Intra-domain big data communication architecture for automatic driving controller

A decentralized, serial interconnected closed-loop architecture with FPGA and SSD storage for autonomous driving systems addresses high-bandwidth data transmission challenges, ensuring fault tolerance and cost-effectiveness in autonomous driving systems.

CN119094480APending Publication Date: 2024-12-06HUBEI UNIV OF AUTOMOTIVE TECH
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411204269.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Existing autonomous driving systems face challenges in managing high-bandwidth and low-latency data transmission between multiple SoCs due to limitations in traditional communication methods, leading to increased complexity, cost, and potential failure points in centralized architectures like PCIe and Ethernet Switches.

Method used

A decentralized, serial interconnected closed-loop architecture with fault-tolerant mechanisms, utilizing FPGA for clock distribution and power startup timing control, SSD storage for redundancy, and high-stability clocks to ensure continuous communication even in case of node failures.

Benefits of technology

The proposed architecture enhances robustness and fault tolerance, reduces hardware costs, simplifies design, and optimizes time and economic costs while maintaining high data transmission efficiency and scalability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119094480A_ABST
    Figure CN119094480A_ABST
Patent Text Reader

Abstract

The invention provides an intra-domain big data communication architecture for an automatic driving controller. The intra-domain big data communication architecture comprises a field programmable gate array (FPGA), a computing unit, an SSD storage unit and a high-stability clock, the computing unit is formed by connecting at least two SoC (System on Chip) in series, and each SoC is provided with a PCIe (Peripheral Component Interconnect Express) interface controller PEU0 and a PCIe interface controller PEU1; the field programmable logic gate array FPGA is used for completing reference clock distribution of the PCIe interface and power supply starting time sequence control of the Root Complex equipment and the End Point equipment; the SSD storage unit is used for backing up and recovering data, and storing and loading an upgrading program to a system chip SoC at the same time; and the high-stability clock is connected with the field programmable logic gate array (FPGA). According to the invention, a tandem interconnection closed-loop architecture without center nodes is adopted, a redundancy and fault-tolerant mechanism is introduced through the center nodes, the robustness and the fault-tolerant rate of the communication system are enhanced, each node is directly connected with two adjacent nodes, and the continuity of communication is ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to an intra-domain big data communication architecture for an autonomous driving controller. Background Art

[0002] As the complexity of autonomous driving functions continues to increase, the amount of data and computing power that vehicles need to process has also increased dramatically. Initial autonomous driving systems usually rely on distributed computing architectures, with multiple sensors (such as lidar, radar, cameras, etc.) and actuators processing data independently. However, distributed architectures have encountered challenges in system complexity and integration efficiency. From the development trend, the automotive E / E architecture will eventually evolve to a central computing architecture, and ultimately centralize the functional logic to a central controller. The hardware architecture upgrade drives the demand for chip computing power to increase exponentially. Cars need to process a large amount of unstructured data such as pictures and videos, and processors also need to integrate multiple data such as radar and video. These all put higher demands on the parallel computing and transmission of on-board processors. GPU is a single instruction, multiple data processing, using a large number of computing units and an ultra-long pipeline, so the CPU is limited in large-scale parallel computing and is relatively good at processing logic control. GPU is a large-scale parallel computing architecture composed of a large number of computing units. It was previously separated from the CPU to process image parallel computing data and is designed to process multiple parallel computing tasks at the same time. Compared with GPU, less than 20% of the CPU chip space is ALU, while more than 80% of the GPU chip space is ALU. That is, GPU has more ALUs for data parallel processing. In order to give full play to the performance of each chip, the CPU+GPU mode with acceleration chip has become a typical AI deployment solution. The CPU is used for general logic control operations, the FPGA acceleration chip improves computing power and promotes the generation of algorithms, and the GPU performs parallel computing of images. The CPU pre-processes the raw data of the sensor and passes it to the GPU. The GPU performs inference of deep learning models, image task processing, high-precision map data analysis and model establishment, extracts feature information and returns it to the CPU for subsequent decision-making and path planning.Compared with a single SoC, using different types of SoCs can maximize overall performance, achieve key task separation and accelerate parallel processing, but it will inevitably increase costs and design complexity. Multiple SoCs require a complex system architecture to coordinate communication and data transmission between them, and it is necessary to select appropriate interfaces (such as PCI e, Ethernet, shared memory, etc.) to connect different SoCs to ensure that they can transmit data with sufficient bandwidth and low latency. For a single 1080p resolution and a frame rate of 30fps, the raw data volume of the camera is 186.6MB / s. The pre-processed image compression data may be reduced to 1 / 10 to 1 / 20 of the original data. A 128-line lidar can generate millions of points per second, and the data processed by the SoC can reach 50MB / s to 100MB / s. The data of radar and inertial measurement units are relatively small, and the data processed by the SoC is between a few KB / s and hundreds of KB / s. Currently, the leading autonomous driving models on the market, such as Ideal L9, Xiaopeng G9, and Weilai ES7, are equipped with more than 30 sensors such as lidar, millimeter-wave radar, camera, ultrasonic radar, and high-precision positioning. The increasing number of sensors has brought about a surge in the amount of on-board data, which not only puts higher requirements on the bandwidth and rate of transmission for big data interaction between SoC chips. Based on the data transmission volume of a single sensor, the data volume generated by the radar and video sensors of an autonomous vehicle reaches 100MByte / s. On top of this, the fused data in the attached ECU is about 50MByte / s. For example, a vehicle equipped with five radar sensors and two video systems needs to manage a massive amount of data of about 1GByte / s during collection and storage. Such a huge amount of data also poses a great challenge to the storage of SoC data. Traditional parallel communication methods can no longer meet the requirements of throughput capacity and faster data processing capabilities for large amounts of heterogeneous data. Therefore, serial communication methods with faster data transmission efficiency and larger bandwidth have become the inevitable choice for big data communication between multiple SoCs.

[0003] To address the above challenges, the research and development of big data communication between multiple SoCs focuses on several key technical areas:

[0004] NVLink: A high-speed interconnect technology developed by NVIDIA, designed for high-speed data transmission between multiple GPUs in NVIDIA's specific ecosystem. It is suitable for high-performance computing and AI application scenarios that require processing large amounts of data. It allows GPUs to communicate in a point-to-point manner, bypassing the traditional PCIe bus, achieving higher bandwidth and lower latency. NVLink can connect two or more GPUs to achieve high-speed data transmission and sharing, providing higher performance and efficiency for multi-GPU systems. Because NVLink uses serial protocols and P2P connections for data transmission, it supports bidirectional data transmission with a bandwidth of up to 32GB / s in each direction. Greater bandwidth and higher rates mean higher economic costs and power consumption.

[0005] PCIe Switch: A device based on the PCIe protocol that uses the PCIe interface to interconnect multiple different SoC computing nodes, supports parallel transmission of multiple devices, communicates with the host through memory-mapped IO, does not rely on the network protocol stack, and directly writes or reads data to a specific memory address, making data transmission more direct. Due to the star topology of the PCIe Switch, the PCIe Switch is located in the center, and all Root Complex devices and End Point devices are directly connected to the switch through independent links. The switch acts as a central hub, responsible for managing and forwarding data flows between all devices. When the central node of the PCIe Switch fails, the entire transmission link will fail and cannot work.

[0006] Ethernet Switch: A network device based on the Ethernet protocol that forwards data packets to the target device according to the MAC address table. It uses ARP or routers. The topology of Ethernet Switch is the same as that of PCIe Switch, both of which are star topologies. When the central node fails, the entire transmission link will not work, and the transmission rate and bandwidth are lower than those of the PCIe bus.

[0007] In order to meet the transmission bandwidth and rate of massive data between multiple SoCs, high-speed serial buses Pcie and Ethernet came into being.

[0008] The ideal next-generation central computing platform (CCU) architecture integrates the three functions of vehicle control, intelligent driving, and cockpit, and integrates the SoC responsible for these three functions onto a PCB board. The high-speed communication of each SoC is supported by PCIe Switch to achieve the expansion of computing power and multi-domain integration. PCIe Switch can meet the real-time big data interaction between SoCs, solve the pain point requirements of high bandwidth and low latency, and achieve data transmission bandwidth of more than 20Gb / s between any end-to-end, and have physical isolation. In the hardware architecture of Chuangshi Automobile's intelligent driving domain controller, multiple SOC chips and MCUs are used to run simultaneously. SOC is responsible for sensor information fusion and decision planning, MCU is responsible for drive execution and safety monitoring, and Ethernet Switch is used as data exchange between different processors. Ecotron's ADAS domain controller EAXVA05 adopts a solution of two Xavier+TC297. In this hardware solution, two Gigabit Ethernet Switches are used to form a closed-loop communication. The two Gigabit Ethernet Switches are connected through the SGMII interface, and the two Xavie and TC297 are connected through the RGMII interface. Jinmai's ADAS controller GDCU35C adopts a high-reliability MCU + high-performance SoC architecture. The MCU uses an Infineon TC397, and the SoC uses two Horizon Journey J5 processors and one CoreChi X9U. The hardware solution uses two Gigabit Ethernet Switches and one PCIeSwitch for high-bandwidth and high-speed communication. The domain controller ODIN 2.0 launched by Furuitai has two SoCs and two safety MCUs built in, and their interconnection is achieved through a high-speed Ethernet Switch.

[0009] The tree structure of PCIe Switch is as follows Figure 6As shown in the figure, the Root Complex device is the root node of the entire tree structure and also the upstream port of the PCIe Switch. All End Point devices are connected to the downstream port of the PCIe Switch. The PCIe Switch, as the central node of the entire communication link, can realize the expansion of PCIe. However, the defects of this tree communication structure are obvious. When the PCIe Switch fails, the entire communication link will fail. In terms of economic cost and board layout and wiring, it will not only increase the economic burden, but also put forward requirements for device layout, differential routing, impedance matching, and heat dissipation in the layout stage, which increases the difficulty of hardware design and development. In terms of delay, the delay of PCIe Switch mainly comes from three aspects. Data needs to be transmitted from the End Point device to the PCIe Switch first, and then from the PCIe Switch to the Root Complex device, which adds a jump point and introduces additional delay. The PCIe Switch needs to handle the forwarding and routing of data packets, which will bring about PCIe Switch processing delay. If there are multiple switches in the data path, each additional switch will cause delay accumulation. In applications with high real-time requirements, the PCIe Switch solution is not well applicable. In terms of transmission bandwidth, although the PCIe Switch can support parallel transmission of multiple devices, the uplink bandwidth of the Switch is limited. If the uplink bandwidth of the Switch is not enough to support the total traffic of all downlinks, the data communication between multiple devices will have a bandwidth bottleneck.

[0010] Ethernet Switch fuzzy topology structure Figure 7As shown in the figure, compared with PCIe Switch, its advantage is that any endpoint can be used for any topology connection, but its defects are also obvious. Ethernet Switch also exchanges data and communicates around Ethernet switching devices as the central node. When the Ethernet switching device fails, the entire transmission path will be unavailable. Although Ethernet and PCIe are both high-speed serial communication methods, the current fastest 10 Gigabit Ethernet transmission rate is 10Gbit / s, while the latest PCIe 6.0 single lane rate can reach 64GT / s, the actual bandwidth can reach 8GB / s, and PCIe can select different lanes according to different scenarios. In terms of transmission rate, PCIe supports direct memory access, allowing devices to exchange data directly with main memory, reducing the data transmission time on the CPU, further reducing latency and improving performance. Although Ethernet can also support DMA, due to the complexity of the Ethernet protocol stack, the overall efficiency is not as high as PCIe. In the 10G automotive Ethernet chip market, there are currently only two manufacturers that can provide 10G or above: Marvell and Broadcom. Since 2023, the short-term price of Broadcom's BCM89 series switch chips has skyrocketed several times, with a single chip costing as much as US$30-40.

[0011] Nvlink is a high-speed interconnection technology developed by NVIDIA for interconnecting GPUs. Compared with PCIe Switch and Ethernet Switch, the biggest advantage of Nvlink is its high bandwidth and low latency. The fourth-generation NVLink connects the host and the acceleration processor at a speed of up to 900GB / s per second, integrating 4 NVlinks. A single NVlink is a duplex two-way channel. Each direction generates 8 different pairs, and each link has a bandwidth of 40GB / s for two-way, a total bandwidth of 80GB / s in one direction, and a duplex of 160GB / s. In order to simplify routing, NVlink supports channel inversion and channel polarity. Its topological structure when multiple SoCs are connected is as follows: Figure 8 As shown in the figure, due to the proprietary nature of NVLink, its disadvantages are also obvious. It is only applicable to NVIDIA GPUs. If you need to connect to GPUs or CPUs from other manufacturers, you need to expand through PCIe, which limits the versatility of NVLink and its compatibility with other brands of devices. NVLink technology cannot be widely used.

[0012] To sum up, it is necessary to make further innovations on existing technologies. Summary of the invention

[0013] In response to the technical problems existing in the above-mentioned background technology, the present invention proposes an intra-domain big data communication architecture for an autonomous driving controller, which is reasonably and ingeniously designed and adopts a serial interconnected closed-loop architecture with decentralized nodes. Redundancy and fault-tolerant mechanisms are introduced through decentralized nodes to enhance the robustness and fault tolerance of the communication system. Each node is directly connected to two adjacent nodes. Even if a node or link fails, data can still be transmitted in the other direction of the serial interconnected closed loop, thereby ensuring the continuity of communication.

[0014] In order to solve the above technical problems, the present invention provides an intra-domain big data communication architecture for an autonomous driving controller, which includes a field programmable logic gate array FPGA, a computing unit, an SSD storage unit and a high-stability clock;

[0015] The computing unit is composed of at least two system chips SoC connected in series, each system chip SoC has a PCIe interface controller PEU0 and a PCIe interface controller PEU1; in the computing unit, PCIe links are established between adjacent system chips SoC and between two system chips SoC at the head end and the end end through the PCIe interface controller PEU0 and the PCIe interface controller PEU1;

[0016] The field programmable gate array FPGA is used to complete the reference clock distribution of the PCIe interface and the power startup timing control of the RootComp lex device and the End Point device;

[0017] The SSD storage unit is used for data backup and recovery of the system chip SoC connected thereto, ensuring that the system chip SoC can be quickly restored to a normal state when a failure occurs or a restart is required, and storing and loading upgrade programs to the system chip SoC to achieve performance optimization and function expansion; the system chip SoC is configured as a root complex device through a PCIe interface controller PEU0 to connect to the SSD storage unit to establish a PCEe link;

[0018] The high-stable clock is connected to the field programmable logic gate array FPGA, and the clock signal generated by the clock is used to drive various logic circuits and clock management modules inside the field programmable logic gate array FPGA; after the high-stable clock is connected to the field programmable logic gate array FPGA, the field programmable logic gate array FPGA realizes clock signal multiplication through a phase-locked loop and outputs the PCIe interface reference clock required by the system chip SoC and the SSD storage unit. After the reference clocks of the system chip SoC and the SSD storage unit are stable and within the allowable fluctuation range, the field programmable logic gate array FPGA starts to manage the power-on startup timing of the system chip SoC and the SSD storage unit; the Root Complex device controls the startup process of the End Point device by managing the reset signal of the End Point device to ensure that the Root Complex device is ready to initialize the PCIe link when the End Point device is started, and the Root Complex device is responsible for controlling the PERST# reset signal line; after the power supply and reference clock of the Root Complex device are stable, the Root Complex device releases the reset signal of the End Point device, and the End Point device starts to start at this time.

[0019] The intra-domain big data communication architecture for the autonomous driving controller, wherein: the field programmable gate array FPGA is composed of chips U5A, U5B and U5C, and a resistor R10; the models of the chips U5A, U5B and U5C are all GW1N-9K-CM64; one end of the resistor R10 is grounded, and the other end is connected to the pin IOL5A of the chip U5B.

[0020] The intra-domain big data communication architecture for the autonomous driving controller, wherein: each of the system chips SoC has a PCIe connection port U8C; the model of the PCIe connection port U8C is Tengrui D2000 / 8, which includes a PCIe interface controller PEU0, a PCIe interface controller PEU1, an external reference clock port CLK_REQ, a data cache port ATB and an external precision calibration port REXT;

[0021] The PCIe interface controller PEU0 and the PCIe interface controller PEU1 both support the PCIe3.0 specification; the PCIe interface controller PEU0 is connected in series with an AC coupling capacitor C67 through the pin PEUO_X1_X16_REFCLKP and is connected to the pin IOB29A of the chip U5B through the AC coupling capacitor C67, and is connected in series with an AC coupling capacitor C68 through the pin PEUO_X1_X16_REFCLKN and is connected to the pin IOB29B of the chip U5B through the AC coupling capacitor C68; the PCIe interface controller PEU1 is connected in series with an AC coupling capacitor C85 through the pin PEU1_X16_REFCLKP and is connected to the pin IOB33A of the chip U5B through the AC coupling capacitor C85, and is connected in series with an AC coupling capacitor C86 through the pin PEU1_X16_REFCLKN and is connected to the pin IOB33B of the chip U5B through the AC coupling capacitor C86;

[0022] The external reference clock port CLK_REQ is a signal port of model D2000 / 8, which is used to request a reference clock signal from a motherboard or other PCIe devices;

[0023] The external precision calibration port REXT is connected to a resistor R16 through a pin PEU0_X1_REXT and is grounded through the resistor R16, is connected to a resistor R17 through a pin PEU0_X16_REXT and is grounded through the resistor R17, is connected to a resistor R21 through a pin PEU1_X1_REXT and is grounded through the resistor R21, and is connected to a resistor R22 through a pin PEU1_X16_REXT and is grounded through the resistor R22.

[0024] The intra-domain big data communication architecture for the autonomous driving controller, wherein: the SSD storage unit is composed of a hard disk M2 connector J2, decoupling capacitors C1, C2, C9 and C19, AC coupling capacitors C4, C6, C11, C12, C15 and C16, an active crystal oscillator U1, a pull-up resistor R1, an absorption resistor R3, and a filter inductor LB1;

[0025] The model of the hard disk M2 connector J2 is 1-2199230-5; the hard disk M2 connector J2 is connected to a 3.3V power supply through pins 3V3_1, 3V3_2, 3V3_3, 3V3_4 and 3V3_5, connected to pin PEU0_X16_TXP0 of the PCIe interface controller PEU0 through pin PERp0, connected to pin PEU0_X16_TXN0 of the PCIe interface controller PEU0 through pin PERn0, connected to the AC coupling capacitor C11 through pin PETp0 and connected to pin PEU0_X16_TXN0 of the PCIe interface controller PEU0 through the AC coupling capacitor C11. PEU0_X16_RXP0, connected to the AC coupling capacitor C12 through pin PETn0 and connected to the pin PEU0_X16_RXN0 of the PCIe interface controller PEU0 through the AC coupling capacitor C12, connected to the pin PEU0_X16_TXP1 of the PCIe interface controller PEU0 through pin PERp1, connected to the pin PEU0_X16_TXN1 of the PCIe interface controller PEU0 through pin PERn1, connected to the AC coupling capacitor C15 through pin PETp1 and connected to the pin PEU0 of the PCIe interface controller PEU0 through the AC coupling capacitor C15 _X16_RXP1, connect the AC coupling capacitor C16 through the pin PETn1 and connect the AC coupling capacitor C16 to the pin PEU0_X16_RXN1 of the PCIe interface controller PEU0, connect the AC coupling capacitor C4 through the pin REFCLKp and connect the AC coupling capacitor C4 to the pin IOB13B of the chip U5B, connect the AC coupling capacitor C6 through the pin REFCLKn and connect the AC coupling capacitor C6 to the pin IOB13A of the chip U5B; one end of the decoupling capacitor C1 is connected to the 3.3V power supply and the other end is grounded; one end of the decoupling capacitor C2 is connected to 3.3V power supply, the other end is grounded; one end of the pull-up resistor R1 is connected to the pin PERST# of the hard disk M2 connector J2, and the other end is connected to the 1.8V power supply; one end of the decoupling capacitor C9 is grounded, and the other end is connected to the pin PERST# of the hard disk M2 connector J2; the active crystal oscillator U1 is grounded through the pin GND, connected to the absorption resistor R3 through the pin OUT and connected to the pin SUSCLK of the hard disk M2 connector J2 through the absorption resistor R3, connected to the filter inductor LB1 through the pin VCC and connected to the 3.3V power supply through the filter inductor LB1; one end of the decoupling capacitor C19 is grounded, and the other end is connected to the 3.3V power supply.

[0026] The intra-domain big data communication architecture for the autonomous driving controller, wherein: the circuit of the high-stability clock is composed of an active crystal oscillator Y1, a decoupling capacitor C42, a resistor R11, a resistor R12 and a filter bead FB1;

[0027] The active crystal oscillator Y1 adopts 50MHz±10ppm, and the specific model is OT322550MJBA4SL; one end of the filter bead FB1 is connected to the power supply VCC1_3, and the other end is divided into two paths, one of which is connected to the resistor R11 and connected to the pin OE_N / ST / NC of the active crystal oscillator Y1 through the resistor R11, and the other is connected to the pin VDD of the active crystal oscillator Y1; one end of the decoupling capacitor C42 is grounded, and the other end is connected to the pin VDD of the active crystal oscillator Y1; the active crystal oscillator Y1 is grounded through the pin GND, connected to the resistor R12 through the pin OUT, and connected to the pin IOL15A of the chip U5B through the resistor R12;

[0028] The active crystal oscillator Y1 outputs a 50MHz high-speed stable signal to the pin IOL15A of the chip U5B through the pin OUT, and then the field programmable logic gate array FPGA multiplies the frequency to generate the 100MHz clock required for the PCI e interface reference clock, and the pins IOB13A, IOB13B, IOB17A, IOB17B, IOB23A, IOB23B, IOB27A, IOB27B, IOB29A, IOB29B, IOB33A, IOB33B, IOB39A, IOB39B, IOB41A, IOB41B, IOR17A, IOR17B, IOR22A, IOR22B, IOR24A, IOR24B, IOR6A and IOR6B on the chip U5B perform PCI e interface reference clock distribution, and at the same time, the output level voltage of all pins on the chip U5A and the chip U5B is configured through the pins VCCO0 / VCCO2 and the pin VCCO1 / VCC3 on the chip U5C.

[0029] By adopting the above technical solution, the present invention has the following beneficial effects:

[0030] The structural design of the intra-domain big data communication architecture for the autonomous driving controller of the present invention is simple and reasonable. It is a decentralized node and low-cost serial interconnection closed-loop backbone big data communication architecture based on PCI e; in particular, the serial interconnection closed-loop architecture with decentralized nodes is adopted, and redundancy and fault tolerance mechanisms are introduced through decentralized nodes to enhance the robustness and fault tolerance of the communication system; each node is directly connected to two adjacent nodes, and even if a node or link fails, data can still be transmitted in the other direction of the serial interconnection closed loop to ensure the continuity of communication. At the same time, the serial interconnection closed-loop architecture with decentralized nodes does not introduce Switch devices, and there will be no delay caused by Switch forwarding and routing. Due to the reduction of devices, the hardware cost can be reduced. At the same time, the layout and wiring time can be saved in PCB design to simplify the design. The power consumption of the entire system is relatively low, and there are significant improvements and optimizations in time, economic cost, schematic design, and transmission delay. Furthermore, the serial interconnected closed-loop architecture of decentralized nodes has good scalability compared to the Switch solution. The expansion of the Switch is limited by the number of ports of the device. When the number of ports is limited, only new Switch devices can be added for expansion. The serial interconnected closed-loop architecture of the patent of the present invention can directly connect the pre-expanded SoC in series at the same time, without the need to add Switch devices.

[0031] The reference clocks of the Root Complex device and the End Point device of the present invention are provided by the Field Programmable Logic Gate Array (FPGA), which is different from the existing PCIe reference clock solution and can realize flexible configuration of reference clocks with different requirements for different devices; at the same time, in view of the startup timing requirements of the Root Complex device and the End Point device interconnected through the PCIe interface, the power on and off timing of the interconnected system chip SoC can be programmed through the Field Programmable Logic Gate Array (FPGA). When the jitter frequency of the reference clock generated by the Field Programmable Logic Gate Array (FPGA) is large, a crystal oscillator or clock generator with lower jitter can be replaced as the clock source of the Field Programmable Logic Gate Array (FPGA), and the reference clock can also be provided by a special clock buffer chip.

[0032] The present invention adopts a large-capacity, high-speed data storage SSD storage unit for storage redundancy, which can realize real-time backup and rapid recovery, and at the same time provides a PCIe interface for firmware upgrade and software update of the system chip SoC.

[0033] Compared with PCIe Switch and Ethernet Switch solutions, the serial interconnection closed-loop architecture adopted by the present invention can produce the following positive effects:

[0034] (1) The topology of decentralized nodes is adopted, and the Switch device is removed. While ensuring the normal communication of big data, redundancy and fault-tolerant mechanisms are introduced to solve the problem that the PCIe Switch and Ethernet Switch solutions cause the failure of the entire transmission link when a single node fails. The expansion of the serial interconnection closed-loop architecture of the present invention can directly connect a new system chip SoC in series at the end without adding a new Switch node. However, the price of a Switch remains high. The PCle Switch chip represented by Broadcom SS26 has soared from US$5,500 to US$20,000. The decentralized node design concept of the present invention can optimize the hardware usage cost;

[0035] (2) The 100MHz reference clock required by the Root Complex devices and End Point devices connected to the PCIe interface is uniformly forwarded and constrained by the high-precision clock FPGA. At the same time, the field programmable logic gate array FPGA can program the startup timing required by the Root Complex devices and End Point devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 It is a schematic diagram of the intra-domain big data communication architecture for the autonomous driving controller of the present invention;

[0038] Figure 2 A circuit diagram of a field programmable gate array FPGA involved in the intra-domain big data communication architecture for an autonomous driving controller of the present invention;

[0039] Figure 3 A circuit diagram of a PCIe connection port U8C of a separate system-on-chip SoC in a computing unit involved in an intra-domain big data communication architecture for an autonomous driving controller of the present invention;

[0040] Figure 4 A circuit diagram of an SSD storage unit involved in the intra-domain big data communication architecture of an autonomous driving controller according to the present invention;

[0041] Figure 5 A circuit diagram of a high-stability clock involved in the intra-domain big data communication architecture of an autonomous driving controller of the present invention;

[0042] Figure 6 This is the tree structure diagram of PCIe Switch;

[0043] Figure 7 It is the fuzzy topology diagram of Ethernet Switch;

[0044] Figure 8 This is the topology diagram of Nvlink's multi-SoC connection. DETAILED DESCRIPTION

[0045] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The present invention is further explained below in conjunction with specific implementation modes.

[0047] like Figure 1 As shown, this embodiment provides an intra-domain big data communication architecture for an autonomous driving controller, including a field programmable gate array FPGA, a computing unit, an SSD storage unit and a high-stability clock.

[0048] The field programmable gate array FPGA is mainly used to complete the reference clock distribution of the PCIe interface and the power startup timing control of the Root Complex device and the End Point device; after the external high-stability clock is connected to the field programmable gate array FPGA, the field programmable gate array FPGA uses a phase-locked loop to multiply the clock signal and output the reference clock required by the PCIe interface of each system chip SoC and SSD storage unit. After the reference clock of the system chip SoC and SSD storage unit is stable and within the allowed fluctuation range, the power-on startup timing of the system chip SoC and SSD storage unit is managed. Generally, the power supply and reference clock of the Root Complex device should be completely stable before the End Point device; the Root Complex device controls the startup process of the End Point device by managing the reset signal of the End Point device, which ensures that when the End Point device is started, the Root Complex device is ready to initialize the PCIe link. The Root Complex device is responsible for controlling the PERST# reset signal line; after the power supply and reference clock of the Root Complex device are stable, the Root Complex device releases the reset signal of the End Point device, and the End Point device starts to start. Figure 2As shown, the field programmable gate array FPGA is composed of chips U5A, U5B and U5C, and a resistor R10; the models of the chips U5A, U5B and U5C are all GW1N-9K-CM64; one end of the resistor R10 is grounded, and the other end is connected to the pin IOL5A of the chip U5B.

[0049] The computing unit is composed of at least two system chips SoC connected in series to form a closed-loop communication architecture, each system chip SoC has a PCIe interface controller PEU0 and a PCIe interface controller PEU1; PCIe links are established between adjacent system chip SoCs and between the two system chip SoCs at the head end and the end end through the PCIe interface controller PEU0 and the PCIe interface controller PEU1. In this embodiment, the computing unit includes four system chips SoC, namely, system chip SoC_1, system chip SoC_2, system chip SoC_3 and system chip SoC_4; wherein, the system chip SoC_1~SoC_4 are four high-performance and large computing power units and are connected in series to form a closed-loop communication architecture; wherein, the system chip SoC_1 is a Root Complex device, the system chip SoC_2 is an End Point device, and the system chip SoC_3 is a Root Complex device; one PCIe controller in the system chip SoC_4 is an EndPoint device, and the other PCIe controller is set as a Root Complex device; the system chips SoC_1~SoC_4 are all connected to an external SSD storage unit through a PCIe interface. When used for big data storage, the system chip SoC should be a Root Complex device, and the SSD storage unit should be an End Point device. The reference clocks of all system chips SoC are uniformly managed by a field programmable gate array FPGA, and the field programmable gate array FPGA multiplies the clock through an external stable high-speed clock. Among them, each system chip SoC can be a CPU or a GPU, the CPU is responsible for general computing and control tasks, and the GPU is responsible for parallel computing and image processing; in this embodiment, the chip model used by each system chip SoC is the Feiteng eight-core D2000; each system chip SoC communicates through a high-speed serial bus PCIe, using decentralized nodes connected end to end, and can achieve a maximum of 32GB / s for each single link. Each system chip SoC has two PCIe controllers, each PCIe controller can be configured as a Root Complex device or an End Point device, and the two PCIe controllers can be configured as different clock sources.

[0050] like Figure 3As shown, the four system chips SoC in the computing unit all have a PCIe connection port U8C; the model of the PCIe connection port U8C is Feiteng Tengrui D2000 / 8, which includes PCIe interface controller PEU0, PCIe interface controller PEU1, external reference clock port CLK_REQ, data cache port ATB and external precision calibration port REXT.

[0051] like Figure 3As shown, the PCIe interface controller PEU0 and PCIe interface controller PEU1 both support the PCIe3.0 specification. The PCIe interface controller PEU0 and PCIe interface controller PEU1 have a total of 34 channels, each of which is an X16 channel and an X1 channel, where the X16 channel can be split into two X8 channels; and when the PCIe interface controller PEU0 is a Root Complex device, the PCIe interface controller PEU1 can be a Root Complex device or an End Point device; when the controller PEUO is an EndPoint device, the PCIe interface controller PEU1 can only be an End Point device. According to this feature, when Tengrui D2000 / 8 is connected to an external storage device, Tengrui D2000 / 8 needs to send a request, so the Tengrui D2000 / 8 end should be a Root Complex device and the storage should be an End Point device. Point device, at the transmitting end of each group of channel lanes, there are AC coupling capacitors C45~C86 with specifications of 220n / 16V and package size of 0402. These AC coupling capacitors remove the DC bias component in the high-frequency digital signal and allow the high-frequency component to pass smoothly. Among them, the PCIe interface controller PEU0 is connected to the SSD storage unit through pins PEUO_X16_RXP[0:1], pins PEUO_X16_TXP[0:1], pins PEUO_X16_RXN[0:1], and pins PEUO_X16_TXN[0:1], and pins PEUO_X16_RXP[8:15], pins PEUO_ X16_RXN[8:15] and pins PEUO_X16_TXN[8:15] are connected to the adjacent system chip SoC_2; the PCIe interface controller PEU1 is connected to the system chip SoC_4 via pins PEU1_X16_RXP[8:15], pins PEU1_X16_TXP[8:15], pins PEU1_X16_RXN[8:15], and pins PEU1_X16_TXN[8:15].At the same time, the PCIe interface controller PEU0 is also connected to the AC coupling capacitor C67 through the pin PEUO_X1_X16_REFCLKP and connected to the pin IOB29A of the chip U5B of the field programmable logic gate array FPGA through the AC coupling capacitor C67, and connected to the AC coupling capacitor C68 through the pin PEUO_X1_X16_REFCLKN and connected to the pin IOB29B of the chip U5B of the field programmable logic gate array FPGA through the AC coupling capacitor C68; the PCIe interface controller PEU1 is connected to the AC coupling capacitor C85 through the pin PEU1_X16_REFCLKP and connected to the pin IOB33A of the chip U5B of the field programmable logic gate array FPGA through the AC coupling capacitor C85, and connected to the AC coupling capacitor C86 through the pin PEU1_X16_REFCLKN and connected to the pin IOB33B of the chip U5B of the field programmable logic gate array FPGA through the AC coupling capacitor C86.

[0052] like Figure 3As shown, the external reference clock port CLK_REQ is a signal port of Tengrui D2000 / 8, which is used to request the reference clock signal of the motherboard or other PCIe devices; the signal of the external reference clock port CLK_REQ is usually used to notify the motherboard of Tengrui D2000 / 8 that a reference clock signal is required when the system chip SoC enters a low power state, and the signal of the external reference clock port CLK_REQ is used to reduce power consumption; when the system chip SoC does not need a reference clock, the motherboard of Tengrui D2000 / 8 can be requested to turn off the reference clock signal, thereby saving energy. A redundant design is made here, and the reference clock signal of the system chip SoC is provided by a field programmable logic gate array FPGA with higher accuracy. The resistors R13, R14, R15, R18, R19, and R20 are selected to be empty and not mounted; R16, R17, R21, and R22 are calibration resistors with a resistance value of 3.01KΩ and an accuracy of 1%. The calibration resistors are used for terminal matching to ensure that the impedance of the transmission line is consistent with the impedance of the signal source and the receiving end, thereby reducing signal reflection and improving signal integrity. The external reference clock port CLK_REQ is connected to the resistor R13 through the pin PEU0_C0_CLKREQ and is connected to the 1.8V power supply through the resistor R13, is connected to the resistor R14 through the pin PEU0_C1_CLKREQ and is connected to the 1.8V power supply through the resistor R14, is connected to the resistor R15 through the pin PEU0_C2_CLKREQ and is connected to the 1.8V power supply through the resistor R15, is connected to the resistor R18 through the pin PEU1_C0_CLKREQ and is connected to the 1.8V power supply through the resistor R18, is connected to the resistor R19 through the pin PEU1_C1_CLKREQ and is connected to the 1.8V power supply through the resistor R19, is connected to the resistor R20 through the pin PEU1_C2_CLKREQ and is connected to the 1.8V power supply through the resistor R20.

[0053] like Figure 3 As shown, the external precision calibration port REXT is connected to the resistor R16 through the pin PEU0_X1_REXT and grounded through the resistor R16, connected to the resistor R17 through the pin PEU0_X16_REXT and grounded through the resistor R17, connected to the resistor R21 through the pin PEU1_X1_REXT and grounded through the resistor R21, and connected to the resistor R22 through the pin PEU1_X16_REXT and grounded through the resistor R22.

[0054] like Figure 4As shown, the SSD storage unit is designed as a storage redundancy, mainly used for fast data backup and recovery of the system chip SoC connected to the SSD storage unit, ensuring that when the system chip SoC fails or needs to be restarted, it can quickly recover to a normal state, and at the same time store and load the upgrade program to the system chip SoC to achieve performance optimization and function expansion. The system chip SoC is configured as a root complex device through the PCIe interface controller PEU0 to connect to the SSD storage unit to establish a PCEe link.

[0055] like Figure 4 As shown, the SSD storage unit is mainly composed of a hard disk M2 connector J2, decoupling capacitors C1, C2, C9 and C19, AC coupling capacitors C4, C6, C11, C12, C15 and C16, an active crystal oscillator U1, a pull-up resistor R1, an absorption resistor R3, and a filter inductor LB1. Among them, the model of the hard disk M2 connector J2 is 1-2199230-5; the active crystal oscillator U1 is an independent crystal oscillator of the M.2 PCIe SSD, which can be used for data processing, cache management, and control logic inside the SSD storage unit. An independent clock source is required for operation scheduling; the high-frequency filter circuit composed of the filter inductor LB1 and the decoupling capacitors C1, C2, and C19 can filter out the high-frequency noise in the 3.3V power supply network, and the filtered clean power supply is used to power the VCC pin of the active crystal oscillator U1; the pins 3V3_1, 3V3_2, 3V3_3, 3V3_4, and 3V3_5 of the hard disk M2 connector J2 are the power supply ends of the SSD storage unit, which are powered by the 3.3V power supply network. The pin RSD_MFG_DATA of the hard disk M2 connector J2 is led out with a test point TP1, the pin RSD_MFG_CLK is led out with a test point TP2, the pin CLKREQ# is led out with a test point TP5, the pin SMB_DATA is led out with a test point TP7, and the pin SMB_CLK is led out with a test point TP8; the test points TP1, TP2, TP7, and TP8 can provide IIC and UART interfaces for subsequent debugging tests, and the test point TP5 can be used to test whether the SSD storage unit is working properly; the pin REFCLKn of the hard disk M2 connector J2 And pin REFCLKp is the reference clock of the PCIe interface given by the field programmable gate array FPGA, and pins PETn1, PETp1, PERn1, PERp1, PETn0, PETp0, PERn0 and PERp0 of the hard disk M2 connector J2 are connected to the system chip SoC in the computing unit; AC coupling capacitors C4, C6, C11, C12, C15 and C16 are placed at the transmitting end of the PCIe transmission link of the SSD storage unit, which can isolate the DC component of the crosstalk and improve the noise tolerance.

[0056] like Figure 4 As shown, the hard disk M2 connector J2 is connected to a 3.3V power supply through pins 3V3_1, 3V3_2, 3V3_3, 3V3_4 and 3V3_5, connected to pin PEU0_X16_TXP0 of the PCIe interface controller PEU0 of the PCIe connection port U8C in the computing unit through pin PERp0, connected to pin PEU0_X16_TXN0 of the PCIe interface controller PEU0 of the computing unit through pin PERn0, connected to AC coupling capacitor C11 through pin PETp0 and connected to pin PEU0_X1 of the PCIe interface controller PEU0 of the computing unit through the AC coupling capacitor C11. 6_RXP0, connected to the AC coupling capacitor C12 through pin PETn0 and connected to the pin PEU0_X16_RXN0 of the PCIe interface controller PEU0 of the computing unit through the AC coupling capacitor C12, connected to the pin PEU0_X16_TXP1 of the PCIe interface controller PEU0 of the computing unit through pin PERp1, connected to the pin PEU0_X16_TXN1 of the PCIe interface controller PEU0 of the computing unit through pin PERn1, connected to the AC coupling capacitor C15 through pin PETp1 and connected to the pin PEU0 of the PCIe interface controller PEU0 of the computing unit through the AC coupling capacitor C15 _X16_RXP1, connected to the AC coupling capacitor C16 through the pin PETn1 and connected to the pin PEU0_X16_RXN1 of the PCIe interface controller PEU0 of the computing unit through the AC coupling capacitor C16, connected to the AC coupling capacitor C4 through the pin REFCLKp and connected to the pin IOB13B of the chip U5B of the field programmable logic gate array FPGA through the AC coupling capacitor C4, connected to the AC coupling capacitor C6 through the pin REFCLKn and connected to the pin IOB13A of the chip U5B of the field programmable logic gate array FPGA through the AC coupling capacitor C6; one end of the decoupling capacitor C1 is connected to a 3.3V power supply, and the other end is connected to the 3.3V power supply. One end is grounded; one end of the decoupling capacitor C2 is connected to a 3.3V power supply, and the other end is grounded; one end of the pull-up resistor R1 is connected to the pin PERST# of the hard disk M2 connector J2, and the other end is connected to a 1.8V power supply; one end of the decoupling capacitor C9 is grounded, and the other end is connected to the pin PERST# of the hard disk M2 connector J2; the active crystal oscillator U1 is grounded through the pin GND, connected to the absorption resistor R3 through the pin OUT and connected to the pin SUSCLK of the hard disk M2 connector J2 through the absorption resistor R3, connected to the filter inductor LB1 through the pin VCC and connected to the 3.3V power supply through the filter inductor LB1; one end of the decoupling capacitor C19 is grounded, and the other end is connected to the 3.3V power supply.

[0057] This high-stability clock is an important component that provides a clock source for the field programmable logic gate array FPGA. The clock signal it generates is used to drive various logic circuits and clock management modules inside the field programmable logic gate array FPGA. Figure 5 As shown, the circuit of the high-stable clock is composed of an active crystal oscillator Y1, a decoupling capacitor C42, a resistor R11, a resistor R12 and a filter bead FB1; wherein the active crystal oscillator Y1 adopts 50MHz±10ppm, and the specific model is OT322550MJBA4SL; one end of the filter bead FB1 is connected to the power supply VCC1_3, and the other end is divided into two paths, one is connected to the resistor R11 and connected to the pin OE_N / ST / NC of the active crystal oscillator Y1 through the resistor R11, and the other is connected to the pin VDD of the active crystal oscillator Y1; one end of the decoupling capacitor C42 is grounded, and the other end is connected to the pin VDD of the active crystal oscillator Y1; the active crystal oscillator Y1 is grounded through the pin GND, connected to the resistor R12 through the pin OUT, and connected to the pin IOL15A of the chip U5B of the field programmable logic gate array FPGA through the resistor R12. The pin OUT of the active crystal oscillator Y1 outputs a 50MHz high-speed stable signal to the pin IOL15A of the chip U5B, which is then multiplied by the field programmable logic gate array FPGA to generate the 100MHz clock required for the PCIe interface reference clock, which is generated by the pins IOB13A, IOB13B, IOB17A, IOB17B, IOB23A, IOB23B, IOB27A, IOB27B, IOB29A, IOB29B, IOB33A, IOB33B, IOB39A, IOB39B, IOB41A, IOB41 on the chip U5B. B. Pins IOR17A, IOR17B, IOR22A, IOR22B, IOR24A, IOR24B, IOR6A and IOR6B are used to distribute the reference clock of the PCIe interface. At the same time, the output level voltage of all pins on chip U5A and chip U5B can be configured through pins VCCO0 / VCCO2 and VCCO1 / VCC3 on chip U5C, which can adapt to different level standards. VCCX is the core power supply, and LV version and UV version are available.

[0058] The working principle of the intra-domain big data communication architecture of the present invention for the autonomous driving controller is:

[0059] Each system chip SoC has two PCIe interface controllers, namely PCIe interface controller PEU0 and PCIe interface controller PEU1; both PCIe interface controller PEU0 and PCIe interface controller PEU1 support PCIe 3.0 specification, and both integrate DMA engine and two channels, one read and one write. PCIe interface controller PEU0 and PCIe interface controller PEU1 have 34 channels in total, each channel is one X16 channel and one X1 channel, where the X16 channel can be split into two X8 channels, and the split mode table is shown in the following table:

[0060]

[0061] The two PCIe interface controllers of each SoC support software configuration as Root Complex device and End Point device. When connected to SSD mass storage, PCIe interface controller PEU0 needs to be configured as EndPoint device. When used for big data communication between SoCs, the master-slave connection relationship between each SoC and two adjacent SoCs can be flexibly configured through PEU1 of PCIe interface controller, which can be configured as Root Complex device and End Point device. Adjacent SoCs are connected in series, and the SoCs at the head and end are also connected in series to form a node-free, multi-path serial interconnection closed-loop link.

[0062] The PCIe links for data transmission between multiple system chips SoCs and the establishment of PCIe links between system chips SoCs and SSD storage units are uniformly managed by field programmable gate arrays FPGA. The PCIe links established on the physical layer are manifested in clock stability, power-on startup of root complex devices and end point devices, and release of reset signals. The reference clocks of the PCIe links between multiple system chips SoCs and SSD storage units adopt a shared clock mode, and the field programmable gate array FPGA multiplies the high-stable clock and provides it uniformly. If there are multiple clock domains, the data transmission across clock domains can be processed through the internal FIFIO buffer of the field programmable gate array FPGA. After the clock is stable, PLL is locked, and jitter meets the requirements, the RootComplex device and the End Point device are powered on and reset. The field programmable logic gate array FPGA performs timing startup programming on the core voltage and interface voltage of the RootComplex device. After the Root Complex device is started, the EndPoint end starts to power on. After the Root Complex device and the End Point device are completed, the reset signal is released. After the link is established, the link layer negotiation between the two communicating parties begins. After the entire link is initialized, the field programmable logic gate array FPGA monitors the PCIe transmission link status. The system chip SoC detects the health of the network connection by regularly querying the interface status. If the link is disconnected or an abnormality occurs, the system chip SoC triggers the corresponding fault handling mechanism, such as restarting the network interface or switching to the reverse path transmission.

[0063] The present invention adopts a serial interconnection closed-loop architecture, which can be used for large data transmission and data storage between multiple SoCs. Different from the PCIe Switch and Ethernet Switch transmission topology solutions, in order to avoid the failure of the communication system caused by a single point failure, the biggest advantage of the present invention is that the serial interconnection closed-loop structure of decentralized nodes is used to replace the centralized Switch structure, and the redundancy and fault tolerance mechanism are introduced through the decentralized nodes to enhance the robustness and fault tolerance of the communication system. Each node is directly connected to two adjacent nodes. Even if a node or link fails, the data can still be transmitted in the other direction of the ring. The decentralized node will not introduce the Switch device, and the direct connection of the system chip SoC of the present invention will not have the routing and forwarding delays caused by the Switch device, and there will be greater resource savings in economic cost and PCB design time. The serial connection mode of the system chip SoC enables the architecture to be flexibly expanded. Compared with the centralized Switch structure, the serial interconnection closed loop can be easily expanded, and the complexity of the network will not be greatly increased by adding nodes to the closed loop. In order to meet the high data throughput and large-scale data storage requirements brought by the high speed and large bandwidth of the serial interconnected closed-loop architecture of multiple system chips SoC, a large-capacity external storage is used to connect a high-performance SSD storage unit, access the external storage device through the PCIe interface, and provide large-capacity storage for multiple system chips SoC. In order to ensure the synchronization and reliability of data transmission and storage in the serial interconnected closed-loop architecture, the global clock network of the field programmable logic gate array FPGA is used to distribute the main clock signal to all PCIe interface controllers PEU0 and PCIe interface controllers PEU1. Compared with the clock distribution chip, the field programmable logic gate array FPGA is used as a synchronization and timing management, which can better adapt to the expansion of the serial interconnected closed-loop architecture. The field programmable logic gate array FPGA can output a clock that meets the frequency, and the clock buffer in the field programmable logic gate array FPGA is used to enhance the driving ability of the clock signal and reduce the attenuation and interference of the signal, without adding any more devices. At the same time, the field programmable logic gate array FPGA can be used as a timing management unit to program the power-on timing between multiple system chips SoC to meet the start and stop sequence of different system chips SoC under different circumstances.

[0064] The present invention has a reasonable conception. It uses a decentralized node and low-cost serial interconnection closed-loop backbone big data communication architecture based on the PCIe interface to avoid the influence of a single computing node or link on the entire transmission path. It adopts the idea of ​​decentralized nodes. Multiple system chips SoC use the PCIe interface for serial interconnection to form a closed-loop communication mode. When a node fails, reverse transmission can be performed through the serial interconnection closed loop. A high-precision and stable FPGA is used instead of a clock buffer to provide a synchronous clock signal to the Root Complex device and the End Point device. Not only the big data exchange between different system chips SoCs is considered, but also the storage after the data exchange is considered. The automotive-grade SSD based on the PCIe interface will provide a data throughput rate greater than 10GB / s, and the massive storage capacity will provide strong support for sensors.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intra-domain big data communication architecture for an autonomous driving controller, characterized in that : The communication architecture includes a field programmable logic gate array FPGA, a computing unit, an SSD storage unit and a high-stability clock; The computing unit is composed of at least two system chips SoC connected in series, each system chip SoC has a PCIe interface controller PEU0 and a PCIe interface controller PEU1; In the computing unit, PCIe links are established between adjacent system chips SoC and between two system chips SoC at the head end and the tail end through PCIe interface controller PEU0 and PCIe interface controller PEU1; The field programmable gate array FPGA is used to complete the reference clock distribution of the PCIe interface and the power startup timing control of the Root Complex device and the End Point device; The SSD storage unit is used for data backup and recovery of the system chip SoC connected thereto, ensuring that the system chip SoC can be quickly restored to a normal state when a failure occurs or a restart is required, and storing and loading upgrade programs to the system chip SoC to achieve performance optimization and function expansion; the system chip SoC is configured as a root complex device through a PCIe interface controller PEU0 to connect to the SSD storage unit to establish a PCEe link; The high-stability clock is connected to the field programmable logic gate array FPGA, and the clock signal generated by the clock is used to drive various logic circuits and clock management modules inside the field programmable logic gate array FPGA; after the high-stability clock is connected to the field programmable logic gate array FPGA, the field programmable logic gate array FPGA realizes clock signal frequency multiplication through a phase-locked loop and outputs the PCIe interface reference clock required by the system chip SoC and the SSD storage unit. After the reference clocks of the system chip SoC and the SSD storage unit are stable and within the allowable fluctuation range, the field programmable logic gate array FPGA starts to manage the power-on startup timing of the system chip SoC and the SSD storage unit; The Root Complex device controls the startup process of the End Point device by managing the reset signal of the End Point device to ensure that the Root Complex device is ready to initialize the PCIe link when the End Point device starts. The Root Complex device is responsible for controlling the PERST# reset signal line; after the power supply and reference clock of the Root Complex device are stable, the Root Complex device releases the reset signal of the End Point device, and the End Point device starts to start.

2. The intra-domain big data communication architecture for an autonomous driving controller as claimed in claim 1, characterized in that :The field programmable gate array FPGA is composed of chips U5A, U5B and U5C, and a resistor R10; the models of the chips U5A, U5B and U5C are all GW1 N-9K-CM64; one end of the resistor R10 is grounded, and the other end is connected to the pin IOL5A of the chip U5B.

3. The intra-domain big data communication architecture for an autonomous driving controller as claimed in claim 2, characterized in that : Each of the system chips SoC has a PCIe connection port U8C; the model of the PCIe connection port U8C is Tengrui D2000 / 8, which includes a PCIe interface controller PEU0, a PCIe interface controller PEU1, an external reference clock port CLK_REQ, a data buffer port ATB and an external precision calibration port REXT; The PCIe interface controller PEU0 and the PCIe interface controller PEU1 both support the PCIe3.0 specification; the PCIe interface controller PEU0 is connected in series with an AC coupling capacitor C67 through the pin PEUO_X1_X16_REFCLKP and is connected to the pin IOB29A of the chip U5B through the AC coupling capacitor C67, and is connected in series with an AC coupling capacitor C68 through the pin PEUO_X1_X16_REFCLKN and is connected to the pin IOB29B of the chip U5B through the AC coupling capacitor C68; the PCIe interface controller PEU1 is connected in series with an AC coupling capacitor C85 through the pin PEU1_X16_REFCLKP and is connected to the pin IOB33A of the chip U5B through the AC coupling capacitor C85, and is connected in series with an AC coupling capacitor C86 through the pin PEU1_X16_REFCLKN and is connected to the pin IOB33B of the chip U5B through the AC coupling capacitor C86; The external reference clock port CLK_REQ is a signal port of model D2000 / 8, which is used to request a reference clock signal from a motherboard or other PCIe devices; The external precision calibration port REXT is connected to a resistor R16 through a pin PEU0_X1_REXT and is grounded through the resistor R16, is connected to a resistor R17 through a pin PEU0_X16_REXT and is grounded through the resistor R17, is connected to a resistor R21 through a pin PEU1_X1_REXT and is grounded through the resistor R21, and is connected to a resistor R22 through a pin PEU1_X16_REXT and is grounded through the resistor R22.

4. The intra-domain big data communication architecture for an autonomous driving controller as claimed in claim 3, characterized in that : The SSD storage unit is composed of a hard disk M2 connector J2, decoupling capacitors C1, C2, C9 and C19, AC coupling capacitors C4, C6, C11, C12, C15 and C16, an active crystal oscillator U1, a pull-up resistor R1, an absorption resistor R3, and a filter inductor LB1; The model of the hard disk M2 connector J2 is 1-2199230-5; the hard disk M2 connector J2 is connected to a 3.3V power supply through pins 3V3_1, 3V3_2, 3V3_3, 3V3_4 and 3V3_5, connected to pin PEU0_X16_TXP0 of the PCIe interface controller PEU0 through pin PERp0, connected to pin PEU0_X16_TXN0 of the PCIe interface controller PEU0 through pin PERn0, connected to the AC coupling capacitor C11 through pin PETp0 and connected to pin PEU0_X16_TXN0 of the PCIe interface controller PEU0 through the AC coupling capacitor C11. PEU0_X16_RXP0, connected to the AC coupling capacitor C12 through pin PETn0 and connected to the pin PEU0_X16_RXN0 of the PCIe interface controller PEU0 through the AC coupling capacitor C12, connected to the pin PEU0_X16_TXP1 of the PCIe interface controller PEU0 through pin PERp1, connected to the pin PEU0_X16_TXN1 of the PCIe interface controller PEU0 through pin PERn1, connected to the AC coupling capacitor C15 through pin PETp1 and connected to the pin PEU0 of the PCIe interface controller PEU0 through the AC coupling capacitor C15 _X16_RXP1, connect the AC coupling capacitor C16 through the pin PETn1 and connect the AC coupling capacitor C16 to the pin PEU0_X16_RXN1 of the PCIe interface controller PEU0, connect the AC coupling capacitor C4 through the pin REFCLKp and connect the AC coupling capacitor C4 to the pin IOB13B of the chip U5B, connect the AC coupling capacitor C6 through the pin REFCLKn and connect the AC coupling capacitor C6 to the pin IOB13A of the chip U5B; one end of the decoupling capacitor C1 is connected to the 3.3V power supply and the other end is grounded; one end of the decoupling capacitor C2 is connected to 3.3V power supply, the other end is grounded; one end of the pull-up resistor R1 is connected to the pin PERST# of the hard disk M2 connector J2, and the other end is connected to the 1.8V power supply; one end of the decoupling capacitor C9 is grounded, and the other end is connected to the pin PERST# of the hard disk M2 connector J2; the active crystal oscillator U1 is grounded through the pin GND, connected to the absorption resistor R3 through the pin OUT and connected to the pin SUSCLK of the hard disk M2 connector J2 through the absorption resistor R3, connected to the filter inductor LB1 through the pin VCC and connected to the 3.3V power supply through the filter inductor LB1; one end of the decoupling capacitor C19 is grounded, and the other end is connected to the 3.3V power supply.

5. The intra-domain big data communication architecture for an autonomous driving controller as claimed in claim 3, characterized in that : The circuit of the high stable clock is composed of an active crystal oscillator Y1, a decoupling capacitor C42, a resistor R11, a resistor R12 and a filter bead FB1; The active crystal oscillator Y1 adopts 50MHz±10ppm, and the specific model is OT322550MJBA4SL; one end of the filter bead FB1 is connected to the power supply VCC1_3, and the other end is divided into two paths, one of which is connected to the resistor R11 and connected to the pin OE_N / ST / NC of the active crystal oscillator Y1 through the resistor R11, and the other is connected to the pin VDD of the active crystal oscillator Y1; one end of the decoupling capacitor C42 is grounded, and the other end is connected to the pin VDD of the active crystal oscillator Y1; the active crystal oscillator Y1 is grounded through the pin GND, connected to the resistor R12 through the pin OUT, and connected to the pin IOL15A of the chip U5B through the resistor R12; The active crystal oscillator Y1 outputs a 50MHz high-speed stable signal to the pin IOL15A of the chip U5B through the pin OUT, and then the field programmable logic gate array FPGA multiplies the frequency to generate the 100MHz clock required for the PCIe interface reference clock, which is generated by the pins IOB13A, IOB13B, IOB17A, IOB17B, IOB23A, IOB23B, IOB27A, IOB27B, IOB29A, IOB29B, IOB33A, IOB33B, IOB39A, IOB39B, IOB41A, IOB41 on the chip U5B. B, pins IOR17A, IOR17B, IOR22A, IOR22B, IOR24A, IOR24B, IOR6A and IOR6B distribute the PCIe interface reference clock, and at the same time, the output level voltage of all pins on the chip U5A and the chip U5B is configured through the power supply of the pins VCCO0 / VCCO2 and the pin VCCO1 / VCC3 on the chip U5C.

Citation Information

Cited By

  • Method, device and equipment for correcting delay between optical interconnection link channels and storage medium

    CN120499537A