“Lattice FPGA and software solutions enable designers to accelerate future-proof models using existing silicon. This article will explore some application examples of Lattice FPGAs and software solutions in computer vision and AI designs at the network edge.
ABI’s research shows that by 2024, the proportion of devices with on-device AI reasoning capabilities is expected to reach 60%. As evidenced by the rapid innovation in AI over the past few years, engineers need to develop more flexible design models in the transition from the cloud to the network edge. The drivers of this trend include the need for ultra-low latency, security performance, and bandwidth limitations and privacy protection.
Lattice FPGA and software solutions enable designers to accelerate future-proof models using existing silicon. This article will explore some application examples of Lattice FPGAs and software solutions in computer vision and AI designs at the network edge.
Why FPGAs are the best choice for network edge computing and AI applications?
FPGAs are inherently flexible and adaptable, making them ideal for edge computing and AI applications.
An FPGA is a parallel computing engine capable of operating at lower clock frequencies and therefore lower power consumption. In addition, the entire architecture also has flexible resources, including DSP, memory, distributed and interconnected programmable logic units, and has many similarities with those new ASICs dedicated to AI. However, unlike ASICs and other processors, the flexibility of FPGAs enables continuous optimization of existing use cases within the system and the introduction of entirely new use cases without requiring new hardware.
When we compare the design cycle of FPGAs to ASICs, we can see that system designers can use FPGAs for multiple iterations to quickly introduce new applications and bring them to market. And system designers using ASICs have to wait for the next iteration to reach the same level of performance, which not only delays time to market, but is also less efficient overall when the system needs to adapt.
At this year’s Embedded Vision Summit, Lattice showed a demonstration based on a CertusPro™-NX device that can run multiple AI engines and concurrent threads in parallel, reducing overall system latency and implementing a full system implementation Higher FPS can be achieved in .
Helping system designers accelerate the development of AI applications
In the second half of the presentation, I briefly describe how Lattice leverages its various software solutions, including the Lattice sensAI™ Solutions Collection, to help system designers (usually software developers, not FPGA experts) in development. Lattice sensAI includes a range of tools, hardware, acceleration IP, software tools, reference designs and demos, custom design services, and our end-to-end solutions built for AI applications in specific end markets.
Included in sensAI is our sensAI Studio design software, which enables system designers to complete use-case verification in hours instead of days or weeks. They can use sensAI Studio to import models already in the software model library or their own models, then perform transfer learning, evaluate the training of the model, collect and label data, configure and test the model, and then perform transfer learning for a specific device on the board. compile.
In the final Q&A section, I talked about how a Lattice agricultural customer used sensAI Studio to build an AI system that detects berries in the field. This example shows that tools like sensAI Studio are really important for system designers who want to build various AI applications, but lack the relevant experience.
Enabling the Next-Generation Smart PC Experience
Today’s PC users are increasingly interested in smart devices that are more sensitive, and they also need strong security to protect their privacy. In addition, they want to experience better collaboration features, such as a great conference call audio and video experience. Laptop system designers face various challenges in bringing different form factors to the market, which also creates system challenges; how do you transfer all this data from the camera to the rest of the system?
The Lattice AI-based solution I presented in my talk can help system designers address this, helping them develop entirely new capabilities, including presence detection, bystander detection, and face framing for video calls. Our solution also enables attention tracking, allowing users to spend up to 45% of the time away from the screen while using the computer, resulting in up to 28% additional battery life.
Contact us to learn more about how our software tools enable AI/ML capabilities at the network edge with low-power FPGAs.