Recently, Axera, an intelligent perception and edge computing chip platform company, has appeared at several heavyweight industry conferences, bringing a rich product matrix and advanced technological layout. Liu Jianwei, Co-founder and Vice President of Axera, has given the latest judgment on the application and development of processors in the intelligent era, and introduced Axera's industry layout and implementation achievements in scenarios such as smart AIoT, ADAS and edge intelligence.

When intelligence becomes ubiquitous, how should enterprises respond to this trend? At the End-side AI Ecosystem Seminar hosted by Arm China at the end of 2024, Liu Jianwei stated that with the explosion of computing power, AI will become a stable and reliable intelligent service infrastructure. "Today's intelligence will be like electricity in the past, which can improve the operational efficiency of various industries and transform the original production processes. In this process, the edge and end sides have also become important carriers of AI." He believes that the accelerated implementation of end-side AI such as intelligent cameras, smartphones, and AI PCs has allowed consumers to truly feel the power of intelligence. "It can be said that the explosion of edge and end-side intelligence has made AI truly ubiquitous."

In fact, artificial intelligence is essentially a cost-driven productivity revolution. While it has greatly reduced the overall operating costs, enterprises also need to consider how to embrace intelligence at a lower cost. Therefore, Axera believes that processors in the AI era need to have a matching instruction set, and should focus energy consumption on data computing as much as possible. The principles designed based on these two points are the operator instruction set + DSA microarchitecture, which can ensure that enterprises can respond to network changes. Based on these two considerations, Axera has launched the more targeted AxeraNeurons AI Processor.

The core of the AxeraNeurons AI Processor is the operator instruction set and dataflow microarchitecture. The underlying layer adopts a programmable dataflow microarchitecture to improve energy efficiency and computing power density. At the same time, its flexibility ensures the completeness of the operator instruction set, supporting various AI applications.

Nowadays, when enterprises introduce intelligent chips, they not only consider computing power, but also have higher requirements for toolchains. At ICCAD-Expo 2024, Liu Jianwei introduced the complete toolchain of the AxeraNeurons AI Processor. He stated that after four to five years of polishing, AxeraNeurons has a mature toolchain. Customers can download it from the network and then reproduce performance and demos with one click, truly achieving quick onboarding within one hour.

At present, the coordination between processor software and hardware has not yet reached a highly efficient state, and enterprises often need joint software and hardware design. In this regard, Liu Jianwei pointed out that enterprises need to make reasonable disassembly when integrating software and hardware, and try to do "subtraction". Simpler software and hardware can present customers with a user-friendly and efficient integrated solution. Ultimately, it is hoped that when the instruction set matures, software design can also be completely separated like today's GPUs and CPUs.

With its proactive layout for the Transformer framework, Axera has a clear plan for its AI processor product line: the AxeraNeurons series of AI processors has covered high, medium and low tiers of computing power, and is widely used in fields such as smart cities, intelligent driving, and smart AIoT, helping enterprises with different computing power needs obtain AI-specific chips suitable for their own business development.

In the smart city scenario, Axera's AI processors solve the problem of high deployment costs for intelligent applications in a series of long-tail scenarios such as data collection by efficiently running large models; in the intelligent driving field, the M76H series products can also perfectly support the BEV multi-task model in intelligent driving applications; in terms of smart AIoT, Axera's global shipment volume has already ranked among the forefront.