KL730 (Kneron Edge AI SoC)

KL730 Architecture Block Diagram

Product Features

Flexible Low-Power Platform

  • Quad Core ARM�� CortexA55 CPU
  • Kneron�� 4th Gen KDP NPU up to 3.6eTOPS
  • Linux and RTOS SDK
  • 12-nm TSMC HPC+ CMOS process

Media & IO

  • High-Efficiency Video Decoding: H.265 and H.264 decompression
  • High speed IO: USB3.0, USB2.0 OTG
  • Memory: 32bit ??LPDDR4 I/F

AI Capability

  • Support Transformer / CNN / LSTM AI architecture
  • Support multiple AI models running

Security

  • FIPS 140-2 level 2 certified security module
  • DES, TDES, AES, SM2/SM3/SM4, SHA-1, SHA-2, ChaCha20-Poly1305, ECDH, ECDSA, RSA, EdDSA

Package

  • TFBGA361
  • Ball Pitch: 0.65mm
  • Dimension: 15mm x 15mm

Technical Specifications

Sensor & Video Input Interface

  • Support Bayer, RGBIR 4x4, YUV sensor
  • Support HDR Sensor
  • 2 channels MIPI sensor input (4-lane for each channel)
  • 4 channels MIPI sensor input (2-lane for each channel)
  • 4 channels MIPI sensor input (2-lane for each channel) + 2 channels DVP sensor input (16-bit for each channel)
  • Support Dynamic stitching

System Performance

  • VIC: 960MP/s
  • ISP & Encoder: 480MP/s, 12M@30fps
  • Built-in dewarping function

AI Capabilities

  • Multiple objects detection
  • Pedestrian detection and Re-Identification
  • Face recognition
  • Sound detection - Baby crying detection

Power Consumption Profile

  • Sensor: 8M@30fps
  • H.264 Video streaming: 8M@30fps + 2M@30fps
  • AI inference frame rate: 30fps, yolov5
  • SoC Power consumption: 2.3W

KL730 yields a 3 to 4 times leap in energy efficiency compared to previous Kneron models. It supports the most advanced lightweight GPT large language models.