For the past 70 years, computers have been designed according to the von Neumann architecture, and data needs to be transferred back and forth between the processor and memory during runtime.
With the development of the times, this working model faces great challenges: in high-concurrency computing scenarios such as artificial intelligence, data transmission back and forth will generate huge power consumption; the current performance improvement rate of memory systems lags behind significantly. Due to the performance improvement speed of the processor, the limited memory bandwidth cannot guarantee high-speed data transmission.
On December 3, Kuai Technology learned that DAMO Academy successfully developed a new architecture chip. This chip is the world's first DRAM-based 3D bonded stacked storage and computing integrated AI chip, which can break through the performance bottleneck of the von Neumann architecture and meet the needs of artificial intelligence and other scenarios for high bandwidth, high-capacity memory and extreme computing power.
In specific AI scenarios, the performance of this chip is improved by more than 10 times, and the energy efficiency ratio is improved by up to 300 times.
In the context of the gradual slowdown of Moore's Law, the integration of storage and calculation has become a key technology to solve the bottleneck of computer performance.
The integrated storage and computing chip is similar to the human brain. It integrates the data storage unit and the computing unit, which can greatly reduce data handling, thus greatly improving computing parallelism and energy efficiency.
This technology was proposed as early as the 1990s. However, due to the complexity of the technology, high design costs and lack of application scenarios, in the past few decades, the industry has paid little attention to integrated storage and computing chips. research is progressing slowly.
The storage and computing integrated chip developed by DAMO Academy integrates multiple innovative technologies and is the world's first chip that uses hybrid bonding 3D stacking technology to achieve storage and computing integration. The memory unit of this chip uses heterogeneous integrated embedded DRAM (SeDRAM), which has the characteristics of ultra-large bandwidth and ultra-large capacity; in terms of computing unit, DAMO Academy has developed and designed a streaming customized accelerator architecture to conduct "end-to-end" recommendation system "Acceleration, including matching, coarse sorting, neural network calculations, fine sorting and other tasks.
Thanks to the innovation of the overall architecture, the chip achieves both high performance and low system power consumption. In actual recommendation system applications, compared with traditional CPU computing systems, the performance of storage and computing integrated chips is improved by more than 10 times, and the energy efficiency is improved by more than 300 times. The research results of this technology have been included in ISSCC2022, the top conference in the chip field, and can be applied to VR/AR, driverless driving, astronomical data calculation, remote sensing image data analysis and other scenarios in the future.
Zheng Hongzhong, a scientist at the Computing Technology Laboratory of DAMO Academy, said: "The integration of storage and computing is a disruptive chip technology. It naturally has the advantages of high performance, high bandwidth and high energy efficiency, and can be solved from the underlying architecture. In the era of Moore's Law, chip performance and energy consumption are issues. The chip developed by DAMO Academy closely integrates this technology with the scenario, achieving a perfect integration of memory, computing and algorithm applications."
It is reported that DAMO. The computing technology laboratory of the institute focuses on the research of chip design methodology and new computer architecture technology. It has achieved a number of leading results and published many papers at top conferences such as ISSCC, ISCA, MICRO, and HPCA.