针对需要高性能处理和低功耗的各种应用(识别、推理、测量、控制和安全),开发了一种多核片上系统(System-on-Chip, SoC)。该SoC集成了3种可综合的处理器:8个CPU(M32R)、2个多组矩阵处理器(Multi-Bank Matrix processors, MBMX)和1个控制器(M32C)。这些处理器分别以1 GHz、500 MHz和500 MHz的频率运行,这3种处理器通过高带宽多层系统总线在芯片上相互连接,8个CPU通过缓存一致性机制连接到一个公共流水线总线上。此外,8个CPU共享1个512 KB的L2缓存以减少内部总线流量,采用了一种具有2读1写计算和后台I/O操作功能的多组矩阵处理器,1 GHz的CPU通过延迟管理网络实现,该网络包括可以适用于任何应用或工艺技术的延迟监视器。可配置异构架构具有9个CPU和2个矩阵处理器,可以将功耗降低45%。
Abstract
A system-on-chip (SoC) has been developed in this study to cater to various applications requiring high-performance processing and low power consumption,such as recognition,in-ference,measurement,control,and security.The SoC integrates three synthesizable processors:eight CPUs (M32R),two multi-bank matrix processors (MBMX),and one controller (M32C).These processors operate at frequencies of 1 GHz,500 MHz,and 500 MHz respectively.The three processors are interconnected on the chip through a high-bandwidth multi-layer system bus.The eight CPUs are connected to a shared pipeline bus via a cache coherence mechanism.Additionally,the eight CPUs share a 512 KB L2 cache to reduce internal bus traffic.The multi-bank matrix processor,which features 2-read-1-write computation and background I/O operations,is implemented using a 1 GHz CPU and a delay management network.This network includes a delay monitor applicable to any application or process technology.The configurable heterogeneous architecture comprises nine CPUs and two matrix processors,enabling a 45% reduction in power consumption.
关键词
时钟延迟调整器 /
CMOS /
可配置处理器 /
延迟监视器 /
异构多核处理器
Key words
clock delay adjuster /
CMOS /
configurable processor /
delay monitor /
heterogeneous multicore processor
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 陈诚,皮志勇,赵英龙,等.基于自适应灾变遗传-循环神经网络的锂离子电池SOC估计[J].电气工程学报,2022,17(1):86-94.
[2] 季永康,景乃锋,王琴.面向异构SoC的串并混合总线结构设计与研究[J].信息技术,2020,44(6):41-45.
[3] 闫瑾.SOC技术与发展预测[J].信息技术,2011,35(9):188-190.
[4] 雷思磊.RISC-V架构的开源处理器及SoC研究综述[J].单片机与嵌入式系统应用,2017,17(2): 56-60,76.
[5] 张发荣.面向算法的SIMD计算机数学模型及其应用[J].电子技术,2022,51(3):48-49.
[6] 陈梦清,陈洋,陈志环,等.路网约束下异构机器人系统路径规划方法[J].自动化学报,2023,49(4):718-730.
[7] NODA H,NAKAJIMA M,DOSAKA K,et al.The design and implementation of the massively parallel processor based on the matrix architecture[J].IEEE Journal of Solid-State Circuits,2006,42(1):183-192.
[8] KANEKO S,KONDO H,MASUI N,et al.A 600-mhz single-chip multiprocessor with 4.8-gb/s internal shared pipelined bus and 512-kb internal memory[J].IEEE Journal of Solid-State Circuits,2004,39(1):184-193.
[9] 左超,陈钱.计算光学成像:何来,何处,何去,何从[J].红外与激光工程,2022,51(2):158-338.