MicroPython Hands-on (01)——After the Spring Festival, I bought a K210 chip AI development board

eagler8

MicroPython Hands-on (01)——After the Spring Festival, I bought a K210 chip AI development board [Copy link]

 

 

Python's openness, simplicity, and integration are in line with the current development stage, which has rapidly promoted artificial intelligence, big data analysis, visualization, and various platform program collaboration. It has been five or six years since the release of Python 3. It took too long from the opposition when it was first released to its gradual acceptance and popularity. However, it may also have a lot to do with national conditions and development needs. In short, more and more people are starting to use Python.

MicroPython is Python running on a single-chip microcomputer. The official development board is PYBoard, but this board is relatively rare and expensive. Fortunately, MicroPython supports a variety of development boards, and we can run MicroPython on other development boards very well, and the effect is the same. Based on 32-bit ARM processors, such as STM32F4 and F7 series, it also supports cc3200, esp32 and esp8266 (common wifi modules-can be used for Internet purposes in the future), Raspberry Pi, Banana Pi and BBC Micro:bit development board, etc.

I like Toutiao and saw an ad by chance. Half a month ago, I paid 8.9 yuan and signed up for a four-day Python training course on Toutiao. Haha, I finally got a start from zero (still got some results, see https://www.sohu.com/a/381128744_120248280 ). Three days ago, I searched for this Electronic Engineering World Forum and immediately registered an ID. I found that this may be the largest platform for learning and communicating MicroPython.......

After the COVID-19 outbreak during the Spring Festival, I bought a development board that supports MicroPython and a domestically produced K210 chip with independent intellectual property rights. I started learning from scratch and tried to do experiments.

This content is originally created by eagler8 , a user of EEWORLD forum. If you want to reprint or use it for commercial purposes, you must obtain the author's consent and indicate the source

eagler8

On September 6, 2018, Canaan Technology launched the world's first mass-produced commercial edge intelligent computing chip based on RISC-V, the K210, which was independently designed and developed. The chip relies on the AI neural network accelerator KPU, which is completely independently developed. It has three major features: independent IP, audio-visual and programmable capabilities, and can fully adapt to the needs of multiple business scenarios. As an edge AI chip independently developed by Canaan Technology, the K210 has both high energy efficiency and flexibility. In terms of computing power, the K210 can provide 1TOP computing power support under the condition of 0.3W, which is fully adapted to the computing power requirements under low power consumption constraints in most business scenarios.

eagler8

In terms of chip integration, Kanzhi K210 adopts an integrated audio-visual design. In terms of machine vision, the chip is based on the self-developed neural network accelerator KPU, which can complete image classification tasks based on neural networks, perform face recognition and detection, and classify the detected targets in real time. In terms of auditory ability, the chip comes with an APU voice processing unit, which can support up to 8 channels of audio data and 16 directions, and can realize functions such as sound source orientation, sound field imaging, beamforming, voice recognition and wake-up without occupying the CPU.

In terms of algorithm customization, Kanzhi K210 shows higher flexibility in programmability. First of all, compared with ARM and other architectures, Kanzhi K210 adopts RISC-V architecture, which has stronger customizability, making it easier for developers to customize algorithms according to specific application scenarios. Secondly, the chip is equipped with FPIOA field programmable IO array, supports mainstream AI programming frameworks such as TensorFlow, Keras, Darknet, PaddlePaddle and Caffe, as well as comprehensive development documentation, which is very friendly to developers. In addition, the chip has a built-in 64-bit dual-core processor architecture, which is divided into computing cores and application cores, which can provide developers with computing resources sufficient to cope with complex business scenarios.

eagler8

K210 system architecture
K210 includes a RISC-V 64-bit dual-core CPU, each core has an independent built-in FPU. The core functions of K210 are machine vision and hearing, which includes a KPU for calculating convolutional artificial neural networks and an APU for processing microphone array input. At the same time, K210 has a fast Fourier transform accelerator that can perform high-performance complex FFT calculations. Therefore, for most machine learning algorithms, K210 has high-performance processing capabilities. K210 has built-in AES and SHA256 algorithm accelerators to provide users with basic security functions. K210 has high-performance, low-power SRAM and powerful DMA, and has excellent performance in data throughput. K210 has a wealth of peripheral units, namely: DVP, JTAG, OTP, FPIOA, GPIO, UART, SPI, RTC, IS, IC, WDT, Timer and PWM, which can meet a large number of application scenarios.

eagler8

Based on the RISC-V open source instruction set
RISC-V (pronounced "RISC-FIVE") is an open instruction set architecture (ISA) based on the principle of reduced instruction set computing (RISC). V stands for the fifth generation of RISC (reduced instruction set computer), indicating that there have been four generations of RISC processor prototype chips before. Each generation of RISC processors was completed under the leadership of the same person, Professor David A. Patterson of the University of California, Berkeley. In contrast to most ISAs, the RISC-V ISA can be used for free in all desired devices, allowing anyone to design, manufacture and sell RISC-V chips and software. Figure 1 shows the four generations of RISC processor prototype chips before. Although it is not the first open source instruction set (ISA), it is important because it is the first instruction set architecture designed to select the appropriate instruction set according to the specific scenario. Based on the RISC-V instruction set architecture, server CPUs, home appliance CPUs, industrial control CPUs and CPUs used in sensors smaller than fingers can be designed.

Compared to most instruction sets, the RISC-V instruction set can be freely used for any purpose, allowing anyone to design, manufacture, and sell RISC-V chips and software. While this is not the first open source instruction set, it is significant because its design makes it suitable for modern computing devices (such as warehouse-scale cloud computers, high-end mobile phones, and tiny embedded systems). The designers took into account performance and power efficiency in these uses. The instruction set also has a large number of supporting software, which addresses the usual weaknesses of new instruction sets. The project started at the University of California, Berkeley in 2010, but many contributors are volunteers and industry workers outside the university. The RISC-V instruction set was designed to be small, fast, and low-power, but it is not overly designed for a specific microarchitecture. As of May 2017, RISC-V has established the userspace instruction set (userspace ISA) of version 2.22, and the privileged instruction set (privileged ISA) is also in draft version 1.10.

eagler8

Kanzhi K210 becomes a model for supporting RISC-V NOMMU
When it comes to the embedded field, RISC-V may be a knowledge point that you absolutely cannot avoid. Although ARM is still the dominant player in the mobile terminal, RISC-V has been given too much meaning by the outside world. The architecture itself is suitable for promising embedded development, and its free and open source features have also attracted much attention in the current international environment, making it a new Internet celebrity in the mobile terminal field. RISC-V is a general trend, and some chips and development modules based on this architecture have emerged in the market, such as the Kanzhi K210 that many developers are using. At the end of last year, a senior engineer named Jean-Luc shared the work of deploying the Linux5.1 system on the K210 processor. At the same time, he also quoted Western Digital's sharing in the article, and Kanzhi K210 was included as a demonstration of supporting RISC-V NOMMU.

In the field of AI, neural networks belong to computing-intensive scenarios, and algorithms such as face recognition are often limited by the power consumption of edge application scenarios. Jianan said that ARM is often accompanied by a lot of energy consumption in the actual computing process, and it is not a very economical architecture. According to public information, Jianan is one of the top five manufacturers in the RISC-V Alliance that use the RISC-V architecture. In the research and development of the Kanzhi series of AI chips, Jianan used the RISC-V RocketChip, which reduced a lot of related workload, saved a lot of R&D manpower costs, and the cost of IP licensing. At present, the Kanzhi K210 and its development modules are favored by many developers and have attracted the attention of top domestic AI teams such as Baidu and Alibaba. For example, Jianan customized the development module PaddlePi-K210 for Baidu's AI development platform PaddlePaddle, opening up the PaddlePaddle model device-side deployment solution. Developers do not need to change the hardware, and can use the public version mold to achieve the sample stage. In the exploration of cutting-edge projects, the Kanzhi K210 is compatible with Alibaba's latest TinyML algorithm model and is an RV platform that can be used to explore TinyML scientific research projects.

eagler8

Maixduino
is based on the MAIX module and is a RISC-V 64 development board for AI + IoT applications. Different from other Sipeed MAIX dev. The mainboard Maixduino is designed in Arduino Uno form factor, with ESP32 module and MAIX AI module onboard. MAIX is a product series specially designed by Sipeed, designed for running AI at the edge. Move AI models from the cloud to devices at the edge of the network, running faster, cheaper and with higher privacy on these devices.

eagler8

eagler8

eagler8

eagler8

eagler8

• CPU: Dual-core 64-bit RISC-V with FPU; 400MHz neural network processor
• QVGA@60FPS/VGA@30FPS Image Recognition
• Onboard ESP32 module supports 2.4G 802.11.b/g/n and Bluetooth 4.2
• Arduino Uno form factor, Arduino compatible interface
• Onboard omnidirectional I[size=75%]2 S digital output MEMS microphone
• 24P 0.5mm FPC connector for DVP camera
• 8-bit MCU LCD 24P 0.5mm FPC connector
• Machine Vision Based on Convolutional Neural Networks

eagler8

• Support self-flying micro SD card holder
• Reset and start buttons; 3W DAC+PA audio output
• Connect the USB Type-C cable to complete the download
• High-performance microphone array processor for machine hearing
• Supports MaixPy IDE, Arduino IDE, OpenMV IDE and PlatformIO IDE
• Supports Tiny-Yolo, Mobilenet, and TensorFlow Lite for deep learning

eagler8

Standard Sipeed_OV2640 camera

OV2640 main parameters
Can support customized FPC length, lens angle (70-160 degrees)Can support customized FPC length, lens angle (70-160 degrees)
Photosensitive array 1632x1232 Maximum format UXGA
IO voltage 1.7V-3.3V Analog voltage 2.5-3.0v (internal LDO to power the core 1.2V)
Power consumption working TBD sleep <20μATemperature
operation -30℃ to 70℃
Stable operation 0℃ to 50℃
Output format (8-bit) YUV/YCbCr4:2:2 RGB565/555/444 GRB4:2:2 Raw RGB DataOptical
size 1/4"
Field of view 70 degrees
Maximum frame rate 15fps SXGA
Sensitivity 1.3V/(Lux-sec)
Signal-to-noise ratio 40 dB
Dynamic range 50
dBBrowsing mode progressive
Electronic exposure 1 line to 1247 lines
Pixel area 2.2μm x 2.2μm
Dark current 15mV/s at 60℃
Working current 40mA

eagler8

Standard ST7789 driver chip 2.4 inch LCD screen (24P 320X240)
1. Module name: LCD display module
2. Model: KD024C-4
3. Models of the same type:
4. Compatible models:
5. Display mode: TFT
6. Display color: 65/262K
7. Resolution: 240*320
8. Dot pitch: 0.153 (H) x 0.153 (V)
9. Viewing angle: 12:00
10. Control IC: ST7789V
11. Display type: Fully transparent, normally white
12. Dimensions: 42.72*60.26*2.6mm
13. Visible area: 38.32*50.56 mm
14. Dot matrix area: 36.72*48.96mm
15. Brightness: 300cd/m2
16. Contrast ratio: 500
17. Interface type: 8/9/16/18-bit 8080 parallel port
16/18-bit RGB interface
3/4-wire SPI interface
18. Number of pins: 24
19. Pin distance: 0.5mm
20. Connection type: FPC plug-in type
21. Working voltage: 3.3V
22. Backlight color and type: White LED backlight
23. Backlight circuit: 4 LEDs in parallel, common anode
If=80mA, Vf =3.2V
24. Service life: 100000h
25. Working temperature: -20----70°C
26. Storage temperature: -30----80°C
27. Quality system certification: ISO9001:2008
28. Product certification: RoHS

eagler8

9zhmke

It looks very high-end and the price is reasonable. Can this microphone array achieve the same voice recognition effect as Tmall Genie X1?

eagler8

9zhmke posted on 2020-3-29 20:00 It looks very high-end and the price is also OK. Can this microphone array achieve the voice recognition effect of Tmall Genie X1?

The standard configuration is this one, 6+1 microphone array I2S interface directly output SEN0325 digital silicon microphone chip MEMS, it is estimated that it cannot reach the level of Tmall Genie, the difference lies in the platform and algorithm

eagler8

Introduction

The microphone array uses the MSM261S4030H0 high-sensitivity digital silicon microphone chip to form a 6+1 microphone array, which can effectively solve practical problems such as noise suppression, echo suppression, reverberation, single or multiple sound source positioning, sound source number estimation, source separation, etc. Through the I2S interface output, the onboard SK9822 is connected in series with the RGB LED, and the color is controlled by two lines.

MSM26S4030H0 is a microphone chip with omnidirectional port and I2S digital output MEMS. I2S is a universal digital audio interface, widely used in DSP and digital audio processing. In traditional audio systems, microphones usually collect audio signals and convert them into analog voltage signals for output. After being converted into digital signals by the ADC of the Codec and encoded, the main control chip processes the audio and inputs the digital audio signals into the main control chip. The MSM261S4030H0 series products have a complete 24-bit I2S audio interface built in. There is no need to add an additional Codec. It can directly connect to the DSP or MCU with fully digital signals, and natively input the digital audio signals into the main control chip, greatly reducing the complexity of signal chain design and the overall cost. It has great advantages in terms of system size, cost, power consumption and anti-interference.

In the consumer application market, the future will be towards personal portable products. This innovative microphone made of silicon material draws on the advantages of semiconductor process technology, so the microphone produced combines high production repeatability, excellent sound performance and flexible expansion performance in the future. Therefore, it is an inevitable trend for low-cost, high-performance MEMS to replace ECM (Electret Condenser Microphone) in the future.

Features

• MEMS microphone: 6 MSM261S4030H0 in array

Sound pressure level: 140 dB

Sensitivity: -26 (dB, dBFS @ 1KHZ 1pa)

Signal-to-noise ratio: 57dB (20kHz bandwidth, A-weighted)

• Output mode: digital I2S output
• Lighting: 12 LEDs form a ring light array, 256 levels of color adjustment, 32 levels of brightness adjustment
• Viewing Angle: 120 degrees
• Two-wire synchronous selection of positive output or negative output RGB three-color LED output
• Each LED consumes 20mA of current per color, and 60mA of current at full color and maximum brightness

Technical specifications

• Dimensions: 78.1mm×88.8mm
• Input voltage: 3.3V±0.2V(DC)
• Input current: >750mA (full brightness)
• Connection method: Support 2*5P 2.54mm terminal and 10P 0.5mm FPC connector
• Working temperature: -30℃~85℃

led2015

This board looks great, it's a good opportunity to learn