Let developers get out of the “code quagmire”, how does AI achieve “unification”?

In the past few years, artificial intelligence, which once only appeared in science fiction movies, broke through the high wall between virtual and reality and entered the life of ordinary people. From the smart recommendation in the mobile APP to the smart city system that can plan the operation of the entire city. Before we know it, AI has entered every corner of our lives.

And with the development of communication technology, faster Wi-Fi 7 and 5G are making connections everywhere. As digital transformation becomes an unstoppable trend, the application scenarios of AI are further expanding.

Nowadays, AI is rapidly expanding from the field of smartphones to new tracks, and AI has become indispensable in the fields of smart cars, XR, various smart hardware and IoT devices.

The rapid expansion of AI has made it hot and a must for manufacturers, and the wider application of AI has brought higher and higher development costs to developers.

Faced with this problem, Qualcomm recently released the Qualcomm AI Stack, an AI software stack. Qualcomm hopes to liberate the productivity of developers, make AI popular in more fields in an easier way, and promote the establishment of a unified technical route.

AI grows savagely, and developers are caught in a “code quagmire”

In the era of the Internet of Everything, the boundaries of artificial intelligence are expanding. AI has found many different applications in different fields. Today, AI has expanded from smartphones to more fields such as automobiles and the Internet of Things, and is playing an increasingly important role.

As an AI leader, Qualcomm, its AI technology has been widely used in many applications.

In the field of smartphones, Qualcomm’s AI technology has supported more than 1.8 billion terminals. Qualcomm’s AI engine has also continued to iterate, and its performance has become more and more powerful. The latest first-generation Snapdragon 8/Snapdragon 8+ platform is equipped with Qualcomm’s seventh-generation AI engine, and its AI performance is 4% higher than the previous generation Snapdragon 888 platform. times.

In the automotive field, Qualcomm cooperates with the world’s leading automakers to promote the implementation of AI in the field of smart cockpits. Qualcomm currently ranks first in in-vehicle connectivity, car connectivity and next-generation infotainment systems, and all major global automakers have chosen the Snapdragon cockpit platform.

The influence of artificial intelligence is expanding at an unprecedented rate.

As the boundaries of artificial intelligence become wider and wider, the work of developers is becoming more complex, and artificial intelligence developers are falling into a “code quagmire”.

In fact, although artificial intelligence covers such a broad field, the needs of different fields are different. There are still challenges to transfer AI in one area to another.

AI in smartphones, IoT, and smart cars differs in power consumption requirements, AI model types, and model deployment methods. At the same time, different scenarios have different requirements for model accuracy, power consumption and delay.

For example, when artificial intelligence is applied in the field of IoT, low power consumption and connection stability are the most important. Since IoT devices generally do not undertake more complex computing tasks, the priority of performance is not so high.

But in an intelligent driving system, in order to ensure traffic safety, the AI ​​model must have very high accuracy. And due to the ever-changing environment on the road, the requirements for AI must also have a low latency to respond to possible emergencies in a timely manner.

Different scenarios, different terminals, and different requirements bring difficulties to developers’ development work.

The needs of different scenarios make developers “tailor-made” for different scenarios. More and more application scenarios and terminal forms of AI are increasing the workload of developers.

As a leader in the field of AI, Qualcomm is building a “uniform” AI ecosystem to simplify the development process of AI and further accelerate the popularization of AI in order to make the work of developers easier.

Artificial intelligence “cars on the same track, books on the same text”

Qualcomm hopes to free developers from repetitive adaptation work, change the status quo that different scenarios require multiple development and adaptation, and achieve “one development, run anywhere”, so that developers can achieve more innovation.

To achieve this goal, there are many challenges. Ziad Asghar, vice president of product management at Qualcomm Technologies, pointed out that due to different needs in different fields and scenarios, AI needs to find a balance in terms of power consumption, accuracy, model type, delay, etc. This will be a very important task. complex work.

A more important challenge comes from IP permissions. Jeff Gehlhaar, vice president of technology at Qualcomm Technologies, talked about how to achieve this goal, how to make all the different IP have the same access rights, so that each IP can choose the corresponding solution in a unified way , is the main challenge to implement such a new development approach.

Qualcomm’s release of the first Qualcomm AI Stack, which integrates AI software development rights across all business lines, is addressing these challenges.

The Qualcomm software stack consists of four layers, which integrate mainstream operating systems, system software, developer libraries and AI frameworks respectively.

From the bottom-level hardware combined with the top-level software and tool optimization, Qualcomm has successfully realized that developers can run across hardware platforms only by developing a unified AI software stack.

The establishment of the AI ​​software stack also laid the foundation for the final realization of a unified SDK.

The Qualcomm AI software stack supports a number of SDKs, such as the Snapdragon Ride SDK for automotive ADAS solutions, the Qualcomm Smart Multimedia SDK for IoT, and the Snapdragon Spaces XR developer platform.

Through this cross-terminal capability, Qualcomm enables the development of artificial intelligence in different fields to achieve “cars on the same track, books on the same text”.

Different from the previous developers who had to develop separately for different hardware, now using the Qualcomm software stack, OEMs or developers only need to develop once, and then use the Qualcomm software stack for optimization in combination with specific use cases, and the development results can be applied to smartphones, In terminal devices in the fields of IoT, automotive, XR, cloud and mobile PC.

For example, successful use cases such as semantic recognition and face recognition that Qualcomm has developed in the AI ​​development of smartphones in the past can be applied to laptops, tablet computers and other terminals equipped with Qualcomm chips through the adjustment and optimization of the software stack. in the product.

The establishment of the software stack has saved developers a huge amount of effort, and this is only the first step taken by Qualcomm. Based on the software stack and its own leadership in multiple fields, Qualcomm has even greater ambitions: it hopes to build a unified technical route and further popularize the influence of AI.

The Internet of Everything promotes the “big unification” of the technology route

The significance of Qualcomm’s release of the AI ​​software stack is not only to simplify the development process and improve the efficiency of developers. Qualcomm has also taken a solid step towards building a unified technology route.

In a world where everything is connected, more and more devices are connected to each other through the network. It is becoming more and more important to build a platform that can plan them together.

At the 2021 investor conference, Qualcomm announced a strategic plan for a “unified technology roadmap.”

This strategic plan unifies the deployment of Qualcomm’s various product lines as a whole. It will start from Qualcomm’s core mobile phone business and eventually provide intelligent, high-performance and low-power systems and wireless combinations for various smart terminals.

Functionally, developers can deploy one-time development to terminals of different forms through the AI ​​software stack, which is consistent with the goal of unified deployment of various business lines established by Qualcomm.

Whether from a functional perspective or from a future planning perspective, the release of the AI ​​software stack will be an important step for Qualcomm to further expand the AI ​​boundary and build a unified technical route.

The powerful functions across terminals also provide convenience for developers and manufacturers to further popularize the application of AI.

We are living in an era of ever-faster and more connected connections.

Qualcomm is using its leadership in many fields to bring unity to massive innovations, and to build a platform that is easier to develop and has broader boundaries in the era of the Internet of Everything, leading the popularization of AI.

The text and pictures in this article are from Lei Feng.com