Mid and Low End FPGAs Become Important Deployment Components

Currently, the industry's attention to AI (artificial intelligence) chips is focused on GPUs or dedicated AI gas pedal chips (such as NPUs and TPUs). However, AI is actually not limited to high performance computing with large computing power, and there is also a huge market space for AI deployed to the endpoints, and thus a considerable number of components that can run AI loads, including FPGAs, which are widely used in the embedded space.

The company with the largest FPGA shipments in the world is currently Lattice. This company has maintained its high earnings growth during the industry downturn of the past two years, and just last month reported full-year 2023 revenue of $737 million, up 12% year-over-year; gross margin of 69.8%. 

At the 2024 Lattice Customer Technology Exchange Conference, Lattice Semiconductor Chief Strategy and Marketing Officer Esam Elashmawi said that Lattice is the world's largest FPGA shipment company, and very early layout of the AI market. The edge AI computing process requires the integration of a lot of sensor data, and the data is ultimately handed over to NVIDIA GPUs to do AI calculations." In this process, Lattice's FPGAs play a role, "Our FPGAs can communicate with different kinds of multi-sensors and translate them into CUDA language and code that NVIDIA GPUs can understand."

Most attention to FPGA and Lattice the company should be very clear, Lattice's FPGA chips, and Intel (Altera) and AMD (XILINX) to focus on the direction of high computing power is different, more inclined to small low-power market. At the end of last year Lattice and NVIDIA announced a partnership in the edge AI market: introducing a new sensor bridge design for AI applications using NVIDIA's Jetson Orin and IGX Orin platforms to accelerate the development of edge AI applications.

As can be seen on the Lattice website, the company's robust portfolio of offerings includes the full-featured Lattice sensAI. Lattice sensAI provides everything needed to evaluate, develop and deploy FPGA-based machine learning and AI solutions - modular hardware platforms, demo examples, reference designs, neural network IP cores, software development tools and custom design services.

An important feature of FPGA devices is the programmability feature, or flexibility, with which developers can create multiple parallel computing pipelines and allocate storage blocks to optimize data transfer. However, FPGAs have a certain design threshold, so it is still necessary to FPGA vendors to provide some help. Esam Elashmawi pointed out that Lattice now has the strongest product portfolio ever, will continue to invest in and introduce a new generation of smaller FPGAs and mid-range FPGAs, and will continue to invest more resources in software tools and solutions.

Frank Xie, senior director of application engineering at Lattice Asia Pacific, said in his presentation that Lattice does performance, power and connectivity better. And customers are looking for products that respond in real time because AI devices on the edge are very close to data collection. Lattice helps customers continuously optimize response time and power consumption, allowing them to achieve better AI capabilities.

The release of Lattice's low-power, mid-range Avant FPGA platform at the end of 2022 marks the company's entry into the mid-range FPGA market along with the Nexus FPGA platform, forms Lattice's "strongest portfolio ever". Avant builds on Nexus' existing market presence, doubles Lattice's SAM market, and allows its products to reach more applications. 2022, when Avant was born, Lattice said Avant was the next step in the company's quest to gain access to the $3 billion SAM market. 

Esam Elashmawi noted that compared to the Nexus FPGA platform, the Avant FPGA platform offers a 5x increase in capacity to 500K logic cells, a 10x increase in bandwidth, and a 30x increase in compute performance. According to the introductory information, Avant is based on a 16nm FinFET platform and offers powerful 25 Gb/s SERDES, hard-core PCI Express and external memory PHY interfaces, and a large number of DSPs for the latest AI/ML and computer vision algorithms.

Lattice Avant is a lower power, higher bandwidth and smaller FPGA product solution. The Avant-G is a general-purpose FPGA for mid-range applications, designed to enable system scalability by providing seamless, flexible interface bridging and optimized computation to meet a broader range of customer needs. Lattice Avant-G devices offer leading-edge signal processing, AI, and flexible I/O - supporting a range of system interfaces - along with a dedicated 2400 Mbps LPDDR4 memory interface.

Designed for high bandwidth and security, the Avant-X advanced interconnect FPGAs have a feature set that can be tailored to customer needs for signal aggregation and high throughput. Lattice Avant-X devices offer up to 1 T/s of total system bandwidth, a PCIe Gen 4 controller with hardcore DMA, and a security engine for encrypting dynamic user data, providing quantum-safe encryption.

Lattice is always trying to improve its ease of use and add more new features, and with the Nexus and Avant families of FPGA products, all of this makes up an ecosystem that is the strongest FPGA portfolio in Lattice's history, and it will get even stronger in the future. With a better FPGA platform coupled with comprehensive tool support, developers can very conveniently realize their AI innovations based on Lattice FPGAs.

If you're interested in this article, you can click through to see our other articles.