A Design Of The Microprocessor With Hardware Multithread For Peripherals Management

Internet of things developed rapidly in recent years, which is widely used in social life, such as smart home, vehicle electronics and medical electronics. Internet of things is a sensor control system essentially, which collects information with a lot of sensors. The sensor device generally has real-time requirement. Traditional microprocessors for peripherals management with real-time requirement have a lot of limitations, such as the need of saving and restoring interrupt context and the randomness of the interrupt response latency. This thesis aims to design a hardware multithread processor for peripherals management based on traditional microprocessor. There are two main design goals, first, reducing the interrupt response latency. Second, reducing the average time required to process the interrupts while multi-interrupts processed in parallel.In order to accelerate interrupt response speed, this thesis proposes an embedded processor design with fine-grained multithread support based on CK802 embedded processor, which supports four interrupts processed in parallel. Interrupt service programs are dispatched to hardware multithread processor directly by interrupt dispatcher without software intervention, which reduces the saving and restoring operations of interrupt context. This thesis also proposes a conditional Round-Robin hardware thread scheduling policy which allows only ready hardware threads scheduling. Compared with Round-Robin scheduling policy, pipeline delay of concurrent interrupt service programs can be hided better, which improves the efficiency of the interrupts handling, thereby reducing the average time required to process the interrupts.Compared with baseline processor, experiments show that interrupt response delay is only one cycle when idle hardware threads exist in multithread processor, which is faster than 12 cycles of CK802. At the same time, the average interrupt handling lime of multithread processor is decreased by 48.80% while multi-interrupts processed in parallel, the cost is the area increased by 15.2%. Multithread processor implemented by this thesis is also compared with multithread processor scheduled by Round-Robin policy, experiments show that the average interrupt handling time is decreased by 15.43%, while the area increased by only 1.3%.