Azure Percept, a new family of AI-enabled edge hardware and services, was the most significant IoT announcement at Microsoft Ignite 2021. Essentially a logic layer on top of Azure IoT, Percept streamlines AI model deployment for low-power edge devices.
But it’s more than that. Taking a step back reveals a bigger trend – the emergence of practical IIoT platforms that can dramatically reduce solution development time and cost.
Why so many enterprise IoT projects fail
I’ve often said that developing industrial IoT solutions is like building Ikea furniture with some parts missing and without instructions. Roanne Sones confirmed this view in her “AI to the Edge” Ignite presentation. Microsoft spoke with over 500 customers about recent IoT projects and found that “less than 20% percent of the proofs of concept that were developed ever moved to production deployment.” 100% of the surveyed customers complained that there are no end-to-end IoT blueprints to follow (no instructions). 90% of them also indicated that matchmaking (finding the missing parts) was a significant complication. Developers have to find, modify and integrate hardware and software components from multiple vendors, stitch them into enterprise services, ensure enterprise-grade security and maintain the whole stack for many years–all without blueprints to follow. Hence, the 80% failure rate.
“Solutions” are not the solution
Today, most IoT suppliers and integrators aim to reduce complexity by offering pre-integrated vertical industry solutions. The sales pitch is, “Come into my world, and I’ll take care of everything for you.” This approach works great when customer business requirements closely align with pre-integrated capabilities. Unfortunately, this is rarely the case. Adapting pre-integrated components to each customer’s unique physical infrastructure, operational processes, workflows and business systems usually requires modifying many pre-integrated components. The cascading effects of these intrusive alterations reduce the advantages of pre-integration. “One size fits all” solutions seldom fit customer requirements right out of the box and customizing pre-integrated components is costly and risky.
Horizontal IoT platforms
IoT platforms combine plug-and-play hardware and software components with rapid domain-specific and customer-specific customization. Although IoT platforms from AWS, Google, Hitachi, Microsoft, Pelion, PTC, Software AG and others have been growing for years, none have solved the entire problem. Consequently, the failure rate for enterprise IoT deployments remains stuck at 80%. Transitioning from 80% failure to 80% success requires horizontal IoT platforms that completely nail down these three key characteristics:
- Works right out of the box – basic horizontal functionality with little or no coding
- Easy customization – the ability to develop with mainstream IT tools and expertise
- Robust ecosystem – a good selection of plug-and-play components for prototyping and POC
Horizontal platforms enable IoT projects to begin with end-to-end, select off-the-shelf components that “just work,” and incrementally customize those components as needed using familiar, IT-friendly development techniques and tools. Ideally, enterprises can add business-specific data handling, logic, and analytical functions without worrying about complicated IoT “plumbing” or using specialized embedded languages and tools. Through combinations of acquisitions and internal development, all major IoT platforms are heading in this direction. Last week’s Azure IoT and Percept announcements tell a compelling story about Microsoft’s progress towards reducing the enterprise IoT project failure rate.
Azure IoT and Percept
After years of steady progress, Azure IoT is getting closer to delivering the IoT platform trifecta – end-to-end functionality a few minutes after unboxing, a good selection of compatible components and easy customization. Percept, a logic layer on top of Azure IoT, significantly advances all three features and is a proof-point for Microsoft’s consistent Azure IoT product vision. Let’s take a closer look at each element of the trifecta.
Microsoft Percept module.Microsoft
Works “right out of the box”
Percept’s primary design goal is to simplify AI model deployment on accelerated IoT devices. Percept Studio, the heart of the offering, integrates Azure AI, Azure Cognitive Services, Azure Machine Learning, Azure Live Video Analytics and Azure IoT management services. The whole stack works end-to-end, from sensors to cloud, requiring no coding when used with the Azure Percept Development Kit. At launch, the kit consists of two hardware modules: an AI-accelerated edge computing device (NXP iMX8m, quad A57) and a Percept Vision camera (Movidus Myriad X MA2025 VPU, 0.7 TOPS). The Vision module is also available separately, as is the Percept Audio module with a built-in array of four microphones.
Unlike other IoT prototyping hardware, the Percept Development Kit is compatible with the 80/20 1010-series industrial building system, often characterized as “The Industrial Erector Set.” Customers can use Percept modules as-is in real-world industrial environments for proofs-of-concept trials and, in some cases, production deployment. Traditional evaluation and prototype boards and boxes designed for use in development labs require custom hardware for field use. In contrast, developers can bolt Percept Development Kit hardware directly into industrial environments, eliminating the need to develop new hardware enclosures and boards.
Microsoft says customers can get pre-built AI applications up and running on the AI-accelerated Percept Development Kit in as little as 10 minutes and create customized functional prototypes in under 30 minutes. I intend to verify this ambitious claim in my lab in the next few weeks. I’ll also test the limits of rapid customization within the constraints of the integrated capabilities.
Even though Percept fires up basic end-to-end functionality in a few minutes, every IIoT project needs some amount of domain-specific and enterprise-specific customization to meet business requirements. Customization complexity is a leading cause of IIoT project failure, so the big question enterprises should ask when evaluating any horizontal IIoT platforms is, “How easy is it to tailor the whole end-to-end stack to my business requirements?”
Horizontal IoT platforms simplify customization by abstracting the esoteric intricacies of embedded programming so that IT personnel can use familiar models, languages, tools and DevOps practices across the whole stack, from devices to clouds. Consistent architecture makes it much easier for compliant devices to seamlessly plug-and-play with high-level, IT-friendly services and development tools, thereby simplifying customization. Here are some of the services that comprise the Azure IoT ecosystem and the device platforms that are compatible with them:
- Azure IoT Hub – Connect, authenticate, provision, update and manage devices
- Azure IoT Central – Connect, build and deploy apps
- Azure IoT Edge – Deploy cloud workloads, AI modules and custom services via standard containers
- Azure Digital Twins – Build digital models of real-world things and processes
- Azure ML – Build, train and deploy machine learning models
- Azure Percept – Hardware and services for deploying AI on accelerated devices, built on Azure Cognitive Services, Azure ML, Azure Live Video Analytics and other services
- Azure Sphere – Hardware and services for secure edge devices
- Azure RTOS – Embedded OS compatible with Azure IoT Services (based on ThreadX)
Azure IoT services simplify device programming, allowing developers to write code that adds business value rather than customize operating systems, network stacks, security and other IoT plumbing. However, in many IIoT use-cases, the available set of compatible IoT device platforms does not support essential features that industrial IoT applications require. Developers then have to figure out how to connect new devices to Azure IoT services, hide the new devices behind gateways such as a Sphere Guardian module or perhaps use another set of services. Today, only a handful of IIoT devices are compatible with Percept (or Sphere), so this is a common problem. However, as the Azure IoT device platforms mature and ecosystems grow, a wider variety of devices makes customization easier. Let’s dig into the ecosystem growth issue a bit more.
Percept’s industry take-up ultimately depends on the size and flexibility of its ecosystem. Only the three hardware devices listed above are available at launch: Dev kit, Vision and Audio. Vision and audio applications are universally applicable in industrial applications, so the initial product offering makes sense. However, industrial IoT applications need a much wider variety of sensors. Some vision applications require specialized cameras such as IR, 3D, high resolution, fast motion, low light, small size and low power. Other IIoT applications need a broad selection of specialized acoustic sensors–seismic, for example. Hundreds of sensor types other than vision and sound are also applicable in AI-based analysis. Azure IoT customers need a large and expanding ecosystem of compatible edge devices.
The Percept partner ecosystem currently includes Aaeon, Ability, Advantech, Arm, Arrow, Aruba, Asus, Avnet, Intel, Lenovo, NXP, TKE, and Sharp/NEC. Asus is the OEM for the three initial Percept modules. Given the excitement around the Percept launch, it’s a good bet that we’ll see additional devices shortly.
Percept is Microsoft’s second Azure IoT hardware device ecosystem. In 2018, Microsoft announced Azure Sphere, a secure device platform built around low-power chips with a prescribed architecture, including a Microsoft-designed security subsystem (Pluton). The MediaTek MT3620 was the first Sphere chip, with development kits from Avnet, Seeed and qiio. At Ignite last week, NXP announced a new Sphere processor for preview late this year, so the Sphere ecosystem is growing. Although Sphere is technically solid, it hasn’t expanded fast enough to make a dent in the “80% fail rate” described above. By providing evidence for the benefits of IoT platform-enabled devices, Percept might have the side effect of increasing interest in Sphere.
Reducing IoT device variability by prescribing and certifying device architecture improves security, compatibility, performance and user experience. The benefits are undeniable, but fewer variables mean less flexibility for chips and devices to address specific business problems and connect with other services. So, for chipmakers and OEMs, supporting Sphere or Percept is a trade-off between the benefits of seamless ecosystem integration and the cost of reduced flexibility. However, the calculus for industrial IoT platform decisions is rapidly changing, from minimizing device cost to increasing project success rates and maximizing overall ROI. In this new light, relatively expensive edge devices are very cheap and integration costs are unacceptably high. The Percept ecosystem is poised to grow, and the Sphere ecosystem might, too. I’ll be watching closely to see how fast these ecosystems expand.
Emerging horizontal IIoT platforms such as Azure IoT have the potential to reduce solution development time and cost. According to Microsoft, customers using Azure IoT with Percept can run pre-built AI applications end-to-end in as little as 10 minutes and build customized functional prototypes in under 30 minutes. These impressive numbers result from standardizing key design elements of each IoT component. Reducing component variability simplifies integration, but customers must choose from a limited number of certified products. Therefore, large-scale adoption of Percept, Azure IoT and other horizontal IoT platforms requires a robust and rapidly expanding set of compatible devices with flexible programming models. It’ll take a few years, but horizontal platforms like Azure IoT can potentially turn the 80% IIoT project failure rate into an 80% success rate.
Disclosure: Moor Insights & Strategy, like all research and analyst firms, provides or has provided paid research, analysis, advising, or consulting to many high-tech companies in the industry, including Microsoft. The author holds no investment positions with any of the companies cited in this article.