K230 Product Brief

Preface

Purpose

This document is intended to provide a general overview of the K230 platform and its features.

Intended Audience

The main people to whom this document (this guide) applies:

• Design and maintenance personnel for electronic products

• Sales personnel for electronic products

• Technical support personnel

Terms

Revision history

Version Number	Author	Revision Date	Revision Notes
V1.0	Zhenming Huang	2023-6-10	Re-format to markdown document based on K230 Brief Datasheet_Complete_Apr_2023_v1

Overview

The K230 chip is the latest generation SoC product in Canaan Technology’s (NASDAQ: CAN) Kendryte® series of AIOT chips. The chip adopts a brand-new multi-heterogeneous unit acceleration computing architecture, integrates two RISC-V C908 computing cores, and has a built-in new generation KPU (Knowledge Process Unit) intelligent computing unit. Multi-precision AI computing power with INT8 and INT16. Supports general AI computing framework. The chip also has a variety of peripheral interfaces, as well as multiple scalar, vector, graphics, and other dedicated hardware acceleration units such as image 2D engine, AI 2D engine, 2.5D GPU, and 3D depth engine. K230 provides full-process computing acceleration for various computing tasks such as images, videos, audio, AI, etc. The chip has many features such as low latency, high performance, low power consumption, fast startup, and high security. K230 can be used in various smart hardware products, such as smart door locks, home smart security cameras, dictionary pens, payment recognition, 3D structured light vision modules, drones, interactive robots, smart home appliances, smart toys, and smart manufacturing, intelligent vehicle cockpit and many other fields.

General Specifications

CPU

• CPU 1: 64bit, RISC-V processor, 1.6GHz, supports RISC-V Vector Extension 1.0 specification, with floating-point units (FPUs), 128bits vector-process-unit, runs RTOS system

• CPU 0: 64bit, RISC-V processor, 800MHz, supports RISC-V 64GCB instruction set, runs LINUX system

KPU

• Support INT8 and INT16

• Support weight sparse compression

• Typical network performance：

  • Resnet 50 ≥ 85fps @INT8

  • Mobilenet_v2 ≥ 670fps @INT8

  • YoloV5S ≥ 38fps @INT8

• Support TVM / TensorFlow / Pytorch / ONNX

• Support quantitative tool chain and profiling tools

• Loss of quantization accuracy <1%

2D Engine

• 2D(GDMA) Engine:

  • OSD/CSC/X-Mirror/Y-Mirror/Rotation（90/180/270）

  • Typical image rotation capability：

    • 2 x 1080*1280 YUV400 @15fps + 1 x 1080*1920 YUV420 @30fps

• Independent AI 2D Engine:

  • Affine/Crop/Resize/Padding/Shift

DPU (Depth Computing Unit)

• Using 3D structured light depth calculation

• Support multiple operators, including Img_check / LCN / SAD / Post_proc / Align / Disptodepth

• The maximum resolution of the landscape image is 1920*1080, and the maximum resolution of the portrait image is 1080*1440

• Typical performance is 1280*800@30fps, 1280*1080@15fps,1920*1080@9fps

Memory

• DRAM: DDR3L / LPDDR4 with 16/32bit, LPDDR3 with 32bit

• DRAM frequency：

  • DDR3L@1600Mbps / LPDDR3@2133Mbps

  • LPDDR4@2667Mbps

• Maximum capacity: 2GB

• SIP mode: 128MB 16bit LPDDR4@3200Mbps

• SRAM：2MB+2MB (default for KPU)

Video Input

• 3 x MIPI CSI

• (1 x 4lane+1 x 2lane) or 3 x 2lane

ISP

• Total throughput：8MP@30fps

• Image 3A / 2DNR / 3DNR / WDR / HDR

• Local Tone Mapping

• Sharpen / BLS / LSC / De-warping

• RGB-IR 4x4 pattern

Video Output

• 1 x MIPI DSI, 1 x 4 lane or 1 x 2 lane

• Resolution: 2MP@60fps

Video Codec

• Video encoding

  • H.264 Baseline / Main Profile / High Profile

  • H.265 Main Profile

  • JPEG / MJPEG

  • Maximum encoding performance:

    • 8MP@20fps

  • CBR / VBR / CQP / ROI encoding

• Video decoding

  • H.264 Baseline, Main Profile, High Profile

  • H.265 Main Profile

  • JPEG / MJPEG

  • Maximum decoding performance:

    • 8MP@40fps

2.5D GPU

• Highly optimized Tessellation Engine, cooperates with the VGlite driver to provide vector graphics drawing functions

• Support rotation, scaling, translation, and format conversion of the input image

• The drawn picture can perform operations such as blending, compression, etc., with the picture in video memory

Audio

• 2 x DAC: 8-192Khz

• 2 x ADC: 8-192Khz

• Support ALC

• Maximum 8 x PDM DMIC inputs

• I2S interface supports 2x2 expansion

Peripherals

• Low-speed interface

  • 5 x UART / 5 x I2C / 6 x PWM

  • GPIO (multiplexed with function pins):

    • 64 x GPIO + 8 x PMU GPIO

    • 50 x GPIO + 6 x PMU GPIO(SIP)

• High-speed interface

  • 2 x USB 2.0 OTG

  • 2 x SDxC: SD3.01, eMMC 5.0

  • 3 x SPI: 1 x OSPI + 2 x QSPI

• WDT / RTC / Timer

Security

• PUF / OTP / TRNG

• Support AES, SHA, RSA, SM2/3/4, ECC

PMU

• Ultra-low power consumption: in deep sleep mode, standby power consumption ≤ 20uW

• Support RTC function in 32KHz

• Support long and short press detection as system shutdown or wakeup signal

• Support IO edge and level detection to generate interrupts

• Support two output IOs to control PMIC and system reset

• Support 512 bits system log when shutdown

Other Features

• Decompression module:

  • decompression bandwidth ≥ 400MB/s, GZIP format

• 4096-point FFT / IFFT unit

• Fastboot: 3A first picture ≤ 400ms

• Working environment temperature: -40 ~ 85°C

• Package：BGA, 0.65mm pitch

  • 13 x 13mm

  • 11 x 11mm (SIP 128MB 16bit LPDDR4)

Block Diagram

Typical Applications

• 3D Face Lock Integrated Application Solution

• Smart Dictionary Pen Application Solution