Résumé
This paper presents a new type of multithreaded and multicore microcontroller. The aim is to provide more computing power for embedded applications, like the ones developed for smart cities. Through parallelism, our microcontroller is able to run multiple tasks, either independent or cooperating, like getting data from sensors, analysing them, taking decisions and activating actuators. Parallelism is handled through multithreading, which is better suited to real-time constraints than more traditional interruptions. The microcontroller Instruction Set Architecture (ISA) is the open source RISC-V RV32I. A prototype of the microcontroller has been implemented through High-Level Synthesis (HLS) tools on a Field Programmable Gate Array (FPGA). The implementation has been tested and evaluated on a set of benchmarks from the embedded application domain. This evaluation shows that multithreading is an effective technique to hide latencies with an average 1.09 Cycle Per Instruction (CPI) on a set of 16 benchmarks taken from the embedded application domain. The microcontroller speed-up has been measured on a distributed version of the matrix multiplication, varying the number of threads to distribute the computation. This measure shows that a 4 core processor running a total of 8 threads is 5.73 times faster than a sequential run.