Published papers

This paper presents a tracking control system combining feedforward and feedback with shaped input, using a benchmark problem applied to a two-mode system. Simulations and experimental validation of the new control system for tracking control of flexible spacecraft including comparison with conventional methods are presented. The results show that the residual vibrations after the moves by the proposed control system are smaller than the conventional methods in the presence of an unknown high-order flexible mode. Sinc function-based command shaping where the spacecraft attitude is controlled along a sequentially commanded trajectory and the residual vibration after the move can be significantly reduced. Sinc function is known as an extra-insensitive function which has no frequency response above a certain threshold frequency. Three feedback controllers are applied to the control system and compared using the benchmark problem. This controller design has several unique characteristics. First, the sinc function-based shaped input is designed to reduce residual vibrations after maneuvers under the existence of unknown high-order mode. Conventional input shapers for flexible spacecraft are not effective for multi-mode system with unknown flexible modes, where actual flexible spacecraft usually has such unknown high-order flexible modes. Second, only the dynamics of the low-order mode need to be considered in designing all the feedforward, feedback controller, and shaped input. Experimental validation is performed by hardware-in-the-loop (HIL) test configuration with newly developed real-time onboard computer to evaluate the feasibility of processing load by the proposed algorithms for real-time processing. Because the proposed algorithms do not contain any iteration algorithm, the proposed algorithms can be implemented in actual satellites with feasible processing load. The assumed two-mode system represents the most simplified multi-mode system with unknown high-order mode, so the similar results can be expected for systems equipped with more complicated flexible structures.

In recent years, embedded real-time systems have grown in complexity and are required to be able to handle both periodic and aperiodic tasks. When an external interrupt occurs via an I/O peripheral, the program counter jumps to the corresponding exception vector. This results in large overhead due to context switching. However, embedded real-time systems are required to execute tasks in real-time with high precision. The Responsive Multithreaded Processor (RMTP), a prioritized SMT CPU, consists of eight hardware contexts treated as eight logical processing cores and has various hardware mechanisms for fine-grained real-time processing on the order of 10 μs. One hardware mechanism implemented in RMTP is an interrupt wake-up mechanism that enables interrupt-triggered thread wake ups. When a thread is designated to be an interrupt handling thread, it starts execution one clock cycle after an interrupt occurs. In this paper, we design and implement a software mechanism which reduces the interrupt response time by using this mechanism and demonstrate the effectiveness of our proposed method by RTL simulations. Evaluation results show that our proposed method achieves high real-time performance by reducing the interrupt response time by up to 82% compared to the baseline.

In this paper, we present 20 symbols for the error correcting 4b/10b line coding scheme, increasing the number of available symbols from the 18 originally proposed. The two additional symbols are used to supplement special control codes and increase compatibility with other line codes such as 8b/10b. In addition, we show that the 20 found is the maximum number of symbols possible for 4b/10b by formulating the problem as the maximum clique problem, an NP-hard problem. Finally, we show that the generalized problem of finding a set with the maximum number of symbols is NP-hard as well.

In real-Time systems such as automobiles, correctness of processing tasks depends not only on its result, but also on time. Real-Time scheduling is a basic approache to achieve real-Time execution in real-Time systems. Fluid scheduling is an optimal real-Time scheduling model, which keeps the execution speed of each task proportional to its processor utilization. However, the speed of a task cannot be controlled arbitrarily on a general processor, making it impossible to implement fluid scheduling directly. Therefore, there have been many studies on pseudo fluid scheduling where the algorithm is based on repeat task execution which speed is constant in an interval. We apply the special mechanisms of Responsive MultiThreaded Processor (RMTP) to the novel control mechanism in order to implement fluid scheduling directly. RMTP is a prioritized SMT processor for real-Time processing, which has 1. a thread speed control mechanism which controls execution speed of each thread and 2. a context cache enabling context switches in four clock cycles. In this paper, by adding a novel thread speed control mechanism, execution speeds of 40 threads are controlled to be constant over short time intervals, directly making fluid scheduling possible.

A spacecraft control system is often designed as a distributed real-time system that consists of processors with required functions along with peripheral devices including sensors and actuators. The system must operate under real-time constraints and be maximally dependable. In this paper, we design and implement Space Responsive Multithreaded Processor (SRMTP) System-on-Chip (SoC) and System-in-Package (SiP) for spacecraft control.

Simultaneous Multithreading (SMT) technology is widely adopted in modern high-end processors to maximize on-chip hardware utilization. In an SMT processor, multiple threads are executed in parallel, sharing hardware resources. This technique aggregates potential efficiency which will not be available in a single thread processor. However, when a data cache miss or a branch prediction miss occurs, every thread competing for hardware resources causes the degradation of hardware utilization. Therefore, instruction fetch policies have been proposed to manage hardware resources efficiently in SMT processors. The fetch policies distribute hardware resources indirectly, through fetch bandwidth control. Conventionally, a fetch policy selects fetch threads based only on the resource usage at the moment, while the characteristics of threads are not exploited on decision. Therefore most conventional fetch control schemes only take effects only after an outstanding event occurs. Capability of resource restriction is limited even with an aggressive fetch control scheme. In this paper, we propose a Fetch Gate Estimator (FGE) that is a fetch gating mechanism based on machine learning, which is implemented as a hardware module. The FGE evaluates each thread to decide whether an instruction fetch from a thread should be gated. The FGE is trained dynamically by the execution statistics resulted from the inferences, so that characteristics of each thread are encoded into a learning model. Thus, the FGE is trained dynamically in parallel with execution of programs. We applied a single layer perceptron as learning model inside the FGE for circuit simplicity, and investigated performance impacts. Evaluation results show that the perceptron-based FGE can train itself from acquired execution statistics to identify inefficient threads adaptively, adjusting resources through the fetch gate control mechanism.

Recently demands for computation intensive applications such as convolutional neural networks (CNNs) have been increasing. In these applications, valid data for computation are allocated in non-continuous addresses. Therefore, common burst memory access pattern results in a low spatial locality of valid data per access. As a result, computation of data parallel execution units degrades in throughput, as computation resource is wasted by computing invalid data. This is especially a problem in embedded systems in which constraints in power consumption provoke a requirement for high computation efficiency. In this paper, we introduce a Data Rearrange Unit (DRU), a hardware unit rearranging computation data to increase spatial locality of valid data. The DRU drastically reduces the main memory access rate and increases computation efficiency by decreasing memory access to reduce power consumption. We demonstrate the effectiveness of our DRU by implementation on the RMTP SoC [1] [2] improving convolution throughput on a data parallel execution unit by a maximum of 94times, while only increasing the total cell area by about 13%.

Fluid scheduling is an optimal real-time scheduling applicable to multiprocessor systems and is a scheduling model so that every real-time task is executed at a constant speed from release time to deadline. It is necessary for fluid scheduling to control the execution speed of each task. But, since the execution speed of the task is invariable in a conventional general purpose processor. Therefore, fluid scheduling is conventionally achieved by repeatedly executing and stopping tasks. However, this method incurs significant overheads due to frequent task switching. On the other hand, fluid scheduling without overhead by using an IPC control scheme has been proposed. The IPC control scheme controls the execution speed of each thread in an SMT processor. We propose a new IPC control scheme to extract thread-level parallelism and improve throughput effectively, also improving the schedulability of the fluid schedule. The evaluation results show that the proposed method improved the total throughput of the fluid schedule.

With the advent of parallel processors for embedded systems, it is very important to design new runtime and system features so that embedded software developers can take advantage of the increased throughput and responsiveness without the hindrances of low-level synchronization and data sharing. In this paper, we design and implement a fully-prioritized, work-conserving, and event-driven parallel computing runtime on Responsive MultiThreaded Processor (RMTP), a simultaneous multi-threading (SMT) processing platform with priority for real-time systems, which implements hardware-based resource allocation and context switching. The proposed runtime achieves low-cost forks and fine-grained prioritized scheduling by using special-purpose hardware functionalities on the RMTP, and also reduces resource over-subscription penalties while allowing for dynamic parallelism by employing a continuation-stealing execution strategy. We evaluate the performance of the proposed system with a set of benchmarks derived from the MiBench suite, and show that meaningful speedup can be achieved even when the degree of parallelism at each level is dynamically varied.

Recently, the demand for computation-intensive applications such as multimedia and AI applications has increased. Data-parallel execution units are typically used for calculations in these applications to increase computational throughput. However, the data required for computation may need to be accessed at discontinuous memory addresses, which can reduce computation efficiency. Generally, normal memory access instructions access data blocks of continuously allocated memory addresses, containing both valid and invalid data for computation. These memory access patterns result in low computation density in the data parallel execution units, wasting computation resources. Therefore, simply increasing the number of data-parallel execution units leads to an increase in wasted computational resources, which will become a significant issue in embedded systems where multiple resource limitations exist. It is essential to improve computational efficiency to perform practical computation in such systems. This paper introduces a Data Rearrange Unit (DRU), which gathers and rearranges valid computation data between main memory and execution units. The DRU improves the performance of multimedia and AI application by significantly reducing the access rate from/to main memory and increasing computation efficiency. It is applicable to most hardware architectures, and its effectiveness can be further enhanced by the execution unit interface that directly connects the DRU to the execution unit. We demonstrate the effectiveness of our DRU by implementation on the RMTP SoC, improving convolution throughput on a data-parallel execution unit by a maximum of 94 times while only increasing the total cell area by about 12.7%.

It is very important to design an embedded real-time system as a fault-tolerant system to ensure dependability. In particular, when a power failure occurs, restart processing after power restoration is required in a real-time system using a conventional processor. Even if power is restored quickly, the restart process takes a long time and causes deadline misses. In order to design a fault-tolerant real-time system, it is necessary to have a processor that can resume operation in a short time immediately after power is restored, even if a power failure occurs at any time. Since current embedded real-time systems are required to execute many tasks, high schedulability for high throughput is also important. This paper proposes a non-stop microprocessor architecture to achieve a fault-tolerant real-time system. The non-stop microprocessor is designed so as to resume normal operation even if a power failure occurs at any time, to achieve little performance degradation for high schedulability even if checkpoint creations and restorations are performed many times, to control flexibly non-volatile devices through software configuration, and to ensure data consistency no matter when a checkpoint restoration is performed. The evaluation shows that the non-stop microprocessor can restore a checkpoint within 5µsec and almost hide the overhead of checkpoint creations. The non-stop microprocessor with such capabilities will be an essential component of a fault-tolerant real-time system with high schedulability.

Recent microprocessors improve performance by extracting various levels of parallelism. Among these, out-of-order processors focus on ILP to improve performance. On the other hand, out-of-order processors consume a lot of power because they fetch and decode many instructions.We propose a Decoded Instruction Cache (DIC), in which the control signals generated by decoding RISC instructions are stored as decoded instructions in the DIC. The scheme improves performance and reduces power consumption because the results of fetch and decode can be reused. The DIC also supports multi-threaded execution, so TLP is also improved.When implemented in a multithreaded RISC processor, the DIC improves IPC by 2.39%.

The rising prominence of RISC-V, despite its inherent lack of multithreading support, has sparked research initiatives to design and implement multithreaded RISC-V processors. In this paper, we present a novel multithreaded RISC-V processor explicitly tailored for real-time systems. Our design allows for dynamic thread manipulation- encompassing creation, execution, halting, deletion, and more- to emulate the behavior of tasks in a real-time operating system’s task queue. We introduce specialized instructions to RISC-V to facilitate these thread control operations. The proposed processor is equipped with eight logical cores, enabling simultaneous execution of up to eight threads. A context cache is integrated to efficiently manage context switches between threads, achieving context switches in four clocks. This feature empowers the processor to concurrently run a greater number of threads than its core count. Simulation results underscore the benefits of our design: multithreaded execution yields an instruction per cycle (IPC) rate that surpasses single-threaded execution by up to 5.5 times. Furthermore, our priority system ensures preferential execution of high-priority threads over their lower-priority counterparts. Consequently, we have realized a multithreaded RISC-V processor capable of prioritized execution across eight threads and concurrent execution beyond its core count.

Today, embedded real-time systems such as automobiles, spacecraft, and sensor networks are parts of social infrastructures. Since a system failure in these systems may lead to a severe accident, these systems should be designed as fault-tolerant systems. If a power failure occurs in an embedded real-time system, values of all flip-flops and main memory are lost, resulting in a system failure.This paper describes the design, implementation, and evaluation of a system-on-chip/system-in-package for a non-stop fault-tolerant real-time system that continues to operate even with an unstable power supply. We designed and implemented a non-stop fault-tolerant real-time system-on-chip (SoC) that integrates a non-stop microprocessor, a non-volatile memory (MRAM), a volatile memory (SRAM), and IO peripherals. We also co-designed and implemented a non-stop fault-tolerant real-time system-in-package (SiP) that integrates the SoC, an FPGA, memories, a USB power delivery (USB PD), DC/DC converters, a potentiometer, and a board-to-board (BtoB) connector at the same time to support functions not included in the SoC but required. We evaluated the bit error rate of store operations and benchmark performance with checkpoint creations on the actual SiP.

In Simultaneous Multithreading (SMT) processors, multiple threads share hardware resources, which improves the overall utilization of the pipeline, but reduces the throughput due to stalls caused by hardware resource shortage and instruction retention of low-priority threads. In real-Time processing, where tasks are processed based on priority, high-priority threads occupy much of the processor pipeline, resulting in lower total throughput compared to SMT which does not use priority.In this paper, we propose a novel learning-based SMT processor for real-Time processing. This processor dynamically changes its priority according to the utilization of hardware resources. The proposed processor has a mechanism for autonomous learning and inference, which enables dynamic control in response to resource demand bottlenecks and their changes. There are two inference methods suitable for hard and soft real-Time systems. It also learns autonomously to make more optimal predictions. Each functional unit of the proposed processor has a counter that indicates the amount occupied at a particular time, the rate, and the total value. The counter allows learning and inference based on resource utilization, throughput, and instruction loss rate. This allows the overall throughput to be increased without compromising the real-Time performance of high-priority threads and allows resources to be allocated to other threads. In addition, simulation results show that the IPC (instructions per clock) rate is higher in soft real-Time systems, where high throughput is required in multiple applications.

Today, the development of computer technology has led to the execution of programs using multiple threads. A context switch occurs when the operating system switches from one thread to another. The context switch saves the information necessary for the computation (context) to memory and restores the context of the next thread from memory. In real-time systems where tasks must be executed within strict time constraints, the overhead of context switches can significantly impact overall system performance. The context cache is a special cache for contexts, used as a way to reduce the overhead of context switches.In this paper, we designed and implemented a context cache for multicore RISC-V processors. This design aims to accelerate context switches within a core and thread migrations between cores, improving system responsiveness and efficiency. Our evaluation shows a substantial reduction of context switch overhead by 99% with context cache. Furthermore, when running on a single core, the performance of parallel programs using the context cache approaches the performance of single-threaded programs because the context switch overhead is negligibly small.

With the advancement of computer technology, modern systems increasingly run multiple programs. A context switch occurs when the operating system switches the executing program from one thread to another. During this process, the system saves the current thread's context to memory and restores the context of the following thread from memory. However, the context switch has significant overhead, which can degrade overall system performance if it occurs frequently. In this study, we design the context cache for a multicore RISC-V processor to reduce the migration overhead caused by context switches between multiple cores. By utilizing a context cache, we aim to accelerate context switches and improve system performance and efficiency. We evaluated the context cache, which reduced the overhead of thread migration. Furthermore, we assessed the trade-off between performance and area by changing its configuration.

In safety-critical systems where failures or faults can lead to accidents, especially in special and harsh environments like those encountered in spacecraft, it is well known that multiple levels of errors can significantly increase the likelihood of system failure. For instance, system-level errors may occur due to cable breaks caused by severe vibrations, while gate-level errors can arise from radiation effects on System-on-Chip (SoC) semiconductors. Furthermore, in such specialized environments, human repair is often not feasible, necessitating that the system continue to operate even when a fault occurs in a part of it. Therefore, this research discusses the design, implementation, and evaluation of a redundant system that combines multi-level redundancy mechanisms to tolerate multiple errors and maintain operation. The implementation was carried out on the Space I/O Core (SIOC), a System-on-Chip equipped with processors and I/O for distributed real-time systems. Additionally, bitstreams were created from the implemented logic, and the performance of the fault-tolerant system was evaluated using an actual FPGA.