Network-on-Chip (NoC) interconnect IP provider enables faster AI training in cloud datacenter
CAMPBELL, Calif. – October 9, 2018 – Arteris IP, the world’s leading supplier of silicon-proven commercial network-on-chip(NoC) interconnect intellectual property (IP), today announced that Enflame (Suiyuan) Technology has purchased multiple licenses of Arteris FlexNoC interconnect IP for use as the on-chip communications backbone of their artificial intelligence (AI) training chips for use in cloud datacenters.
"Arteris FlexNoC interconnect IP is the only interconnect that would allow our AI chips to achieve their high bandwidth requirements while also meeting our QoS requirements. Using Arteris NoC technology allows our architecture to take maximum advantage of state-of-the-art HBM2 memories to avoid system-level data starvation, which is major problem with less efficient AI training chips.”
Enflame is a China-based startup that has received pre-Series A investment from Tencent Holdings. It is developing AI training systems-on-chip (SoC) that implement neural networks. These chips use multiple hardware accelerators to increase the speed and accuracy of neural network training while reducing data center power consumption.
Enflame is a China-based startup that has received pre-Series A investment from Tencent Holdings. It is developing AI training systems-on-chip (SoC) that implement neural networks. These chips use multiple hardware accelerators to increase the speed and accuracy of neural network training while reducing data center power consumption.
Enflame chose Arteris IP FlexNoC interconnect for their AI chips because it:
- Enables easy creation of regular topologies used in AI chips, such as rings and meshes.
- Implements means to multicast and/or broadcast data on the interconnect, which can help reduce off-chip memory accesses.
- Allows high bandwidth off-chip communications with 1024-bit wide data access to high-bandwidth memories (HBM) complying with the JEDEC JESD235A HBM2 specification.
- Aids determinative end-to-end quality-of-service (QoS) capabilities, guaranteeing bandwidth and latency deadlines for critical on-chip communications.
“Arteris FlexNoC interconnect IP is the only interconnect that would allow our AI chips to achieve their high bandwidth requirements while also meeting our QoS requirements,” said Arthur Zhang, COO of Enflame. “Using Arteris NoC technology allows our architecture to take maximum advantage of state-of-the-art HBM2 memories to avoid system-level data starvation, which is major problem with less efficient AI training chips.”
“Enflame’s choice of Arteris FlexNoC interconnect IP is proof of our NoC technology’s advantages in complex neural network-based AI and machine learning chips,” said K. Charles Janac, President and CEO of Arteris IP. “Today’s AI training and inference chips require our NoC technology to ensure efficient data sharing amongst tens or even hundreds of on-chip processing elements and hardware accelerators.”
About Enflame
Enflame (Suiyuan) Technology is a China-based startup developing cloud-based deep learning chips for AI training platforms. Founded in March 2018, the company focuses on developing chips based on a self-developed unique architecture design to achieve high computational power and efficiency with low cost for deployment in major machine learning frameworks. The company has two research and development centers in Beijing and Shanghai.
Enflame (Suiyuan) Technology has completed its Series Pre-A funding led by Tencent Holdings Ltd with nearly CN¥340 million (US$50 million). Proceeds of this round will be used for research and development of its cloud-based AI chips and relevant software investments.
About Arteris IP
Arteris IP provides
network-on-chip (NoC) interconnect IP to accelerate system-on-chip (SoC) semiconductor assembly for a wide range of applications from AI to automobiles, mobile phones, IoT, cameras, SSD controllers, and servers for customers such as
Samsung,
Huawei / HiSilicon,
Mobileye, and
Texas Instruments. Arteris IP products include the
Ncore cache coherent and
FlexNoC non-coherent interconnect IP, the
CodaCache standalone last level cache, and optional
Resilience Package (ISO 26262 functional safety) and
PIANO automated timing closure capabilities. Customer results obtained by using the Arteris IP product line include lower power, higher performance, more efficient design reuse and faster SoC development, leading to lower development and production costs. For more information, visit
www.arteris.com
Contact:
Kurt Shuler
Arteris IP
+1 408 470 7300
Email Contact
