New Processor for Edge AI "AM62A Series"

Texas Instruments (hereafter, TI) offers a large number of microcomputers and processor products. Among them, this time we will introduce the latest processor "AM62A series" equipped with Arm® Cortex®-A53.

This product falls under the series of Sitara™ processors. It combines camera-based image processing with low consumption and edge AI functions such as object detection and object recognition, enabling the development of HMI (Human Machine Interface) devices with various analytical functions.

If you are interested in this product, please feel free to contact us from the bottom of the page.

Features of the AM62A series

AM62A series block diagram
*Click the block diagram to open the TI website.

AM62Ax functional block diagram
Source – Texas Instruments AM62Ax Sitara™ Processors datasheet
https://www.ti.com/jp/lit/ds/symlink/am62a7.pdf

The AM62A series is designed for vision applications requiring 1-2 cameras. It integrates an H.264/H.265V video codec, an image sensor processor (ISP) with RGB-IR support, and an AI accelerator, making it suitable for various vision applications.

We will introduce the features of this product and examples of actual usage, so please refer to the functional block diagram on the right.

Use cases in vision applications

Application requirements, etc., depend on the use case. Vision applications require uplinking video data and choosing an appropriate image sensor.

The built-in H.264/H.265 encoder/decoder provides high-performance encoding/decoding capabilities up to 4K@60fps, allowing H.265 encoded clips to be uplinked via network. In addition, the built-in ISP supports sensors with resolutions up to 8 megapixels, enables image processing functions such as denoising, demosaicing, defective pixel correction, gain and white balance, and offers a variety of image sensor choices.

The table below introduces application examples when using the AM62A series and use cases of the required specifications.

Application example

resolution

frame rate
[FPS]

camera
shutter type

Surveillance camera

2 to 8 megapixels

10~30

rolling

machine vision

5 megapixels or more

60~

global

infrastructure camera

1 to 5 megapixels

5~15

global

drive recorder

2-5 megapixels

15~30

global / rolling

sports camera

2 to 8 megapixels

60~

rolling

Barcode reader

1 to 5 megapixels

10~30

rolling

AI accelerator that can operate with low power consumption

A major feature of this product is that it has a built-in deep learning accelerator. It uses a combination of TI's C7x DSP and a custom Matrix Multiply Accelerator (MMA) for performance in Neural Networks (NN), especially Convolutional Neural Networks (CNN), which are common in vision AI. has improved significantly. When using C7x DSP and MMA and operating at a maximum of 1GHz, the maximum computing performance is 2TOPS.

Power consumption is less than 2W at full load at room temperature and less than 5W at 125°C, enabling use in cost and power constrained applications.

 

TI has prepared various tools for edge AI, so if you are interested, please refer to the following articles.

How to simplify your embedded edge AI application development

Real application example

Here are some real world application examples that leverage the features introduced above.

video doorbell

Responsiveness to theft and identity verification is important for video doorbells and home security systems. Therefore, it is necessary to analyze real-time video data locally to reduce false positives and to respond faster and more reliably. However, power and size constraints limited AI processing performance, sacrificing real-time responsiveness.

The AM62Ax, in an 18x18mm package, provides the performance required for video doorbells, as previously mentioned, allowing a higher level of human and object detection to be implemented.

smart retail

Smart retail is a new style of shopping where customers select their purchases and leave the store without paying at the cash register.

The application's system utilizes an AI model derived from object detection and a barcode scanner to identify the items a customer purchases as they add items to their shopping cart and ultimately exit the store.

Smart retail applications can process data locally to reduce response times during transactions and improve data security. Running AI models locally does not require a network connection to cloud resources (data is not sent to the outside world), so security against unauthorized access from outside can be improved.

As with video doorbells, power consumption is a major design challenge. The AM62Ax enables fruit and vegetable sorting, anomaly detection, orientation detection, and barcode identification while reducing power consumption.

Development environment

Hardware Evaluation Environment (EVM)

AM62A evaluation board has been released.

SK-AM62A-LP

Software Evaluation Environment (SDK)

A dedicated SDK (Software Development Kit) is provided.

PROCESSOR-SDK-AM62A

 

There is also a dedicated page for software developers, so even first-time users can start evaluation smoothly.

Academy for AM62Ax (8.6.0.0 v1)

 

Summary

The AM62A series is designed for vision applications requiring 1-2 cameras. It combines camera-based image processing with low power consumption and edge AI functions such as object detection and object recognition, enabling the development of HMI (Human Machine Interface) devices with various analytical functions.

Inquiry

If you have any questions about the "AM62A series", please contact us from the following. the Company product representative will get back to you.

Related information

Information on analog products

In addition to digital products, the Company also handle many analog products. Please take a look at the Company original catalog.