Home /

Product Application

/Ingenic T40, focusing on the new SVIoT track

Ingenic T40, focusing on the new SVIoT track

August 24, 2021

"Ingenic is the first chip company to focus on the SVIoT track wholeheartedly, and provides full-stack core technologies such as chips, AI computing power, development platforms and solutions."


Recently, Ingenic brought a chip called T40 to the market for the first time around the concept of SVIoT.


SVIoT, stand out from AIoT


Smart Vision for the Internet of Things (SVIoT), namely SVIoT, Smart Vision for the Internet of Things (SVIoT).

As its name suggests, it is developed from the Internet of Things track.

SVIoT has the same origin with traditional IoT/AIoT, and both belong to the category of IoT, and both belong to the category of smart hardware, and both involve chips, products, algorithms, solutions, and networking.

However, SVIoT is also different from the latter.

IoT is dominated by wireless interconnection (WIFI/BT/...) and MCU, with small devices as the core; AIoT is a grand concept, and it seems that both mobile phones and PCs can be AIoT.

SVIoT focuses on vision and has clear positioning.

From the perspective of information acquisition, vision is the most important means of perception. About 80% of human information comes from the eyes. Although SVIoT is a subdivision of AIoT, as the main entrance to the AI era, computer vision has extremely rich application scenarios in the AI field, and it is also a track with great commercial value.

The visual industry is indeed the first to enter the explosive period. Smart home, digital city, smart doorbell, security monitoring, car regulation perception, industrial vision, video conferencing, 3D vision, visual medical, somatosensory entertainment, service robots, etc. Currently, the hottest typical markets all have SVIoT, and under focus, they stand out.

SVIoT takes visual imaging, video processing, AI algorithm analysis, controllable cost, low power consumption, end-to-cloud interconnection and big data as its core technologies. It develops faster and has higher technical content than traditional IoT.

SVIoT has higher requirements for chips. General MCU+WIFI cannot meet performance requirements, and the cost of mobile phone APs is too redundant. This track is suitable for companies that truly master the core IP and has extremely high requirements for strength and execution.

Ingenic is just such a chip company focusing on the SVIoT track. They provide integrated solutions from chips, to AI computing power, to development platforms, to algorithm integration.

Today, Ingenic's chips on the SVIoT track have been shipped on a large scale.


T40, born for SVIoT


Rooted in the field of intelligent vision for many years, Ingenic has designed a chip T40 specifically for SVIoT.

"In the field of AI security, T40 is the best substitute for mid-to-high-end products."

In battery cameras, Ingenic has already won more than 50% of the market. In Ingenic's view, this data is expected to increase with the support of this T40.

T40 specifically defines specifications for SVIoT:

1. XBurst2 dual-core CPU

2. Starlight surveillance-level dual-camera ISP

3. 4K (2169P) video encoding

4. Maximum 8T computing power AI-Engine

5. Built-in DDR

6. Support 1080P screen display

7. The maximum power consumption is less than 1.2W (including DDR)

1.jpg


With dual-core XBurst2 and T40 enhanced RISCV coprocessor, the startup speed, performance, power consumption, and first frame effect are further improved. The 8T computing power of the AIE NPU makes the T40 far ahead in the computing power of the same level of chips. 4K video and ISP capabilities, rich front and rear IO interfaces, completely cover a variety of differentiated vision applications.


The third-generation starlight-level ISP, high-quality imaging capability




Full-scene, all-weather, and high-quality imaging capabilities are the housekeeping skills of the T40.

Jerry, IC Director of Ingenic Video Division: In response to some industry pain points in effect, T40's ISP has made targeted enhancements. For example, light source correction, skin tone restoration, 3D noise reduction and anti-smearing, sharpening enhancement, ADR adaptive, global transparency enhancement, WDR/HDR, color noise reduction, strong light suppression, binocular synthesis, etc., T40 is all carried out A large number of effective breakthroughs.

2.2.jpg



T40 WDR effect, the face and perspective are clearly visible, and the details in the dark are rich;

3.3.png



T40 starlight night vision full color, brightness ≤0.01lux, smooth noise, rich colors and clear details.


High computing power and agile AI development platform




Up to 8T computing power, which is a big killer of the T40, thanks to the AI engine (AIE) behind it. It is reported that this computing power is the highest in the industry on current chips of the same level.

"This is derived from Ingenic's in-depth mastery of CPU/multimedia computing technology for more than ten years. AIE has achieved ultra-high computing power, ultra-low power consumption, and ultra-strong flexibility. This is a unique advantage that is difficult to replicate in the industry."

AI-Engine (AIE) is Ingenic's complete set of AI acceleration engines. AIE references a variety of DSA design concepts to achieve multi-dimensional acceleration of DL/CV algorithms:

4.4.jpg



The AIE specifications on T40 are as follows:

Up to 8T ops computing power, mixed quantification of int16/in8/int4/in2, and the measured utilization rate can reach more than 70%;

512bit SIMD vector acceleration instruction set, flexible to use;

Coprocessor unit, optimized for pre- and post-processing;

Efficient RAM pool, deeply optimized memory bandwidth throughput;

The measured operating power consumption is lower than 500mW, and the energy consumption ratio reaches the lowest level of 0.05W/T.

Magik is a full-stack development platform for end-side AI applications. In addition to the basic tool chain, it also contains other rich auxiliary development resources, and supports post-quantization and more advanced quantitative perception training (QAT).

5.5.jpg



Magik integrates model training, optimization conversion, and deployment reasoning, and provides tools such as model checker, tuning device, and performance analyzer:

Fully support mainstream frameworks such as pytorch/tensorflow/mxnet/caffe/onnx;

Support quantitative perception training (QAT) and post-quantization;

Support 2/4/8/16 precision mixed training and conversion optimization;

Open the full-process code of common algorithms and the backbones of the classic network.

The end-level algorithm itself presents diversity and fragmentation. Hefei Junzheng uses AIE+Magik to effectively break through the computing power bottleneck of end-side AI and obtain obvious product landing advantages:

The MAC utilization rate is increased by 1 to 2 times, and the inference speed is increased by 2 to 4 times;

The RAM/ROM consumption of algorithm operation is reduced by 50%~70%;

Bandwidth is reduced by 35%~85%;

Fever is reduced by 40%~80%.


Algorithm empowers, easy to use is king



When Ingenic algorithm director Sandy commented on Magik, he said: "Computer power and tools are just the foundation. It is the kingly way to truly apply the algorithm to various scenarios and obtain advantages in cost, efficiency, power consumption, etc.!"

It is understood that AIE+Magik has successfully served more than dozens of customers, including some well-known industry brands, traditional algorithm companies, industry monitoring companies, large Internet brands, telecom operators and start-up geeks.

Ingenic has united a series of algorithm partners to provide a complete algorithm empowerment scheme for various scenarios, which greatly accelerates the product landing process. Currently, Ingenic T40 is directly or indirectly cooperating with a series of algorithm suppliers, more than a dozen scene application manufacturers, and more than one hundred related projects.

With the cooperation of all parties, manufacturers based on T40 applications can quickly produce new algorithms. In this way, under the domino of industry, manufacturers can quickly cover new scenarios and empower new industries.

In traditional applications, such as vehicle recognition, license plate recognition, pedestrian detection, face recognition, text recognition, perimeter prevention, object detection, motion detection, etc. have been very mature. In addition, there are many new and innovative scenarios. In mass production, such as family and pet identification, gun and ball linkage, OCR, bone behavior analysis, etc.


6.6.png



Sandy said: "Magik has successfully served more than a dozen algorithm customers so far, and our customer's fastest record is to complete an algorithm transplant in two weeks."


How low can the ultra-low power design be?




Power consumption is an eternal topic in the field of smart hardware.

Low power consumption is one of Ingenic's long-term inherent technical genes. From the educational electronics/PMP in the early years, to the previous wearable watches, to the current SVIoT, the Ingenic chip has always been the leader in power consumption. T40 naturally follows this advantage.

T40's power consumption can still be controlled in an excellent range when the AI computing power is fully utilized. According to different scenarios, the measured power consumption data is as follows:


QQ截图20210824144716.png



Among them, the reasoning power consumption of the AI engine is less than 500mW, and the lowest energy consumption ratio has reached a shocking 0.05W/T level.

Regarding power consumption, Jerry said: "In the past few years, in terms of power consumption, the T series chips have not yet lost the PK."

With complete product attributes, mass production is still possible under core shortage? !


Based on T40, Ingenic provides a variety of complete solution references, software codes, third-party algorithms, and cloud platform resources, including AI smart security camera solutions, low-power camera solutions, gun-ball linkage solutions, VSLAM solutions, face panel solutions, Video conference program, etc. Unconsciously, high-quality service has become one of Junzheng's core competitiveness in the intelligent vision market.


Under the background of the shortage of semiconductor supply, Ingenics has adjusted some peripheral materials for related products that can be seamlessly switched to minimize the adverse effects caused by shortages.

At the same time, Ingenic makes every effort to ensure the safety of T40 supply. Regarding the supply situation, Mick, the marketing director of Ingenic China, said: "T40 just avoids the hardest hit area of upstream supply. We are very confident of large-scale supply in the future, and customers are welcome to actively open cases!"

With many advantages such as chip performance, AI computing power, development environment, ultra-low power consumption, mature solutions, and security of supply, T40 will move forward on the new SVIoT track.

We are Ingenic agents, welcome to consult, there are more Ingenic master chip articles to share "Ingenic T40 Smart Video Processor" "Ingenic T31(H.265+AI) Intelligent comprehensive Omni-balanced video processor"

Product Categories

New Products

Please tell us your needs

Indasina, a group corporation composed of factories, is committed to a variety of electronic products and the overall customized solutions.

partners 8
Partners7
Partners6
Partners5
Partners4
Partners3
Partners2
Partners1
Leave a message Request A Free Quote
If you are interested in our products and want to know more details,please leave a message here,we will reply you as soon as we can.