AliBaba (NYSE: BABA) Announces Breakthrough in Chip Design
https://pandaily.com/alibabas-damo-...acked-in-memory-computing-chip-based-on-dram/
"
Alibaba’s DAMO Academy Successfully Develops World’s First 3D Stacked In-Memory Computing Chip Based on DRAM
Categories:INDUSTRY Pandaily Posted on December 3, 2021
On Friday,
Alibaba Cloud announced in a social media post that its DAMO Academy has successfully developed a 3D stacked In-Memory Computing (IMC) chip.
Alibaba Cloud claims this is a breakthrough that can help overcome the von Neumann bottleneck, a limitation on throughput caused by the standard personal computer architecture. It meets the needs of artificial intelligence (AI) and other scenarios for high bandwidth, high capacity memory and extreme computing power. In the specific AI scenario tested by Alibaba, the performance of the chip is improved by more than 10 times.
With the outbreak of AI applications, the shortcomings of the existing computer system architecture are gradually revealed. The main problems are that, on the one hand, processing data brings huge energy consumption. Under the traditional architecture, the power consumption required for data transmission from memory unit to computing unit is about 200 times of that of computing itself, so the real energy consumption and time used for computing are very low.
On the other hand, the development of memory technology lags far behind that of the processor. At present, the computing power of the processor is increasing at a rate of 3.1 times every two years, while the performance of the memory is only improved by 1.4 times every two years. The performance of data storage technology greatly affects the speed of data transmission.
Alibaba Cloud believes that IMC is currently the best way to solve the above problems. Similar to the human brain, IMC stores data in RAM and processes it in parallel, supplying real-time insights that enable businesses to deliver immediate actions and responses.
This technology was proposed as early as the 1990s. However, due to the complexity of technology, high design cost and lack of application scenarios, the research progress of IMC in the industry has been slow in the past decades. Nowadays, Alibaba’s DAMO Academy hopes to solve the industry problem of computing power bottleneck through self-developed innovative technologies.
SEE ALSO: Alibaba DAMO Academy Creates World’s Largest AI Pre-Training Model, With Parameters Far Exceeding Google and Microsoft
In addition, the IMC has broad application prospects. For example, in VR/AR, unmanned driving, astronomical data calculation, and remote sensing image data analysis, IMC can give full play to its features of high bandwidth and low power consumption.
At present, the research results of this chip have been included in the International Solid-State Circuits Conference 2022, the foremost global forum for presentation of advances in solid-state circuits and systems-on-a-chip. In the future, DAMO Academy hopes to make further breakthroughs in IMC technology and gradually realize typical applications to help build the technological ecosystem.
TAGS:
ALIBABA-DEVELOPED AI CHIP |
ALIBABA∗ |
CHIPS |
DAMO ACADEMY
"
AliBaba and its DAMO Academy will present their new technology at the following upcoming conference in San Francisco:
ISSCC 2022
2022 International Solid-State Circuits Conference
February 20-24, 2022
San Francisco Marriott Marquis Hotel
Info & Registration:
https://www.isscc.org/
Download the ISSCC 2022 Advance Program
According to the CONFERENCE TECHNICAL HIGHLIGHTS in the above draft program, the following infos are related:
On Sunday, February 20th, the day before the official opening of the Conference, ISSCC 2022 offers:
• A choice of 12 Tutorials, or
• A choice of 1 of 2 all-day Advanced-Circuit-Design Forums:
“Compute-in-X (CiX): Overcoming the Data Bottleneck in AI Processing”
“Chip Design for Low-Power, Robust, and Secure IoT Devices”
The 90-minute tutorials offer background information and a review of the basics in specific circuit- and
system-design topics. In the all-day Advanced-Circuit-Design Forums, leading experts present state-of-the-
art design strategies in a workshop-like format. The Forums are targeted at designers experienced in the
technical field.
...
-----------------------------------------------------------------
SESSION 15, Tuesday February 22nd, 2022
ML Processors
Session Chair: SukHwan Lim, Samsung Electronics, Hwaseong-si, Korea
Session Co-Chair: Sophia Shao, University of California, Berkeley, Berkeley, CA
...
2:00 PM
15.3
COMB-MCM: Computing-on-Memory-Boundary NN Processor with Bipolar Bitwise Sparsity Optimization for Scalable Multi-Chiplet-Module Edge Machine Learning
H. Zhu* 1 , B. Jiao* 1 , J. Zhang* 1 , X. Jia 1 , Y. Wang 1 , T. Guan 2 , S. Wang 2 , D. Niu 2 , H. Zheng 2 , C. Chen 1 ,
M. Wang 1 , L. Zhang 1 , X. Zeng 1 , Q. Liu 1 , Y. Xie 2 , M. Liu 1 , *Equally-Credited Authors (ECAs)
1 Fudan University, Shanghai, China; 2 Alibaba DAMO Academy, Shanghai, China
...
-----------------------------------------------------------------
SESSION 29, Wednesday February 23rd, 2022
ML Chips for Emerging Applications
Session Chair: Jun Deguchi, Kioxia, Kawasaki, Japan
Session Co-Chair: Jae-sun Seo, Arizona State University, Tempe, AZ
1:30 PM
29.1
184QPS/W 64Mb/mm^2 3D Logic-to-DRAM Hybrid Bonding with Process-Near-Memory Engine for Recommendation System
D. Niu 1 , S. Li 1 , Y. Wang 1 , W. Han 1 , Z. Zhang 2 , Y. Guan 2 , T. Guan 3 , F. Sun 1 , F. Xue 1 , L. Duan 1 , Y. Fang 1 ,
H. Zheng 1 , X. Jiang 4 , S. Wang 4 , F. Zuo 4 , Y. Wang 4 , B. Yu 4 , Q. Ren 4 , Y. Xie 1
1 Alibaba DAMO Academy, Sunnyvale, CA
2 Alibaba DAMO Academy, Beijing, China
3 Alibaba DAMO Academy, Shanghai, China
4 UniIC, Xian, China
...
-----------------------------------------------------------------
...
ISSCC is sponsored by IEEE and SSCS, the International Solid-State Circuits Conference – ISSCC – is the foremost global forum for presentation of advances in solid-state circuits and systems-on-a-chip. The Conference offers a unique opportunity for engineers working at the cutting edge of IC design and application to maintain technical currency, and to network with leading experts.
