=" style="width:100%px" class="fr-fic fr-dib" /></a></li></ul>

Join Prof. Fei Chen as he explores advanced bimanual manipulation and teleoperation techniques shaping the future of intelligent human-like robots in this expert-led robotics session.

Bimanual Manipulation and Teleoperation of the CURI Human-like Robot

'Link-bots' can move, explore, cooperate without sensing or computation

Robots made of linked particle chains

<h2 dir="ltr" id="isPasted">Introduction</h2>Service robotics is a fast-growing field, with intelligent, adaptive, and autonomous systems becoming increasingly prevalent across industries. As these robots become more sophisticated, engineers face more complex challenges when it comes to achieving real-time perception, precise motion control, and environmental awareness. TDK Corporation, a leading Japanese electronics company, has developed a comprehensive sensor ecosystem to address these challenges. The result is a modular and scalable platform for next-generation robots.TDK’s sensor ecosystem is integrated with the ROS 1 and ROS 2 framework, and is a unified solution for robotic developers. Drawing inspiration from the sensory capabilities often imagined in anime and manga, this technology brings us closer to creating robots that are highly perceptive and responsive. By combining cutting-edge sensor technologies and flexible algorithms, TDK is leading the way for a new era of intelligent robotics that can interact with humans and their environment. &nbsp;<h2 dir="ltr">The TDK Sensor Ecosystem: A Comprehensive Robotics Solution</h2>The core of TDK’s ecosystem is a collection of modular sensors. Each sensor is engineered to perform a specific function, yet they are designed to work together to improve robot performance.Motion sensing is handled by 6-axis and 9-axis MEMS-based Inertial Measurement Units (IMUs), like the ICM-42688P. These units provide the high-precision tracking necessary for balance, navigation, and posture correction. Environmental factors are monitored through gas, humidity, and barometric pressure sensors. For example, the IPC-10111 capacitive barometric pressure sensor detects changes in air quality and altitude, which is particularly valuable in industrial contexts.Robots also gain a sense of touch and proximity through capacitive and piezoelectric touch sensors. These components allow machines to detect objects and interact safely with humans by responding with tactile feedback that makes interactions more intuitive. Additionally, a layer of acoustic and spatial awareness is provided by digital microphones (like the ICS-43434) and ultrasonic Time-of-Flight sensors (including CH101 and CH201). These components enable functions like obstacle avoidance and voice interaction.1This comprehensive sensor suite enables flexible hardware configurations, which allows developers to customize robots for specific tasks. The plug-and-play integration with ROS also reduces development time and improves cross-compatibility between components. TDK's multi-sensor fusion techniques merge data from different types of sensors to give robots a holistic perception of their environment, like the multi-sensory awareness often portrayed in the futuristic robots of pop culture.<h2 dir="ltr">Motion Sensing and Adaptive Locomotion</h2>TDK’s IMUs play a very important role in robotic motion control by enabling precise movement and posture management. These sensors allow robots to move smoothly, maintain balance, and even prevent falls, particularly in humanoid designs.When IMUs are integrated with other sensing technology, like LiDAR and vision-based SLAM (Simultaneous Localization and Mapping), it improves robotic positioning and mapping. This sensor fusion approach allows robots to navigate complex environments with greater accuracy and adaptability.Robotic systems benefit from advanced algorithms like Kalman filtering, which compensates for sensor drift and ensures fluid transitions between movements. In AI-enhanced applications, IMUs also support predictive trajectory planning, so that robots can adapt to chaning environments in real time. These capabilities are demonstrated in robots like ANYmal, developed by ANYbotics (a TDK Ventures portfolio company), which uses TDK sensor tech to operate autonomously in industrial settings and cluttered environments.2These capabilities are especially apparent in robots like ANYmal, developed by ANYbotics, a company backed by TDK Ventures. ANYmal demonstrates advanced locomotion and navigation in industrial environments and is a real-world application for TDK’s sensor technologies.<h2 dir="ltr">Environmental Awareness: Enhancing Robot Sensing Capabilities</h2>TDK’s environmental sensors also have real-world applications. These sensors significantly enhance robots’ ability to detect and interact with their surroundings and operate safely around humans. TDK’s environmental sensors support this by providing real-time data on air quality, humidity and atmospheric pressure:<ul><li dir="ltr">Gas Sensors&nbsp;— The TCE-11101 MEMS-based CO2 gas sensor detects harmful substances with high accuracy, which is important in industrial and healthcare environments.</li><li dir="ltr">Humidity Sensors — TDK's humidity sensors, featuring ±5% RH accuracy, enable adaptive climate control in smart homes and workplaces.</li><li dir="ltr">Pressure Sensors — The ICP-10111 capacitive barometric pressure sensor, integrated into TDK's RoboKit1, allows for altitude detection and atmospheric pressure monitoring.</li></ul>Beyond atmospheric sensing, TDK sensors enhance object detection and give robots a reliable, non-contact method for navigation and interaction:<ul><li dir="ltr">Object Detection — TDK's SmartSonic ultrasonic Time-of-Flight (ToF) sensors, such as the ICU-20201, provide non-contact object detection and navigation up to 5 meters in any lighting condition.</li><li dir="ltr">Voice Detection&nbsp;— The ICS-43434 digital I²S microphones enhance spatial awareness in voice-activated robots.3</li></ul>With these advanced sensing capabilities, robots can better adapt to their environment and interact with humans in a more intuitive way.<h2 dir="ltr">Intelligent Human-Robot Interaction with TDK’s Sensor Fusion</h2>Human-machine interaction is more engaging with sensors that understand gestures, proximity, and touch. TDK’s sensor fusion technology also facilitates more natural and intuitive robot-human interactions. Touch and proximity sensing is especially relevant for consumer-facing robotics, where intuitive feedback helps build using trust:<ul><li dir="ltr">Haptic Touch Sensors — TDK's piezoelectric actuators provide high-definition haptic feedback for enhanced touch interactions in robotic applications.</li><li dir="ltr">Gesture-Based Control — The CH101-00ABR ultrasonic ToF sensor enables gesture-based control and user input recognition with a configurable 180° field of view.</li></ul>Through TDK’s RobotKit 1 development platform, integrated with ROS and TDK’s adaptive learning models, robots can tailor their responses over time, learning from repeated interactions. This personalized behavior improves functionality and provides more dynamic human-machine interaction:<ul><li dir="ltr">Holistic Awareness — The ICM-42688-P 6-axis IMU, combined with other sensors in the RoboKit1, provides real-time situational awareness.</li><li dir="ltr">Adaptive Learning&nbsp;— TDK's RoboKit software, integrated with ROS, enables personalized interactions using adaptive learning models.</li><li dir="ltr">Social Applications&nbsp;— The combination of IMUs, microphones, and ToF sensors in the RoboKit1 platform enhances social robotics applications.</li></ul>These advancements bring us closer to the kind of seamless human-robot interactions often portrayed in science fiction, where robots can understand gestures, emotions, and intentions with remarkable accuracy.<h2 dir="ltr">ROS 1 and ROS 2: The Software Framework Unifying TDK’s Sensor Ecosystem</h2>The Robot Operating System (ROS) is a critical standardized platform for robotic perception, planning, and control. TDK’s ROS 1 and ROS 2 extension includes pre-configured sensor libraries and real-time data synchronization, streamlining development and performance.This software foundation supports critical functions like real-time sensor processing, AI-enhanced decision-making, and cross-platform compatibility. Developers can use the same ecosystem to build robots for very different applications, from smart retail kiosks to autonomous healthcare aides, without having to reinvent the platform for each use case.<h2 dir="ltr">Real-World Applications of TDK’s Robotics Sensor Platform</h2>As highlighted throughout this series, TDK’s sensor platform is already making a big impact in the real world. Its technologies are used in robotics systems for healthcare, retail, personal assistance, and more. The investment by TDK Ventures in companies like ANYbotics further underlines the company’s commitment to driving innovation in robotics.By enabling smarter, safer, and more adaptable machines, TDK is solving today’s engineering problems and laying the groundwork for human-robot coexistence.<h2 dir="ltr">Conclusion</h2>TDK's sensor ecosystem is more than just a set of components—it's a modular, scalable, and AI-driven solution that is already shaping real-world robotics. As innovation continues and ethical considerations evolve, we are closer than ever to a future where intelligent, responsive robots integrate seamlessly into our daily lives—much like the imagined robotic companions of anime and manga.Learn more about TDK’s innovative advancements at <a href="https://www.tdk.com/en/index.html">https://www.tdk.com/en/index.html</a>.<hr>Resources:<ol><li>Qualcomm Robotics RB3 Development Kit. Accessed from <a href="https://invensense.tdk.com/technology/robotics/" id="isPasted">https://invensense.tdk.com/technology/robotics/</a></li><li>TDK IMU Product Page. Accessed from <a href="https://product.tdk.com/en/products/sensor/mortion-inertial/imu/index.html" target="_blank" rel="noopener noreferrer">https://product.tdk.com/en/products/sensor/mortion-inertial/imu/index.html</a></li><li>“Sensor solutions that enable advanced control of service robots”. Accessed from <a href="https://product.tdk.com/en/techlibrary/solutionguide/sensor-for-service-robots.html" id="isPasted">https://product.tdk.com/en/techlibrary/solutionguide/sensor-for-service-robots.html</a></li></ol>

This article is part 4 of a 4-part article series on TDK's advanced robotics.

A Platform for Smarter Robots: How TDK's Sensor Ecosystem Powers Next-Gen Robotics

As a doctoral student at ETH, Moritz Geilinger was immersed in theoretical considerations about the planning and controlling of robot movements when a very practical request from Swiss Post landed on his desk. The question was: which tasks in the package handling process can be automated?Together with his colleague Simon Huber, who was writing his doctoral thesis next door, Geilinger examined Swiss Post’s sorting systems. The two researchers were astonished to find that although the packages were automatically sorted on the conveyor belts, they were unloaded and reloaded onto the lorries by hand. This is partly due to the initial costs, says Geilinger: “Automation systems are set up individually for specific tasks, which takes a lot of time and resources.” A more flexible system was therefore needed in order to achieve additional automation.<h2>Making robots smarter</h2>That is precisely the work of the researchers in Professor Stelian Coros’s Computational Robotics Lab – namely, making robots smarter and more adaptive. The group was founded in 2017 with the two future young entrepreneurs as its first doctoral students. “The special thing about the lab is that we all came to robotics from computer science,” says Geilinger.He himself developed what’s known as a differentiable physics simulation. “This allows us to predict what happens after a specific action, such as how the position of objects changes when the robot approaches them,” he explains. In this context, differentiable means that the simulation shows what would happen if the position of another object were moved. Huber, for his part, worked on a technique known as differentiable collision avoidance.Armed with this knowledge, the two researchers jointly developed an algorithm that allows robots to pack objects of different sizes and shapes as tightly as possible together, allowing them to be transported cost-effectively from one place to another.<iframe width="640" height="360" src="https://www.youtube.com/embed/mFihv9M3pL0?&amp;wmode=opaque&amp;rel=0&amp;enablejsapi=1&amp;origin=https://www.wevolver.com" frameborder="0"></iframe><h2 id="isPasted">Enabling robots to work together</h2>What makes this solution so unique? “We can get a robot ready for defined tasks within minutes,” says Geilinger. According to the researcher, the software also allows the robots to work in coordination with one another, for example to lift a package between three of them if necessary. This removes the need to set up expensive new infrastructure for new tasks. Instead of larger and heavier robots, several smaller ones are used instead.“The request from Swiss Post was a real stroke of luck for us,” says Geilinger. This practical task served as the trigger for founding Flink Robotics as a spin-off in 2023. Geilinger assumed the role of CEO, and his colleague Simon Huber that of CTO. Their professor, Stelian Coros, was the third person to join the company, in the capacity of an adviser.A year later, the young entrepreneurs had already signed up six employees and received the Venture Kick Award, which comes with prize money of CHF 150,000. What was the secret to their successful launch?<h2>A team of bright minds</h2>“It might sound a bit overconfident, but we’re a team of bright minds whose proximity to research means we deal with state-of-the-art technology and can solve practically any problem in the field of robotics,” says Geilinger. “At the same time, it’s important for us to be very close to the problem, or rather to the end customer, in order to find out why certain tasks haven’t been automated yet.” <img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1743085152879-image.imageformat.1286.1867601149.jpg" style="width:100%px" class="fr-fic fr-dib" />“Thanks to our proximity to research, our team can solve practically any problem in the field of robotics,” says Flink CEO Moritz Geilinger (back row, third from right).  (Image: Nicole Davidson / ETH Zurich)Flink Robotics is therefore neither a technology company that is building an all-capable robot nor a system integrator that is simply solving a specific problem. “We try to focus on a concrete application for which there is high demand, and then refine the general technology while resolving the use case,” Geilinger explains.<h2>Countless potential customers</h2>Regarding the example of loading packages at a distribution hub, he says: “We’re currently seeing a great deal of interest – not only from Swiss Post but also from Österreichische Post and DHL.” As soon as the technical problem is resolved, he adds, there will be an extremely scalable business model to hand – particularly with a view to the 13 variously sized Swiss Post sorting offices and the over 30 DHL distribution centres in Germany.Another interesting area of business is the final packaging of products. “Ultimately, products are often still packed into cardboard boxes and loaded onto pallets by hand,” says Geilinger. The young entrepreneurs have already visited chocolate manufacturers and producers of face creams that manufacture hundreds of different products, operating perhaps six production lines in a single factory. With classical robots, this would require repeated reprogramming and adaptation to each product.<h2>Managing company growth</h2>In other words, the young entrepreneurs have no shortage of potential customers. The big challenge for them at present is to juggle all of these different tasks. They are continuing to develop their algorithms for existing customers and to carry out test installations on site.  At the same time, new employees need to be introduced. “Time and time again, we have to turn customers away because we don’t have the capacity to initiate further developments,” says Geilinger.So far, Flink Robotics has been very fortunate with its staff given that computer science and robotics graduates are also particularly sought-after by global companies in Zurich. “We attract very specific personalities who feel comfortable in a small, agile team, who like to build things themselves and who feel it’s important for their own work to have a directly visible impact,” says the young CEO.At Flink Robotics, the staff don’t just carry out simulations in the office. “Every few weeks, we load the robots into the car and drive to a distribution hub, where we see what we’ve actually programmed and how much better the robot has become,” says Geilinger enthusiastically.It is there, on site at the customers’ premises, that the long-term success of Flink Robotics will be decided. “The big challenge for venture companies in robotics is to actually put a new technology into practice rather than just demonstrating it in the office,” says Geilinger.<a></a><h2>Looking ahead with confidence</h2>The researcher is confident about the future: “The objective we’ve set ourselves is for our software to be in use in production in a year’s time.” That will probably involve growing the team further. “Carefully, as before,” says Geilinger. At the same time, Flink Robotics also wants to expand to other business areas, such as those relating to final packaging.The aim is for Flink Robotics to be a well-known name in the package handling industry and other relevant industries across Europe within three years. “Flink Robotics will stand for a new breed of automation that is less fixed, less preprogrammed, and much more adaptive and flexible.”

Moritz Geilinger (left) and Simon Huber, the founders of Flink Robotics in their laboratory.  (Image: Nicole Davidson / ETH Zurich)

The ETH spin-off Flink Robotics wants to revolutionize the handling of packages. Its founders Moritz Geilinger and Simon Huber have developed software that allows robots to work together and quickly take on new tasks.

Robotics Software

Engineers Wiki. Most Asked Questions.

IoT Security Solutions: Threat Landscape, Architecture, and Engineering Best Practices

IoT Gateway Architecture: Edge vs. Cloud, Protocol Translation, and Deployment Patterns...

TPU vs GPU: A Comprehensive Technical Comparison

TPU vs GPU: Comprehensive Technical Comparison

AI Infrastructure Engineering: Building Enterprise-Scale Machine Learning Systems...

What is SCADA?: The Basics you Need to Know

NPU vs TPU: Understanding the Key Differences in AI Hardware Accelerators

ORGANIZATIONS. SHAPING THE INDUSTRY.

Nordic Semiconductor

EPFL

High Tech Campus Eindhoven

ETH Zurich

University of Michigan

TAGGED WITH embedded software