With cloud computing now so ubiquitous, and edge computing so well established, why, then, is endpoint computing becoming so important?
In a word: Latency.
If we're going to make true advancements toward meaningful artificial intelligence (AI) and machine learning (ML) in smart homes, smart health and smart cities, lag and latency is unacceptable.
Efficient AI inference demands efficient endpoints that can infer, pre-process, and filter data in real time. Unlike computing in the cloud or at the edge of the cloud, the endpoint represents, as Renesas CEO Hidetoshi Shibata has said, "the true point of action." For example, a home appliance that can predict when it will need maintenance or a voice-interactive wearable that can warn a user of a possible heart anomaly is more useful if it applies ML algorithms at the endpoint with near-zero latency.
And it is precisely at the endpoint where Renesas microcontrollers (MCUs) shine. AI models, trained on industry-standard networks, are embedded on chip to offer design engineers the performance, bandwidth, and responsiveness to effectively realize and enhance emerging smart applications. Advanced endpoint-compiler software enables customers "to test capabilities purely in the cloud before porting them over to our boards, and then, finally, to implement these seamlessly on our chips," Mr. Shibata told McKinsey analysts recently.
"While also providing the flexibility to change the functionality or algorithms," he said, "Such dynamically configurable hardware architectures let customers enjoy the benefit of hardware-processing speeds."
Indeed, users of smart homes, smart health and smart cities applications cannot or will not wait even a few milliseconds for AI data to be processed. Consequently, the latency inherent in transferring data to the cloud threatens to undermine progress in those consumer-focused areas. This may sound obvious, but few competitive MCUs are designed to enable fast processing.
As ML inference moves to device endpoints, "this integrated AI will be the foundation that powers a complex combination of 'sense' technologies to create smart applications with more natural, 'human-like' communication and interaction," Dr. Sailesh Chittipeddi, president and CEO of Renesas Electronics America, recently wrote in Embedded Computing magazine.
"In addition," wrote Dr. Chittipeddi, "A convergence of advancements around AI accelerators, adaptive and predictive control, and hardware and software for voice and vision open up new user interface capabilities for a wide range of smart devices."
Let's take a look at what those might be in the smart home, health and city segments.
Smart Home Applications
The intelligent home is in many ways a "sensor-rich" application. Smart homes will use sensors to collect and process all manner of data, from environmental information to user activity. With the guiding principle of providing "total convenience," these applications tend to use AI for predictive analytics and for customizing user interaction with the home environment.
For example, a refrigerator will learn over time to adjust its temperature settings. Or the TV will learn to change the audio output based on what's happening in the room at the time -- so voices and music are not deafened during parties or cooking. Voice interfaces can also be used to teach or train devices, as well as to provide information when an appliance is offline.
Smart Health Applications
Smart health applications will be even more demanding when it comes to latency. A pacemaker that can predict when a battery needs to be changed could well save lives. An electrocardiogram (ECG) sensor that can warn a user of a heart anomaly in real time could save lives as well. ML algorithms embedded in an ECG sensor could detect disruptions in heartbeat or rhythm and sound the alarm immediately.
AI algorithms are also being used to predict when an insulin pump is likely to run out of a patient's specific dosage and to anticipate when a dose is needed. Otherwise, the pump might not deliver the right dosage at the right time, putting patients at risk.
Smart City Applications
In a smart city, sensors can collect all manner of information -- from pollution levels to traffic delays -- that could be used to forecast traffic conditions and even determine air quality alerts.
Smart city applications will be equally demanding of low latency. A smart streetlamp that can detect traffic density and adjust its brightness accordingly to prevent traffic accidents could save lives. An AI and ML-infused surveillance camera that can detect alterations in traffic flow, which may mean an accident has occurred and can alert emergency services immediately could also save lives.
In scenarios like city traffic, drivers would benefit from AI alerts that warn of slowdowns ahead. Smart parking garages could use AI algorithms to monitor a car's status and accurately charge or discharge a vehicle depending on a user's preferred levels of privacy or security.
In terms of cities and traffic control, AI algorithms embedded in smart traffic lights could similarly be able to notify drivers, pedestrians and bikers of slowing or stopping traffic ahead. Once again, the information would be relayed in near-zero latency. Smart vehicle routing -- say, to avoid heavy congestion -- may be the most challenging application of all.
Flexibility is Key
As ML algorithms come to dominate smart home, health and city applications, the speed at which data is processed will be more critical than ever. For companies like Renesas, that means delivering the most intelligent and flexible MCUs that optimize data processing speed while also providing the flexibility to change functionality or algorithms.
The same MCU that enables fast processing today may support a completely different algorithm tomorrow. That makes the MCU platform, which includes a flexible CPU, an ideal choice for enabling endpoint ML applications.
Find the right MCU with this easy-to-use MCU selection tool, or download the Renesas MCU guide app and move your designs to the endpoint today!