Every year, after the CES dust settles, I take a step back and rank the most disruptive IoT-related technology trends. This year’s list signals the approach of a significant inflection point in the IoT growth curve.
These three developments accelerate IoT’s ongoing transformation from a hodge-podge of custom solutions into a growth industry built on platforms that plug-and-play within a software continuum extending from cloud services to edge devices. Plug-and-play platforms defragmented PCs and smartphones in the 2000s. The same thing is happening with IoT platforms, but the process is slower and more complicated because of the extreme diversity of devices.
Let’s look at each trend from a business perspective to see how smarter devices, standard device platforms, and open networks remove growth barriers and make the IoT more investable across all market segments.
Typical IoT devices have surprisingly little processing power because they act as “peripherals” to services running in the cloud or on-premises. These services ingest and analyze sensor data, trigger appropriate actions, send commands back down to devices, and interface with other services. A new generation of smarter edge platforms capable of autonomously running complex application software is now disrupting the peripheral model, which has been the norm for over 20 years.
Autonomous on-device analysis decreases response times while improving reliability and privacy for applications such as anomaly detection, predictive maintenance, wake-word recognition, image classification, and gesture control. Smarter devices also simplify development, shorten TTM, and reduce operating costs by enabling mainstream software development methods that do not require specialized embedded programming techniques. This year, four disruptive technologies combine to accelerate the edge intelligence trend.
Here’s how IoT solution developers leverage these technologies to build increasingly intelligent edge applications.
Larger processors enable on-device applications to analyze and act on sensor data locally instead of sending it to the cloud for processing. Bigger chips also reduce device development costs and TTM by supporting advanced programming methods and tools such as cloud-native development and containerized software. Development productivity also increases as we move away from low-level embedded techniques and embrace modern, cloud-native software engineering methods.
Neural processors and accelerators (NPUs) enable edge devices to run surprisingly large ML inference applications. Until recently, math-heavy AI algorithms were practical only on large processors. New ML-accelerated chips enable small IoT devices, including battery-powered ones, to run ML inference applications locally without depending on cloud services. Welcome to the brave new world with AI integrated into the things all around us.
ML development on small devices has always been challenging because mainstream cloud-native machine learning code, models, and tools are not optimized for small platforms. In 2019, the tinyML Foundation started a new way of thinking about on-device ML by sharing ideas and experiences for ML applications on low-power devices. ML developers found ways to target small (down to 100kB and below), low-power (milliwatt) devices using familiar languages, tools, libraries, and workflows. Leveraging these concepts, Edge Impulse and other startups provide cloud-based development environments and run-time libraries that directly support many small IoT platforms, with or without NPUs. Cloud-native ML development brings AI functionality to low-power, always-on devices at the edge of the network.
IoT-specific SoCs are already available that integrate many of the function blocks needed to build complete IoT devices. This year, we see a new wave of highly integrated chips with all the functions typically needed for many IoT devices, including wireless radio subsystems (Wi-Fi, 802.15.4, Bluetooth), math accelerators, I/O, more RAM, more flash, and built-in security. New SoCs such as the Silicon Labs MG24 and NXP IW612 are good examples. Arm takes a higher-level approach, providing its ecosystem partners with chip design guidance through initiatives such as Arm Total Solutions for IoT, Arm System Ready, and Corstone. Arm’s Project Cassini and Project Centauri offer consistent system and security software, further accelerating chip development.
When I tell the “edge intelligence” story, developers often ask about device cost. “Sounds great, but how do I justify more expensive devices?” Although complex chips are more expensive than simple ones, the benefits of new application features, efficient development, built-in security, faster TTM, and longer product life offset the higher cost. Strong business cases make edge intelligence the new “north star” driving IoT innovation across all market segments.
IoT Device platform convergence
Developing code on small IoT devices is often like returning to the 1990s or even the 1980s. Embedded device OSes and programming techniques evolved on small, highly constrained chips designed for dedicated purposes. So, the computational performance, memory, storage, and system features in typical embedded SoCs are the bare minimum required to run a specific application. That’s not usually enough horsepower to support the mainstream OSes, software tools, and DevOps commonly used on general-purpose platforms. Embedded software remains stuck in the past with customized OSes, DIY system configurations, waterfall development, and a dearth of re-usable software components. Consequently, IoT developers spend too much time working on system-level code instead of focusing on applications.
Agile DevOps, standard Linux distros, cloud-native development, microservices, containers, and serverless code are already available on larger edge platforms with powerful processors that can run off-the-shelf, general-purpose OS distributions. These modern tools and techniques dramatically improve developer productivity and solution quality while reducing development cost and risk. The business benefits are predictable because we’ve seen similar scenarios before. Windows, Linux, IOS, and Android defragmented PC and smartphone operating systems, creating large-scale, multivendor software and hardware markets. Similar defragmentation is already happening on embedded MPUs (microprocessors), but MCUs (microcontrollers) are on a slower path. Let’s cover MPUs first.
MPUs for embedded applications are scaled-down versions of PC and smartphone processors. MPUs usually run embedded versions of Linux, but off-the-shelf distributions are too big. Squeezing Linux down to an appropriate size requires creating a unique OS configuration for each solution with only the packages needed for the target platform and applications. OS building, customization, testing, debugging, updating, and long-term support are very costly and don’t add any solution value or product differentiation. Also, there’s no ISV market or app store for IoT devices because each IoT solution has a unique software environment. IoT solution development is expensive, slow, and risky primarily because device software engineers have to build custom OSes and write system code instead of just focusing on applications.
IoT platform OS customization isn’t going away for the foreseeable future, but new automated build systems that simplify Linux customization are now available. These new tools (1) do not require deep Linux expertise, (2) are not hardware dependent, and (3) allow continuous builds with long-term support. Two open-source build systems – Buildroot and Yocto – have been around for many years. While Buildroot is simple and fast, Yocto supports continuous integration and uses a layer model for modular customization. But both of these build systems require considerable Linux expertise, have a steep learning curve, and don’t provide an IoT-friendly update service.
This situation opens up some lucrative business opportunities. For instance, Foundries.io is a cloud service that leverages Yocto to simplify building, testing, securing, deploying, and maintaining custom Linux platforms. Application developers without deep Linux skills can use FoundriesFactory to build maintainable, updateable OSes from standard distributions. The tools are hardware independent with support for Arm, x86, and RISC-V processors from multiple chip companies – no hardware lock-in. And, there’s an OSTree-based versioning system for incremental updates with OTA cloud delivery – no re-imaging of devices in the field. Cloud-native developers can start building component-based applications right away using Docker Compose – no need for system-level programming. FoundriesFactory also integrates with multiple device management systems and cloud frameworks. The business model makes sense, too, charging a fixed subscription fee that’s much less than the cost of one Linux expert – and no per-unit fees. The combination of automated Linux builds, managed security, incremental OTA updates, and long-term support allows solution developers to focus on adding value and avoid undifferentiated system programming.
MCUs optimized for dedicated, low-power applications are simpler and more architecturally diverse than MPUs. MCUs don’t have the advanced features needed to run Linux, such as virtual memory support, so using general-purpose Linux-based OSes is not an option. Instead, there are dozens of specialized MCU OSes with compelling attributes for specific platforms and use-cases.
MCU OS defragmentation is already underway, and the leading OSes have support from hyperscale cloud frameworks and chip ecosystems. FreeRTOS (AWS), Azure RTOS/ThreadX (Microsoft), and Zephyr (Linux Foundation, Wind River Systems, Intel, NXP, Google, Meta, Linaro, Qualcomm, and others) are three examples, and there are many more. I expect clear winners to emerge next year due to natural selection. Although we’ll still have many RTOSes for the foreseeable future, cloud-based configuration tools and update services for MCUs can deliver the efficiencies described above for MPUs. However, it’ll take longer because of the greater diversity of hardware and OSes.
Matter revolutionizes consumer IoT
CES 2022 product announcements mark the end of the consumer IoT “connectivity wars.” After 20 years of incompatible radios, proprietary protocols, vertical product silos, annoying gateway (hub) devices, bafflingly confusing device onboarding procedures, and unnecessarily high device costs, the consumer electronics industry is rapidly adopting two new practical, scalable, and open connectivity standards – Matter and Thread.
- Matter – “Lingua franca” for the Internet of Things
- Thread – IP-based mesh network for low power devices
Thread is a low-power mesh network like Zigbee and Z-Wave, but it sends data using internet protocols (IP) compatible with Wi-Fi and Ethernet. Matter defines the messages that travel over those IP networks so that devices from any manufacturer can communicate with any Matter-compatible ecosystem – Alexa, Google Home, or HomeKit, for example. The vision of buying a smart home device from any manufacturer with confidence that it’ll work with your existing network and chosen ecosystem(s) is finally becoming a reality.
Rapid adoption is likely because Amazon, Apple, Google, and over 200 companies are on board and announcing products. Beginning this year, consumers can buy connected products such as door locks, window shades, light switches, thermostats, and cameras that plug-and-play with existing home networks and ecosystems. There’s no vendor lock-in, nothing extra to buy, no complicated hub to configure, and setup is a few clicks on a smartphone app. At CES, dozens of companies announced products supporting Matter and Thread, and hundreds more are on the way. The trend towards Matter and Thread is rapidly becoming a gold rush as influential companies stake claims.
Although Matter and Thread target consumer applications, Thread is already used in industrial verticals, and Matter is influencing other standards organizations to create similar IP-based protocols for commercial domains. IP-BLiS, for instance, aims to align commercial building automation industry alliances around a common IP-based application framework. We’re finally putting the “I” in “IoT.”
Summary: Plug-and-play IoT
IoT growth lags projections because most endpoint devices are constrained, fixed-function platforms with customized embedded software stacks communicating over dedicated, non-interoperable networks. AI-capable SoCs, converged platforms, and interoperable networks address these growth barriers, transforming IoT devices from underpowered embedded gadgets to scalable compute nodes that use modern software tools and DevOps. This fundamental change to IoT economics plays out over several years, but we’ll look back on 2022 as the start of the IoT plug-and-play era.
Note: Moor Insights & Strategy writers and editors may have contributed to this article.