BeagleV-Ahead

• T-Head TH1520 AI system-on-chip processor
  • Quad-core XuanTie C910 open-source RTL RISC-V (RV64GCV ISA) CPU @ 2GHz
    • 3 issue, 8 instruction, 9-12 stage pipeline
    • In-order fetch, out-of-order launch, out-of-order completion and in-order retirement
    • 64KB I-cache, 64KB D-cache per core
  • AI NPU with 4TOPS INT8 @ 1GHz
    • Support for TensorFlow, ONNX, Caffe
    • Feature extraction accelerator
  • Image signal processor
  • H.264/H.265/JPEG video encoder and decoder
  • 2D accelerator
  • 3D Vulkan/OpenCL/OpenGLES accelerator (IMG?)
  • C906-based (RV64IMA[FD]C[V] ISA) audio processor subsystem
• 4GB LPDDR4
• 16GB eMMC
• BeagleBone AI-64 form-factor open-source development board
  • ~2xSuperSpeed-USB type-A host (stacked)~
  • SuperSpeed USB type-C dual-role with power input (no power output)
  • Gigabit Ethernet with speed/connection indicators
  • HDMI type-A connector
  • TBD SDIO WiFi
  • uSD card cage
  • 2xCSI ribbon cable connectors
  • 1xDSI ribbon cable connector
  • 2x46-pin expansion headers with 4-pin HS-USB extension
  • 2x8-pin "mikroBus shuttle" ribbon-cable connector for audio and sensors
  • 3-pin TTL-level debug UART connector
  • Fan power connector (fan is optional)
  • 6x LEDs (1 power, 5 programmable)
  • 3x push-buttons (power, reset, boot/user)
  • 5V barrel jack
• BeagleBoard.org Debian Linux image
  • Tracking mainline u-boot/kernel with T-Head patches as needed
  • Package feeds for custom binaries for TensorFlow Lite, etc.
  • Familiar and flexible graphical user interface environment
  • Self-hosted Visual Studio Code IDE via web-browser
  • USB-based virtual network for "single-cable development"