21st Century Business Herald reporter Zhang Sainan reports from Shanghai
With the rapid development of AI, its demand for computing power chips is increasing, which also drives chip manufacturers to continuously upgrade.
Recently, at the developer conference held by Lattice Semiconductor, it announced that it will continue to expand its product line and launch multiple new hardware and software solution updates, including the latest version of specialized solutions for artificial intelligence (AI), embedded vision, security, and factory automation, adding new features and functionalities.
"With the increase in throughput and direct ports of automotive and industrial data, existing products in the market may no longer be able to support the latest memory interfaces. Based on this background, and considering the computing power requirements of AI, we have optimized structures such as DSP (note: digital signal processing) to ensure low power consumption while optimizing applications in industry and automotive." Introduction by Xu Honglai, President of the Asia Pacific Region of Lattice Semiconductor Shanghai Co., Ltd.
It is worth noting that this developer conference also showcased the achievements of Lattice as the world's third-largest FPGA supplier in building an ecosystem in recent years, inviting relevant leaders from Nvidia, BMW, and Meta to attend and share their cooperation.
New upgrade for AI
Driven by new market demand, Lattice continues to expand its product line. This time, Lattice has launched two new innovative mid-range FPGA series products based on the Lattice Avant mid-range platform - Lattice Avant-G and Lattice Avant-X, respectively, for general design and advanced interconnection, providing low-power, advanced interconnection and optimized computing power for mid-range applications in the communication, computing, industrial, and automotive markets.
Among them, Avant-G general-purpose FPGA aims to achieve system scalability by providing seamless and flexible interface bridging and optimized computing. This device provides leading signal processing and AI, flexible I/O, supports a range of system interfaces, and offers a dedicated LPDDR4 memory interface of 2400 Mbps.
Avant-X advanced interconnect FPGA is designed to achieve high bandwidth and security, and its feature set can be customized according to customer needs for signal aggregation and high throughput. This device provides a total system bandwidth of up to 1 T/s and PCIe with hardcore DMA; Reg; The Gen 4 controller, along with a security engine for encrypting dynamic user data, provides quantum security encryption functionality.
The 21st Century Business Herald learned from the press conference that samples have been provided for two products, and the latest released design software, Lattice Propel and Lattice Radiant, also support both devices.
In addition to hardware product upgrades, Lattice has also released updates to four solution sets, including Lattice SensAI for AI, Lattice mVision for embedded vision, Lattice Sentry for implementing security functions, and Lattice Automation for automated factories. These updates include upgrading the accelerator engine to improve performance, expanding IP and reference designs, adding new security features, and implementing more industry standards.
Xu Honglai, Asia Pacific President of Lattice Semiconductor Shanghai Co., Ltd., explained to 21st Century Business Herald reporters that the core of the mid-range product upgrade launched this time is the "significant improvement in data processing capabilities.", "In terms of hardware structure, we have also made many improvements in DSP related computing power, not only increasing the number of DSPs, but also accelerating the improvement of DSP computing power with the introduction of new technologies. Next, we will update AI oriented solutions based on the Avant platform."
According to Pu Xiaoshuang, the on-site technical support director of Lattice, "We will continue to work hard and continuously launch new products in the field of edge AI. We will also provide different solutions for the interconnection between sensor generation and cloud, and strengthen the improvement of elastic security mechanisms to meet everyone's considerations for network security."
Large scale expansion of the ecosystem
It is worth mentioning that Lattice demonstrated its ability in ecological construction at this conference, with BMW, Meta, Nvidia and other manufacturers present to share. "Since 2018, Lattice's ecological environment has been continuously expanding. Due to continuous updates and an increase in solutions, our ecological environment has increased by five times. In addition to developing our own IP, we also have many IP partners, as well as establishing partnerships with other semiconductor manufacturers and some reference design platforms." Pu Xiaoshuang said.
What makes the outside world curious is how Lattice and Nvidia collaborate? Xu Honglai explained to 21st Century Business Herald reporters, Nvidia has built a platform using GPUs, but this chip design cannot optimize the integration of more sensors and requires more chip resources to be focused on computing power rather than interfaces. In this case, Lattice's chips play a role in expanding the input/output interfaces of its sensors to connect more other sensors or devices to achieve AI functions. In addition, at the edge, Lattice's FPGA can do some things Data preprocessing to reduce memory and costs
For Xu Hong, besides NVIDIA, Lattice has also cooperated with other GPU manufacturers, mainly due to the flexibility of FPGA. "Other semiconductor devices are not our competitors, but our partners."
For the collaboration with Meta, it will make the market more focused on the application of FPGA in the metaverse. Xu Honglai introduced to the reporter, "Our cooperation with Meta mainly focuses on servers, using our FPGA for hardware control of the server platform, mainly in terms of security, as well as some remote control cooperation."
Behind the expansion of Lattice's ecosystem is the development of the entire FPGA market. According to Future Market Insights, the global market size of FPGA chips is expected to reach approximately $13.9 billion by 2032.
Laidi Si introduced that from the development of programmable devices, there have been 5 billion FPGA devices on the market in the past decade, and their quantity will continue to grow in the future. Lattice's current application scenarios have expanded from traditional communication, network communication, and storage to industrial and automotive applications, and updated applications have also appeared in video image processing.
In order to meet new demands, since 2018, Lattice has achieved innovation in three aspects: first, improving chip speed, second, improving agility, and third, extending the lifecycle. In the future, the layout on AI will be of utmost importance to Laidi Si. "Whether in industry, healthcare, or automobiles, I believe that AI will be ubiquitous, which is definitely a key focus of our promotion, not only in devices but also in solutions," said Xu Hong.
"FPGA still has a large market in AI training due to its architecture characteristics. Some applications do not necessarily need GPU, but can be replaced by FPGA. Especially in edge computing, due to technology regulation and supply chain constraints, all FPGA suppliers may be an opportunity to use different GPUs, FPGAs, including some other devices to optimize their evaluation and application of AI computing power. With the rise of the entire AI, I believe that All kinds of devices can participate in the calculation He said.