12.4. Overview of Journey 5

Journey 5 (J5) is an AD-specific (autonomous driving specific) algorithm processor with low power consumption and high-performance launched by Horizon Robotics in 2021.

In terms of computing power, J5 is equipped with 8-core Cortex-A55 CPU, dual-core Vision P6 DSP, and dual-core BPU (Brain Processing Unit) algorithm processor. The BPU accelerator is designed with a new generation of Bayes architecture that provides 128 TOPS of BPU computing power.

Note

  • The BPU is an acceleration core developed by Horizon Robotics. With the scenarios requirements of autonomous driving in the design process, it performs collaborative optimization of software and hardware architecture at three aspects: algorithms, computing architecture, and compiler, which improves several times of the computing power performance with no change in power consumption.

  • The J5 processor has a built-in BPU core with Bayes architecture, which greatly improves the support for advanced CNN networks and reduces the DDR bandwidth usage. It also provides the capabilities such as real-time pixel-level video segmentation and structured video analysis.

In terms of image processing capability, the J5 processor includes two high-performance ISP image processing modules, each of which supports 2x4k/8M@30fps image processing, multi-frame exposure wide dynamic (HDR), and provides high quality images in terrible lighting scenes. The ISP module also supports various mainstream CFAs, such as RGGB, RCCC, RGBIR 2x2, RCCB, RCCG and RYYCy, etc.

The J5 ISP also supports features including black level correction, defective pixel detection and correction, spatial noise reduction, chromatic aberration correction, lens shading correction, demosaicing, etc. We also offer some proprietary ISP modification tools for the customers to adjust ISP parameters and image effects.

Please contact Horizon technical supports or sales if you have any questions.

For better usage of the J5’s algorithm capabilities, you will get complete development board kits, algorithm development platforms, and various AD solutions based on the J5.

You can refer to the following relevant sections for more information.

The following sections describes the system and peripherals of the J5 processor.

12.4.1. Processor

The functional block diagram of the processor is shown in the following figure:

../../../_images/j5_function_frame.png

  • CPU: J5 processor uses an octa-core Cortex-A55 as the main CPU.

  • BPU: J5 processor uses the dual-core BPU with Horizon Gen-3 Bayes architecture.

  • DSP: J5 processor contains two programmable Vision P6 DSP modules, which can be used for CV calculation acceleration, which can also be used for neural network computation. The DSP supports SIMD and VLIW computing architectures to improve computing efficiency, with frequencies up to 650 MHz. The DSP can be configured to 256 8x8 MACs for a total of 0.67 TOPS.

Note

The two built-in BPU cores in the J5 processor provide pretty flexible multi-core scheduling capabilities for model scheduling at the business level.

Single Frame & Single Core: By manually scheduling a single model inference to a single BPU core, the high-priority models for full-image detection can be physically isolated from the low-priority models for scenarios such as detection + classification.

In the case where resources such as DDR has not yet reached the bottleneck, single frame single core can guarantee the isolation of the scheduling and inference on two BPU cores.

Multi-threaded, Multi-frame, Dual-core: This scenario can double the inference FPS. However, for larger models, multi-threaded, multi-frame, and dual-core may not be able to double the FPS due to the consumption of dual-core scheduling.

For more technical details on how to efficiently use the dual-core BPU, refer to BPU SDK API Manual.

J5 processor launched by Horizon Robotics is designed in strict compliance with functional safety requirements and procedures, integrating multiple safety mechanisms for fast detection and response to hardware failures, which complies with ASIL-B standards.

12.4.2. On-processor Peripherals Interfaces

  • Support 4x 2.5G MIPI CSI RX

  • Support 2x 2.5G MIPI CSI TX

  • Support 1x DVP

  • Support 8x I2C

  • Support 4x SPI master and 2x SPI slave, with 1x SPI master and 1x SPI slave multiplexed

  • Support 4x UART

  • Support 4x CANF-FD

  • Support 2x 1Gb Ethernet, with eth0 supporting TSN

  • Support 1x PCIe Gen3 2-Lane

  • Support 16 LPWM and 2x PWM

  • Support 1x eMMC and 1x SD/SDIO

  • Support 1x OSPI to connect NOR/Hyper/NAND flash

  • Support 64bit LPDDR4/4X memory

  • Support 2x I2S