Storage solutions for AI from Innodisk for edge applications

Innodisk, Eindhoven, 28.09.2021 - Artificial Intelligence (AI) and the Internet of Things (IoT) are merging into what is now called AIoT. Edge Computing shifts processing power to where data is generated and IoT devices collect it. AI reduces latency, enables efficient data processing, and requires innovative solutions in the Edge space. Conditions vary greatly, and the demand for optimized memory solutions for AI Edge applications increases with the rising data volumes. The number of Edge Computing devices continues to grow and is expected to have increased by 226 percent by 2025 compared to today.

Transmitting all collected data to the cloud can lead to long latency times. Even with steadily increasing connection speeds, such as with new 5G technology, the exponentially growing data cannot be processed efficiently. Latency increases, and overall system performance suffers. AI counters this and drives new technological innovations such as optimized urban traffic, improved financial services, and enhanced public safety. KIoT components are used in highly diverse conditions and can be installed anywhere—from vehicles, onboard aircraft, to factories and oil rigs in deserts. This requires manufacturing components that are flexible and adaptable. AI reduces the human factor in decision-making processes. System integrators must therefore ensure high-quality components, as in the event of an accident, the human element is eliminated, potentially complicating clear accountability.

Pure IoT collects data with little or no computation and analysis. All data is sent to the cloud without prior analysis, causing a significant increase in data volume there. However, not all data holds equal importance. For security footage, for example, sending all images with unchanged backgrounds to the cloud is inefficient. Only data with moving persons or objects are relevant. Sending all data for cloud analysis would waste bandwidth that could be used for other applications. AI at the network edge can require high computational power to ensure adequate performance. Standard storage components can meet these demands but are not designed for harsh conditions at this location. For example, a traffic monitoring system on the street is exposed to temperature fluctuations between day and night, summer and winter; vehicle systems must withstand shocks and vibrations, and industrial environments can have elevated pollutant levels, etc.

An AIoT application at the network edge is usually a small IPC with an embedded rugged industrial CPU. For real-time data analysis, this CPU needs appropriate support such as Flash storage and DRAM. Industrial-grade memory components are essential for implementing AI at the Edge. The first step is to identify risks at each data collection point. Components can then be tailored to the specific requirements of the application.

Cities are growing, which increases traffic and leads to more congestion. Monitoring and adjusting traffic based on real-time data greatly improves safety and traffic flow. This is achieved through strategically placed surveillance devices. The first level of analysis is performed by local AI platforms in the city's outskirts, including vehicle detection and traffic flow evaluation. Each platform can decide how to process the data based on the analysis, e.g., whether vehicle numbers are rising and congestion is imminent. All relevant data can then be sent to the cloud, where actions like rerouting traffic, changing speed limits, and adjusting traffic signals can be implemented based on the data.

Fleet management can also be significantly optimized through AI. Monitoring a large fleet can be challenging, but many operational improvements are possible, such as reducing fuel costs, vehicle maintenance, and overall driver behavior monitoring. Current tracking systems typically rely on GPS, which is not available everywhere. For example, in tunnels, GPS cannot provide data. Similarly, inside buildings or areas with poor satellite reception, GPS is unusable for determining elevation or position. Alternative data sources can provide insights into vehicle position: for instance, vehicle speed and tire rotation rates can be continuously monitored and recorded. An onboard AI platform can then calculate the vehicle’s position at a given time by using these parameters to compensate for incomplete GPS data. This technology is known as Automotive Dead Reckoning or DR. Eventually, data can be transmitted wirelessly back to the vehicle.

Autonomous driving assumes that the driver no longer intervenes in road events. However, traffic situations constantly change, with many unexpected factors. An autonomous vehicle must be able to make decisions within seconds during sudden changes. While humans rely on their senses, robots collect data via numerous sensors to form a coherent overall picture in real-time. Cloud data is irrelevant here because of latency delays, which make it too late for decision-making. The onboard AI platform performing these complex calculations depends on components that operate reliably under all weather and physical conditions without performance loss. To avoid accidents involving autonomous vehicles, devices must work with minimal error probability and sufficient support levels.

AI will become prevalent in many aspects of our lives, replacing human operators in various scenarios, further increasing the need for robust systems capable of handling all relevant environmental conditions. Equipping AI platforms with industrial-grade memory solutions ensures hardware durability. These memory solutions are key components in building the future IoT. As a leading provider of industrial embedded Flash and memory solutions, Innodisk focuses on AI, IoT, and Edge Computing solutions. In developing various DRAM and SSD technologies, fault tolerance and system recovery are prioritized, as the number of system failures should approach zero—especially in Edge Computing environments.