Transforming data at the source of collection minimizes latency and enables optimized processing for time-critical applications.
Artificial intelligence (AI) at the edge of the network is a cornerstone that will influence the future direction of the technology industry. If AI is an engine of change, then semiconductors are the oil driving the new age that is being defined by machine learning (ML), neural networks, 5G connectivity and the advent of blockchain, digital twins and the metaverse.
Despite recent disruptions to the chip industry due to supply chain and, more recently, macroeconomic factors, the confluence of AI and the Internet of Things (IoT) known as AIoT is poised to shift the world from cloud-centric intelligence to a more distributed intelligence architecture.
It is expected that a staggering 73.1 zettabytes of data is expected to be generated by IoT devices in 2025, according to IDC Research. As a result, endpoint data will increase at a CAGR of 85% from 2017 to 2025, driving intelligence from the cloud to the endpoint to run AI/ML workloads within tiny machines (TinyML). Some of the applications that are seeing the most disruption include the development of “voice as a user interface” to improve human-to-machine communication, as well as environmental sensing and predictive analytics and maintenance. Major growth segments include wearables, smart homes, smart cities and intelligent industrial automation.
What are the benefits of embedding intelligence at the endpoint? Many industrial IoT applications operate within environments constrained by memory capacity, limited computing and battery power and sub-optimal connectivity. Moreover, these applications often require real-time responses that may be mission and system critical. Expecting such devices and applications to operate in a cloud-centric intelligence architecture just does not work.
This is where the power of embedding intelligence at the endpoint is evolving from standard industrial IoT implementations to what we are calling AIoT for industrial applications.
Transforming data at the source of collection minimizes latency and enables optimized processing for time-critical applications. Because data is not processed and transported over the network, the security concerns related to transfer and flow of data are greatly minimized. Another advantage is that data handling can be linked with root-of-trust at the endpoint, making the implementation impervious to attacks. Since data processing is handled at or very close to the source, we can fully leverage data gravity and reduce the power consumption associated with turning on radios or moving data through the network.
Our commitment to our customers is to lead the industry in endpoint computing technology with a broad range of MCUs and MPUs. Already this has enabled designers to leverage our ecosystem of IoT and AI/ML building blocks by tapping into a technology ecosystem that features more than 300 building blocks of commercial grade software provided by Renesas’ trusted partners.
Our growing AIoT portfolio also explains our recent acquisition of Reality AI, a new platform powering edge and endpoint AI in industrial IoT applications using Renesas processors. Reality AI automatically searches a wide range of signal-processing transforms and generates custom machine learning models, while retaining traceability in its approach and offering valuable hardware design analytics. The models run on nearly every MCU and MPU core available from Renesas – with new ones added constantly.
This puts an incredibly powerful tool into the hands of designers to help them solve their most difficult problems, because the model development is specifically for non-visual sensing use cases and based on advanced signal processing math and edge deployment. This enables advanced analytics capable of supporting full hardware design and complete frameworks, including data collection, instrumentation, firmware and ML workflows. Other solutions simply generate algorithms and models that often account for only 5% of typical project costs, while ignoring the other 95% of development expenses.
This approach to AIoT design allows developers to reduce unscheduled equipment downtime, improve production efficiencies and perform sophisticated quality assurance tasks that are costly or difficult to replicate in the current testing environment.
In a real-world use case tested under 51 different environmental and load conditions in a three-ton residential HVAC system, Reality AI was able to achieve a greater than 95% accuracy when detecting and distinguishing single fault conditions. The test also detected indoor and outdoor air-flow blockage and charge faults as small as 5% from OEM specifications in both heating and cooling modes.
The convergence of AI and IoT for industrial applications is a megatrend with significant potential. The acquisition of Reality AI unlocks the potential of combining advanced signal processing with AI at the edge and supported by Renesas’ hardware, software, tools and ecosystem to provide all the building blocks you need to unleash your creativity.
Sailesh Chittipeddi is executive vice president and general manager of the IoT and Infrastructure Business Unit at Renesas.