Today, the Chinese multinational technology company, Alibaba Dharma Institute, has announced that it has successfully developed the world’s first DRAM-based 3D bonded stacking storage-calculation integrated chip.
The official of Alibaba Cloud broke the news in a statement released. According to the official, the DRAM-based 3D bonded stacked storage-calculation integrated chip was the world’s first of its kind.
The storage-computing integrated chip is now the greatest approach to overcome the above challenges; it works similarly to the human brain, combining the data storage and computation units into one, substantially decreasing data handling and enhancing parallelism and computing efficiency.
This technology was proposed as early as the 1990s, but industry development on combined storage and processing chips has been slow in recent decades due to the technology’s complexity, high design cost, and lack of application scenarios.
Dharma Academy now plans to use self-developed revolutionary technology to alleviate the industry’s processing power issue.
The storage-computing integrated chip also has a wide range of applications in terminals, edge terminals, and the cloud. For example, the storage-computing integrated chip can benefit from high bandwidth and low power consumption in scenarios like VR/AR, unmanned driving, astronomical data processing, and remote sensing picture data analysis.
The Rationale for the Development: The drawbacks of traditional computer system design, such as power walls, performance walls, memory walls, and other issues, have increasingly surfaced with the development of artificial intelligence application scenarios.
Firstly, data processing consumes a lot of energy. The power required for data transfer from the memory unit to the computing unit is around 200 times that necessary for processing, hence the real energy consumption and time spent for computation are relatively low in traditional architecture.
Secondly, memory technology is lagging significantly behind processor technology. The processor’s computing power is currently increasing at a pace of 3.1 times every two years, but memory performance is only increasing at a rate of 1.4 times every two years. The latter’s performance significantly impacts data transmission speed, which is also regarded as the Achilles’ heel of traditional computers.
There are three technical paths to achieving storage and calculation integration:
To begin, processing near memory is handled by an independent computing chip that is separate from the memory chip.
Second, Processing In Memory: The calculating process is accomplished by an independent calculation unit housed inside the memory chip, and the storing and calculation units are separate.
Finally, Memory execution calculation (Processing With Memory): The calculation operation is completed by the storage unit inside the memory chip; the storage unit and the calculation unit are totally integrated, and there is no separate calculation unit.
In the long run, the storage-computing integrated technology will also become a key technology for brain-like computing.
According to Dharma Academy, the final test chip demonstrates the obvious benefits of this storage computing technology and architecture:
“By minimizing the distance between the storage unit and the computing unit, it can enhance bandwidth, reduce data handling costs, alleviate data processing bottlenecks, and completely match the bandwidth/memory needs of the data center recommendation system.” Dharma Said.
The results of this chip’s research have been published in the premier chip conference, ISSCC 2022. Dharma Academy aspires to dominate in-storage computing technology in the future and gradually optimize common applications and ecosystems.
