Embedded Real-Time Systems

Nobuyuki Yamasaki (Professor, Department of Information and Computer Science)

In modern society, all sorts of systems have embedded electronic devices such as sensors, actuators, and the microprocessors that control them. Many of these are interconnected via networks, forming a system as a whole. It is no exaggeration to say that modern society itself is a massive collection of various systems. Specifically, a trend is emerging where numerous sensors are attached to controlled objects (such as robots, automobiles, manufacturing equipment, factories, buildings, cities, the natural environment, and people) to monitor and control them. Simultaneously, the sensor and control data are collected in the cloud or similar platforms, and this collected data is then analyzed for optimal control. Recently, artificial intelligence (AI), particularly deep learning, has been used for this big data analysis, drawing significant attention.

On the other hand, the side that actually controls the devices is referred to as embedded or edge computing. When controlling actuators and other components, high-precision processing in real time is desired. Furthermore, since many of these devices are battery-powered, low power consumption is also expected. In other words, it is not enough to simply perform computations and communications; it is also required to control the quality of service (QoS) of these operations and their trade-offs. In many cases, a compact size is also desirable for embedding them into devices.

We are continuing our research and development with a focus on the infrastructure side of this embedded field. Specifically, on the hardware side, we are conducting R&D on microprocessors (CPUs) and networks that can achieve QoS. For real-time processing microprocessors, we are developing the RMT PU (Responsive Multithreaded Processing Unit) , which assigns priorities to threads (programs) and executes them concurrently, enabling real-time processing according to scheduling theory without overhead. Similarly, for real-time communication, we are developing the Responsive Link , a real-time communication link that allows real-time scheduling theory to be directly applied to communications by assigning priorities to packets and allowing them to overtake others at each node. The Responsive Link has been adopted as the international standard ISO/IEC 24740. Furthermore, we are conducting R&D on System-on-Chip (SoC) technology, which integrates these real-time processing CPUs ( RMT PU ), real-time networks ( Responsive Link ), interfaces (I/O) for controlling sensors and actuators, memory, and more onto a single VLSI. For example, the Responsive Multithreaded Processor (RMTP) in Figure 1 integrates all these functions onto a 10 mm square VLSI.

Furthermore, we are also conducting R&D on System-in-Package (SiP), a substrate that integrates components that cannot be integrated onto an SoC—such as main memory (DRAM), persistent storage (flash memory), power supply ICs, various sensors, and connectors—at an ultra-high density. For example, the RMTP SiP in Figure 2 integrates nearly all the functions of a PC, plus real-time processing functions, real-time communication functions, and various I/O interfaces, into a 20 mm square package.

Here, "real-time" does not mean "fast" or "immediate." Processing with some kind of time constraint (such as a deadline or period) is called real-time processing, and similarly, communication with time constraints is called real-time communication. To achieve real-time performance, we not only research theories and algorithms for real-time processing and communication but also conduct R&D on hardware architectures for computation and communication based on these theories, as well as software architectures, schedulers, and operating systems based on the algorithms. We believe that by conducting complementary R&D on all of these aspects—not just theory, not just hardware, and not just software—simultaneously, we can build more practical and high-performance real-world systems. We then apply these research outcomes to high-end embedded real-time systems such as robots and spacecraft. Figure 3 shows an example of distributed control in a spacecraft. Controllers ( RMTP SoC/SiP) are embedded in each sensor and actuator, which are interconnected by a real-time network (Responsive Link) . The entire system is controlled through distributed control as they communicate with each other. Furthermore, based on the data obtained from these real-world systems, we are devising new next-generation theories and architectures. Once these are realized, it will become possible to control various systems with high quality.