Based on architectural innovation, Houmo Intelligent Lights up the industry’s first large-scale computing power AI chip with integrated storage and computing

On May 23, Houmo Intelligent announced that its self-developed AI chip with integrated storage and computing power in the industry was successfully lit up and successfully ran through the intelligent driving algorithm model.

Based on architectural innovation, this chip uses SRAM (Static Random Access Memory) as an integrated storage and computing medium. Through the deep integration of storage units and computing units, it achieves high performance and low power consumption. The sample computing power reaches 20TOPS, which can be extended to 200TOPS, the energy efficiency ratio of the computing unit is as high as 20TOPS/W. This is the first AI chip in the industry based on a strict in-memory computing architecture, with AI computing power reaching tens of TOPS or higher, and supporting large-scale visual computing models. Compared with the large computing power chip under the traditional architecture, this chip has significant advantages in terms of computing power and energy efficiency ratio.

(The first chip of Houmo Intelligence lights up the scene)

The chip adopts a mature 22nm process, which can effectively control the cost while improving the energy efficiency ratio. In addition, in terms of flexibility, this chip not only supports mainstream algorithms on the market, but also supports different customers to customize their own operators, which is more suitable for high-speed iteration of algorithms.

(Homo intelligent chip development board)

In edge-side high-concurrency computing scenarios such as intelligent driving, in addition to high computing power requirements, there are also high requirements for chip power consumption and heat dissipation. At present, the performance improvement speed of the memory system in the conventional architecture chip design lags far behind the performance improvement speed of the processor. The limited memory bandwidth cannot guarantee high-speed data transmission and cannot meet the computing needs of high-level intelligent driving. Secondly, the data transmission back and forth will generate huge power consumption. Based on this chip, Houmo Intelligent has run through the multi-scene and multi-task algorithm model in the intelligent driving scene for the first time on the in-memory computing architecture, providing a new technical path for high-level intelligent driving. In the future, it is possible to provide P-level computing power with W-level power consumption to better meet the needs of the era of high-level intelligent driving.

(Homo Intelligence, based on its self-developed storage and computing integrated AI chip with large computing power, successfully ran through the intelligent driving algorithm model)

At the beginning of its establishment, Houmo Intelligent has firmly taken the road of differentiated innovation. It is also the first start-up enterprise in China to design high-power AI chips through the innovation of the underlying architecture. Any subversive innovation will face extremely high technical challenges. R&D personnel need to redesign circuits, cell arrays, tool chains, etc. according to traditional memory devices, and must break through various physical and structural technical problems. The successful lighting of the chip marks a key breakthrough of Houmo Intelligent in the engineering implementation of the integrated technology of large computing power storage and computing.

Dr. Wu Qiang, founder and CEO of Houmo Intelligence, said that the lighting of the industry’s first large-scale computing power AI chip is the best verification and recognition of Houmo’s technology and engineering capabilities. But this is only the first step we have taken. In the future, we will continue to maintain awe, continue to explore and innovate, deliver products with differentiated competitive advantages, and create greater business value for customers! “