A high-speed serially expanded FPGA cluster for ASIC prototyping
By adopting high-speed serial extension ASIC prototype verification technology in FPGA cluster, using asynchronous bus and high-speed serial data links, the status monitoring and data synchronization between multiple FPGAs are achieved, and the hardware complexity in the existing technology is solved. The problem of insufficient asynchronous clock support reduces the hardware requirements of FPGAs, improves data throughput and system scalability.
Patent Information
- Application Number
- CN202311057722.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-22
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2043-08-22
AI Technical Summary
The existing technology is difficult to implement the status monitoring of multiple boards and multiple FPGAs in large-scale FPGA clusters, especially in asynchronous clock environments, and high-speed serdes function is required to support data recovery and signal import and export, resulting in complex hardware design and correct FPGA chips have high requirements, but domestic FPGA chips do not have this capability.
ASIC prototype using high-speed serial extension is used to verify the FPGA cluster, and a data link is formed through the high-speed serial interface of the main FPGA node and the slave FPGA node. It uses an asynchronous bus for monitoring and control, and realizes data synchronization and interconnection between multiple FPGAs. Reduce hardware requirements for FPGAs and support asynchronous clock environments.
Simplifies the difficulty of building a large-scale FPGA prototype verification cluster, reduces the requirements for FPGAs, achieves high replication and high scalability, supports stable data synchronization and interconnection of multiple FPGAs, improves data throughput, and meets the Requirements for ASIC prototype verification systems.
Smart Images

Figure CN117271415B_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to the technical field of ASIC chip design, and in particular to a high-speed serially extended ASIC prototype verification FPGA cluster. [Background technology]
[0002] In the current large-scale FPGA prototype verification cluster, how to realize the status monitoring of multiple boards and multiple FPGAs in the board, including whether the TDM (time division multiplexing) circuit is in normal status, whether there are bit errors in the transmission process, whether the PLL (phase-locked loop) in each FPGA is locked normally, and quickly realize the export of the signals that need to be observed in the DUT (Design under Test) in each FPGA, force control of some of the signals, and write data to the internal RAM if it needs to be imported later. At present, the general method is to deploy one or more main control boards in the cluster design to realize the monitoring and data import and export of multiple FPGAs based on technologies such as PCIE (PCI-Express, peripheral component interconnect express, a high-speed serial computer expansion bus standard). This method is firstly complex in hardware design, and at the same time, it has high requirements on the FPGA chip above, and needs to have high-end serdes (short for SERializer serializer / DESerializer deserializer, a mainstream time division multiplexing (TDM), point-to-point (P2P) serial communication technology) function, but most domestic FPGA chips currently do not have this capability.
[0003] High-speed serial data recovery technology: "A method for realizing high-speed serial link data recovery based on FPGA", author Qu Wei and Sun Zhian, published in "Computer Measurement and Control" 2011(019)001, proposed a design method to complete the data recovery at the receiving end through FPGA chip logic, using local clock and phase-locked loop to sample high-speed serial link data at different phases, and effectively judge the data through logic design, solving the problem of asynchronous clock data in serial link, and realizing effective recovery of high-speed serial link data up to 400Mb / s.
[0004] Using the keywords "FPGA" and "prototype" to search the Chinese patent database, the following related patents were retrieved.
[0005] Related search result 1: Application (patent) number, CN202110628338.1, name, segmentation and verification method, device, electronic device, storage medium, discloses a segmentation and verification method, device, electronic device, storage medium, applied to the field of electronic design automation technology, the segmentation scheme includes: classifying the nodes in the chip design, merging the classified nodes, and segmenting the new nodes after the merging process, so as to set the segmentation boundary on the connection line with trigger drive; by optimizing and adjusting all segmentation boundaries to the connection line with trigger drive, the accuracy and efficiency of the segmentation and verification of the chip design can be improved. The implementation method of the related patent 1 is based on the TDM circuit under the premise of fully synchronous clock in the FPGA cluster, not a serial circuit based on DR high-speed data recovery, and cannot support asynchronous clock. The application scenario of the related patent 1 is mainly based on synchronization, that is, multi-FPGA segmentation of ASIC design under the premise of fully synchronized clock in FPGA cluster. If you need to realize status monitoring, observation of some signals and import of some data in large-scale FPGA clusters, you need to use high-speed circuits such as PCIE serdes in FPGA. It cannot be used for design segmentation of large-scale ASIC design in FPGA clusters based on AXI bus as the cutting point, and cannot be used for status monitoring, observation of some signals and import of some data in large-scale FPGA clusters at the same time.
[0006] Related search result 2: Application (patent) number, CN202011043967.X, name, a design partitioning method for multiple FPGAs, discloses a design partitioning method for multiple FPGAs, and relates to the field of chip logic design technology; including: S1, modeling the logic circuit as a hypergraph: configuring the group interconnection size resources and the cutting group balance factor; S2, hypergraph coarsening: clustering closely connected vertices together to construct a subgraph; S3, initial partitioning: traversing the clustered subgraphs according to the partition size and the cutting group balance factor using the greedy algorithm rules to distribute them to different partitions; S4, optimization adjustment: calculating the movable vertex benefits according to the FM algorithm, making adjustments, and optimizing the cutting size; it can achieve an optimal partitioning result under the interconnection constraint rules between FPGAs during the partitioning process of the multi-FPGA prototype system. The implementation method of the related patent 1 is based on the TDM circuit under the premise of fully synchronous clock in the FPGA cluster, not the serial circuit based on DR high-speed data recovery, and cannot support asynchronous clocks. The application scenario of the related patent 1 is mainly based on synchronization and multi-FPGA segmentation of ASIC design under the premise of fully synchronized clock in FPGA cluster. If you need to realize status monitoring, observation of some signals and import of some data in large-scale FPGA clusters, you need to use high-speed circuits such as PCIE serdes in FPGA. It cannot be used for design segmentation of large-scale ASIC design in FPGA clusters based on AXI bus as the cutting point, and cannot be used for status monitoring, observation of some signals and import of some data in large-scale FPGA clusters at the same time.
[0007] The present invention aims at the technical problems that the prior art needs high-speed serdes to be implemented and relevant clocks need to be synchronized, and makes technical improvements to the ASIC prototype verification FPGA cluster. [Summary of the invention]
[0008] The purpose of the present invention is to provide a large-scale FPGA prototype verification cluster with high replicability and high scalability.
[0009] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is a high-speed serial expansion ASIC prototype verification FPGA cluster, including a master FPGA node and a plurality of slave FPGA nodes, the FPGA node including a high-speed serial interface constructed based on a high-speed serial data recovery technology, the master FPGA node high-speed serial interface is connected to the slave FPGA node high-speed serial interface, the slave FPGA node high-speed serial interface is connected to the next slave FPGA node high-speed serial interface, to form a high-speed serial data link; the high-speed serial interface of each slave FPGA node is extended to an asynchronous bus through a bus bridge, which is used to monitor and control the ASIC prototype verification test design of the slave FPGA node; the master FPGA node is used to send and receive data frames through the high-speed serial data link to each The slave FPGA node ASIC prototype verifies the design to be tested for monitoring and control, and the data frame includes a node ID field, a TX field for controlling the slave FPGA node ASIC prototype verifies the design to be tested, and an RX field for monitoring the slave FPGA node ASIC prototype verifies the design to be tested; each slave FPGA node is used to send, receive and parse the data frame through a high-speed serial data link, and the slave FPGA node pointed to by the node ID writes the data frame TX field status into the slave FPGA node ASIC prototype verifies the design to be tested through an asynchronous bus, obtains the slave FPGA node ASIC prototype verifies the design to be tested The state is written into the data frame RX field, and the slave FPGA node not pointed to by the node ID transparently transmits the data frame to the next FPGA node through a high-speed serial data link.
[0010] Preferably, a new slave FPGA node is constructed according to the need for ASIC prototype verification, and the new slave FPGA node is inserted into the high-speed serial data link.
[0011] Preferably, the asynchronous bus is an AXI asynchronous bus, an APB asynchronous bus, an AHB asynchronous bus or a CHI asynchronous bus.
[0012] Preferably, the high-speed serial data link is a physical channel formed by serially connecting all FPGA nodes through a pair of LVDS TX signals and a pair of LVDS RX signals.
[0013] Preferably, each slave FPGA node high-speed serial interface has a PRBS-based self-detection capability.
[0014] Preferably, the data frame further includes a fixed frame header field.
[0015] Preferably, the monitoring of the design to be tested from the FPGA node ASIC prototype verification includes cyclic redundancy check monitoring.
[0016] The high-speed serially expanded ASIC prototype verification FPGA cluster of the present invention has the following beneficial effects: 1. It can effectively simplify the construction difficulty of a large-scale FPGA prototype verification cluster and reduce the requirements for FPGA, thereby supporting prototype verification clusters implemented by different FPGAs; 2. It has high replicability and high scalability, and can conveniently and efficiently support the construction and expansion of FPGA prototype verification clusters; 3. The required pin cost is small, and on the basis of clock synchronization between multiple FPGAs in the cluster, based on high-speed serial data recovery technology, a higher data throughput can be achieved, which can fully meet the needs of the FPGA prototype system and realize stable data synchronization and interconnection of multiple FPGAs.
Brief Description of the Drawings
[0017] Figure 1 The schematic diagram is a high-speed serial expansion ASIC prototype verification FPGA cluster architecture.
[0018] Figure 2 It is a high-speed serial expansion ASIC prototype to verify the data layer communication flow chart of the FPGA cluster. [Specific implementation method]
[0019] The features and exemplary embodiments of various aspects of the present invention will be described in detail below. In the detailed description below, many specific details are proposed to provide a comprehensive understanding of the present invention. But it is obvious to those skilled in the art that the present invention can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present invention by illustrating examples of the present invention. The present invention is by no means limited to any specific configuration and algorithm proposed below, but covers any modification, replacement and improvement of elements, parts and algorithms without departing from the concept of the present invention. In the accompanying drawings and the following description, known structures and technologies are not shown to avoid unnecessary ambiguity to the present invention.
[0020] Example
[0021] This embodiment implements a high-speed serial expansion ASIC prototype verification FPGA cluster.
[0022] The cluster communication link layer of this embodiment is based on high-speed serial data recovery technology, and mainly constructs a high-speed serial expansion circuit with high reproducibility and high scalability. The cluster of this embodiment realizes status monitoring of each FPGA in the cluster and high-throughput reading and writing of internal signals based on an asynchronous bus. The cluster of this embodiment only requires a small number of FPGA IOs (this is a relative concept, which depends on the scale of the overall cluster. A lot of FPGA IOs may also be required, but it is still small compared to the scale of the cluster). There is no need to integrate high-speed serdes circuits such as PCIE and related processing circuits in the FPGA to achieve this. The cluster of this embodiment also does not need to synchronize the clocks of all FPGAs, which greatly reduces the difficulty and requirements for the FPGA cluster.
[0023] The cluster data layer of this embodiment is based on a multi-layer data protocol that is easy to use in FPGA clusters, which is convenient for data expansion in large-scale FPGA clusters and is very easy to convert to asynchronous bus protocols commonly used in the industry, such as AXI (Advanced eXtensible Interface) or APB (Advanced Peripheral Bus), AHB (Advanced High Performance Bus) or CHI (Coherent Hub Interface); the cluster of this embodiment uses the above-mentioned data layer protocol to quickly generate such serial circuits in the FPGA cluster, and automatically insert, allocate, connect and split them into FPGAs in the cluster; it solves how to automatically insert high-speed serial expansion circuits in the cluster, and very conveniently and quickly builds a large-scale FPGA prototype verification cluster with high serial expansion.
[0024] FPGA: (Field Program Gate Arrary) Field programmable gate circuit, widely used in ASIC prototype verification and simulation acceleration platforms. Currently, domestic FPGA is relatively backward.
[0025] LVDS: (Low Voltage Diffential Signal) low voltage differential pair, with good signal quality and high speed, can work up to 1.2Gbps.
[0026] Serdes: A high-speed serial transceiver that can currently operate at up to 50Gbps and is the basis of high-speed circuits such as PCIE / USB / MIPI.
[0027] Prototyping: Prototype verification. As ASIC designs become larger and more complex, it is necessary to build large-scale FPGA clusters to split the ASIC into multiple FPGAs, simulate ASIC behavior, and perform functional verification and software development.
[0028] PRBS: Pseudo-Random Binary Sequence, also known as pseudo-random code, is often used in testing high-speed serial channels.
[0029] bus-brg: bus-bridge.
[0030] Figure 1 This is a high-speed serial expansion ASIC prototype verification FPGA cluster architecture diagram. Figure 1 As shown, the following embodiment takes the AXI asynchronous bus as an example to illustrate the specific implementation principle of the cluster in this embodiment.
[0031] The cluster structure of this embodiment is described as follows:
[0032] (1) Build an FPGA prototype cluster environment (deploy 1 master FPGA node and multiple slave FPGA nodes (FPGA1, FPGA2, etc.)) and configure the ASIC prototype verification environment so that the FPGA cluster system of this embodiment meets high availability!
[0033] (2) The master FPGA deploys a high-speed serial master interface, and each slave FPGA node deploys a high-speed serial slave interface. The slave FPGA node is expanded according to the needs of ASIC prototype verification, and automatically inserts the high-speed serial slave interface and automatically connects to form a high-speed serial data link.
[0034] (3) The high-speed serial slave interface of each slave FPGA node is extended to an asynchronous bus, such as an AXI asynchronous bus, through a bus bridge to achieve monitoring connection of the DUT (Design under Test) and monitoring connection of the status of large-scale FPGAs (master FPGA node and each slave FPGA node).
[0035] (4) The master FPGA node controls all FPGA nodes in the cluster based on a multi-layer data protocol by controlling a high-speed serial data link.
[0036] Figure 2 This is a high-speed serial expansion ASIC prototype to verify the FPGA cluster data layer communication flow chart. Figure 2 As shown in the figure, this implementation of large-scale FPGA prototype verification cluster master FPGA node control data communication process, the main steps are:
[0037] Step 1: Based on the high-speed serial data recovery technology, a high-throughput high-speed serial data link is constructed for data transmission and reception. Based on the high-speed serial slave interface, the slave FPGA node can be serially extended to further expand the high-speed serial data link.
[0038] Step 2: Construct a multi-layer data protocol at the asynchronous bus data layer of the high-speed serial data link to facilitate data parsing by each FPGA node in the large-scale FPGA prototype verification cluster and facilitate cluster expansion.
[0039] Step 3: At the cluster software level of this embodiment, a high-speed serial slave interface can be automatically inserted into each slave FPGA node in the FPGA cluster, automatically connected to form a data chain, and automatic segmentation of ASIC segmentation is supported. At the software level, each module is automatically generated according to the configuration and inserted into each FPGA, which is the core technical feature of this embodiment.
[0040] Step 4: In the FPGA cluster built according to the same high-speed serial slave interface configuration file as above, the high-speed serial data module of each FPGA node can monitor, calibrate, parse and transparently transmit the asynchronous bus information.
[0041] Step 5: For the large-scale FPGA prototype verification cluster deployed as above, the master FPHA node can parse and analyze the bus based on its own protocol layering, so as to monitor the internal status of each slave FPGA node. At the same time, it can import the signals that need to be controlled and written from the FPGA node, and export the signals that need to be monitored from the FPGA node.
[0042] Wherein, step 1 includes the following steps:
[0043] Step 1.1: Each FPGA node only needs to use a pair of LVDS TX and a pair of LVDS RX signals as the physical channels for data link interconnection in large-scale FPGA clusters.
[0044] Step 1.2: Through the LVDS differential pair, based on high-speed serial data recovery technology, it has a strong ability to resist frequency deviation. In a large-scale FPGA cluster, there is no need for each FPGA to use a synchronous clock, realizing a high-throughput serial transceiver circuit.
[0045] Step 1.3: The FPGA node high-speed serial data module has a self-detection capability based on PRBS (pseudo-random code). It can be switched to the test mode for self-detection or switched to the working mode.
[0046] Step 1.4: The FPGA node high-speed serial data module has certain data buffering and forwarding capabilities, which can be used to achieve data synchronization and interconnection of multiple FPGA nodes.
[0047] Wherein, step 2 includes the following steps:
[0048] Step 2.1: The entire data path adopts a multi-layer data protocol. The data link layer adopts the format of data frames. The data frames have fixed frame header data to facilitate data recovery.
[0049] Step 2.2: The high-speed serial data link network layer mainly parses the FPGA node ID byte data contained in the data frame. When the ID of the current FPGA node is detected, the next layer of parsing is performed inside the FPGA node. If it is not the ID of the current FPGA node, it is directly transparently transmitted to the next FPGA node through the high-speed serial data link.
[0050] Step 2.3: The high-speed serial data link transmission layer converts the AXI write address, write data, and read address related channel data from the TX direction data, and fills the write status, read data and other channel information into the RX direction data.
[0051] Wherein, step 3 includes the following steps:
[0052] Step 3.1: In this embodiment, the cluster control software reads the configuration information of the FPGA cluster and obtains the number of FPGAs required in the FPGA cluster.
[0053] Step 3.2: In this embodiment, the cluster control software automatically instantiates the same number of high-speed serial data modules in the top level according to the number of FPGAs used in the cluster.
[0054] Step 3.3: The cluster control software of this embodiment allocates a specific ID number to each FPGA, and transmits the ID to the instantiation module (high-speed serial data module) as the unique identification number of this module.
[0055] Step 3.4: The cluster control software of this embodiment automatically connects these modules to form a data link.
[0056] Step 3.5: The user divides the ASIC into sections and distributes them to the FPGA node according to the subsequent allocation strategy. The ASIC prototype verification signals required for observation are connected to the AXI asynchronous bus restored by the high-speed serial data module.
[0057] Step 3.6: In the subsequent platform multi-FPGA ASIC prototype verification segmentation of the cluster control software of this embodiment, the script automatically segments the high-speed serial data module into the corresponding FPGA according to the ID number.
[0058] Step 3.7: At the same time, the cluster control software of this embodiment automatically connects the CRC (Cyclic Redundancy Check) monitoring status and other information related to the TDM (Time Division Multiplexing) module generated after the segmentation to the specific position in the AXI bus restored by the high-speed serial data module.
[0059] Wherein, step 4 includes the following steps:
[0060] Step 4.1: The large-scale FPGA cluster is built according to the hardware configuration file, and the LVDS required by the serial circuit in each FPGA is physically connected through Cable.
[0061] Step 4.2: After the platform is powered on, the system is first in test mode to perform PRBS-based path connectivity detection and bit error rate detection on the entire link to ensure the reliability of the data link.
[0062] Step 4.3: After the platform passes the self-test, the circuit detects the bus packet data according to the bus protocol and analyzes it. When the FPGA node ID is detected, it performs a further analysis internally and converts the bus into AXI-related channel data. If it is not the ID of the FPGA node, it is directly transparently transmitted to the next FPGA node internally.
[0063] This embodiment can effectively simplify the difficulty of building a large-scale FPGA prototype verification cluster, reduce the requirements for FPGA, and thus support prototype verification clusters implemented by different FPGAs.
[0064] This embodiment is highly reproducible and scalable, can effectively support cluster expansion, and can very conveniently and efficiently support large-scale FPGA cluster construction for ASC prototype verification.
[0065] At the same time, the FPGA pin cost required by this embodiment is small, and on the basis of clock synchronization between multiple FPGAs in a cluster, based on high-speed serial data recovery technology, a higher data throughput can be achieved, which can fully meet the needs of the ASIC prototype verification system and realize stable data synchronization and interconnection of multiple FPGAs.
[0066] Those skilled in the art will appreciate that all or part of the steps to implement the above embodiments may be accomplished by hardware, or may be accomplished by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, wherein the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.
[0067] The above is only a preferred embodiment of the present invention. It should be pointed out that ordinary technicians in this technical field can make several improvements and supplements without departing from the principle of the present invention. These improvements and supplements should also be regarded as the scope of protection of the present invention.
Claims
1. A high-speed serially expanded ASIC prototype verification FPGA cluster, characterized by: The invention comprises a master FPGA node and several slave FPGA nodes, wherein the FPGA node comprises a high-speed serial interface constructed based on a high-speed serial data recovery technology, wherein the high-speed serial interface of the master FPGA node is connected to the high-speed serial interface of the slave FPGA node, and the high-speed serial interface of the slave FPGA node is connected to the high-speed serial interface of the next slave FPGA node to form a high-speed serial data link; the high-speed serial interface of each slave FPGA node is extended to form an asynchronous bus through a bus bridge, which is used to monitor and control the ASIC prototype verification of the slave FPGA node to be tested; the master FPGA node is used to monitor and control the ASIC prototype verification of each slave FPGA node to be tested through the high-speed serial data link. , control, the data frame includes a node ID field, a TX field for controlling the slave FPGA node ASIC prototype verification design to be tested, and an RX field for monitoring the slave FPGA node ASIC prototype verification design to be tested; each slave FPGA node is used to send, receive and parse the data frame through a high-speed serial data link, the slave FPGA node pointed to by the node ID writes the data frame TX field status into the slave FPGA node ASIC prototype verification design to be tested through an asynchronous bus, obtains the slave FPGA node ASIC prototype verification design to be tested status and writes it into the data frame RX field, the slave FPGA node not pointed to by the node ID transparently transmits the data frame to the next FPGA node through a high-speed serial data link.
2. The high-speed serially expanded ASIC prototype verification FPGA cluster according to claim 1, characterized in that: A new slave FPGA node is constructed according to the need for ASIC prototype verification, and the new slave FPGA node is inserted into the high-speed serial data link.
3. The high-speed serially expanded ASIC prototype verification FPGA cluster according to claim 1, characterized in that: The asynchronous bus is an AXI asynchronous bus, an APB asynchronous bus, an AHB asynchronous bus or a CHI asynchronous bus.
4. The high-speed serially expanded ASIC prototype verification FPGA cluster according to claim 1, characterized in that: The high-speed serial data link is a physical channel formed by connecting all FPGA nodes in series through a pair of LVDS TX signals and a pair of LVDS RX signals.
5. The high-speed serially expanded ASIC prototype verification FPGA cluster according to claim 1, characterized in that: Each slave FPGA node high-speed serial interface has PRBS-based self-detection capability.
6. The high-speed serially expanded ASIC prototype verification FPGA cluster according to claim 1, characterized in that: The data frame also includes a fixed frame header field.
7. The high-speed serially expanded ASIC prototype verification FPGA cluster according to claim 1, characterized in that: The monitoring of the design to be tested from the FPGA node ASIC prototype verification includes cyclic redundancy check monitoring.
Citation Information
Patent Citations
Design segmentation method for multiple FPGAs
CN112183007A
Segmentation and verification method and device, electronic equipment and storage medium
CN113255273A
AXI Stream protocol high-speed parallel-serial conversion system based on Aurora improved link protocol
CN113886299A
High-performance hardware acceleration and algorithm verification system and method
CN113961505A
Multi-chip joint verification method and device for large-scale ASIC (Application Specific Integrated Circuit) chips
CN114860519A
