Machine Learning is progressing at a rapid rate with new Deep Learning algorithms able to solve historically difficult problems using data-driven design principles. This is especially exciting in IoT where the rapid increase in connected devices has led to an explosion in the amount of data being generated at the edge. These new algorithms are playing a critical role in advancing the next phase of the IoT revolution.
There are many reasons why this approach is being embraced in many markets and areas such as smart cities, smart homes, the Industrial Internet of Things (IIoT), wearables, and more. According to one market research firm’s projections, the AIoT market is expected to increase from US$5.1 billion in 2019 to US$16.2 billion by 2024, growing at a CAGR of 26 percent1).
A recently published Price Waterhouse Cooper (PwC) article shows the drivers of IoT growth and benefits of artificial intelligence. Decreasing costs are among the benefits of AIoT. However, increases in device proliferation and venture capital (VC) spending as well as convergence of information technology (IT) and operational technology (OT) and big data and the cloud/fog are also occurring.
With decreasing cost in several areas as a benefit, it is no surprise that many systems developers are interested in taking advantage of these combined capabilities in their next designs. To accelerate the development of differentiated AIoT products, Infineon Technologies has released the ModusToolbox™ Machine Learning (ML). This design tool enables deep learning-based workloads on Infineon’s PSoC™ microcontrollers (MCUs). ModusToolbox™ ML is a new feature in ModusToolbox™ Software and Tools that provides middleware, software libraries and special tools for designers to evaluate and deploy deep learning-based ML models.
AIoT addresses system barriers in IoT. The original design concept of simply moving all of the data generated on the edge to the cloud for analysis and machine learning has run into three fundamental barriers: privacy, reliability and latency. To reduce these barriers, system designers have changed the location of the ML algorithms that typically have run on the cloud at the edge. A great example to think about are voice-based smart assistants.
Firstly, when you interact with an assistant, the time it takes to make a round-trip to get the answer is generally a poor user experience since that is not the natural way for humans to interact. Secondly, reliability and bandwidth of internet connection is also critical, especially when these assistants run on wearable devices such as smart watches and you don’t always have a perfect, reliable connection to the cloud. Thirdly, with the proliferation of these assistants everywhere, privacy is always top-of-mind and trusting service providers with sensitive voice data is always a challenge. Running these algorithms efficiently on edge eliminates these barriers and allows AIoT products to scale much more rapidly.
ModusToolbox™ ML allows developers to use their preferred deep learning framework, such as TensorFlow, and deploy them directly to PSoC™ MCUs. In addition, the tooling helps designers by optimizing the model for embedded platforms by reducing the size and complexity using a variety of techniques such as quantization.
ModusToolbox™ considerably simplifies the development of IoT products which use Wi-Fi and Bluetooth/Bluetooth Low Energy IoT products in combination with RTOS system microcontrollers, such as those from the PSoC™ family. Developers can use the integrated middleware and code examples to easily connect their IoT products with leading cloud software platforms or proprietary cloud services.
To reduce AIoT development time, ModusToolbox™ also contains solutions to support popular ecosystems and cloud management tools like Pelion Cloud Management and Mbed™ OS, Amazon Web Services (AWS) IoT and FreeRTOS™ SDK as well as Infineon AnyCloud IoT. In addition, it provides specific tools like the low-power assistant, multi-radio smart coexistence, secure authentication, and over-the-air updates to reduce the time and expense required when launching high-value, high-quality products onto the market. This provides flexible, easy-to-use tools and solutions for creating smart devices of the future.
One of the other key features this toolset brings is helping system designers visualize how these optimization techniques impact model performance, so they can make the right trade-offs between performance vs the size/complexity of running the model efficiently on a PSoC™ MCU.
To help system designers get started quickly, code examples and IoT-focused development kits are provided to have a smooth developer experience that reduces the complexities system developers face when developing AIoT applications. These typically require a seamless Machine Learning workload integration, along with compute, connectivity, and cloud domains. ModusToolbox™ ML can address these aspects by giving developers the ability to simplify their design and drop this functionality into any existing cloud or connectivity examples today.
Figure 1. Iterations for optimization are a key part of ModusToolbox ML.
ModusToolbox™ ML delivers an unmatched developer experience that reduces the complexities system developers face when developing AIoT applications. These applications typically require a seamless Machine Learning workload integration, along with compute, connectivity, and cloud domains that ModusToolbox™ ML can address.
ModusToolbox™ is available for download here. In addition, code examples are shown in the ModusToolbox™ GitHub Repository. To access online documentation, online videos, and regular live developer trainings, join the Cypress Developer Community.
Click here for more information about Infineon’s machine learning solutions.
Click here for more information about Infineon IoT solutions.
