Black Sesame, Blue Ocean, Tenstorrent use Ncore and FlexNoC in AI chips
Typically in a AI chip, there are traditional CPUs and a lot of accelerators and customised processing elements forming a neural network. These elements communicate with each other in real-time, where huge amount of data transferring between them. AI chip designers can save on-chip networking design effort by using some of the available Inter-chip networking IP.
Ncore and FlexNoC from Arteris is one such Ips being used by some of the latest AI chips. Mobileye has used Ncore and FlexNoC to connect processing elements inside it.
Ncore is designed to efficiently connect various components within a SoC, including the CPU, GPU, accelerators, memory subsystems, and I/O interfaces. FlexNoC is configurable and scalable network-on-chip providing flexible and efficient communication and interconnects IP blocks within an SoC, such as processors, memories, peripherals, and accelerators. In neural networks, FlexNoC can be used to interconnect different IP blocks involved in AI processing. Ncore and FlexNoC can be explored to use as interconnect Ips across nodes in a distributed system or a network of interconnected devices
Black Sesame licensed Arteris FlexNoC interconnect IP and the accompanying FlexNoC Resilience Package for use in its automotive advanced driver assistance systems. China-based leading artificial intelligence (AI) SoC start-up semiconductor Blue Ocean Smart System licensed Arteris FlexNoC Interconnect IP and the accompanying AI Package for use in next generation systems-on-chip (SoC) that hardware accelerate artificial intelligence (AI) neural network inference and training.
David Zeng, VP of Engineering of Black Sesame. “Arteris IP is the world leader in NoC interconnect technology and a pioneer in automotive and AI markets, and Arteris FlexNoC IP and the Resilience Package are helping us to more quickly develop our ISO 26262-compliant systems.”
“From prior experience, I knew that Arteris interconnect IP was the best solution to construct complex high frequency and high bandwidth on-chip interconnects that were backend friendly for easier timing closure. The addition of the AI Package allows us to finely tune our chip architecture using multicast write semantics which greatly reduce off-chip memory accesses while using little die area and consuming much less power,” said John Rowland, President of Blue Ocean Smart System. “Our use of Arteris FlexNoC and the AI Package has been key to turning our architectural dreams into system-on-chip reality.”
John Rowland also praised “The Arteris IP team and technology have been crucial to optimizing our on-chip dataflow for our unique AI chip architecture. Arteris IP has been a great partner providing best-in-class technology and world-class technical support.”
The newest customer who has licensed Ncore and FlexNoC interconnect Ips is Tenstorrent. Tenstorrent designed advanced AI chips using open source RISC-V processor architecture to design chiplets for handling specific data streams and applications.
It is written in an article at softei.com saying "Using automated topology generation for mesh, ring and torus networks, the FlexNoC 4 AI enables SoC architects to generate AI topologies automatically and also edit generated topologies to optimise each individual network router, if desired." and "The FlexNoC 4 AI intelligent multicast optimises the use of on-chip and off-chip bandwidth by broadcasting data as close to network targets as possible. This allows for more efficient updates of DNN weights, image maps and other multicast data."
Read the full article at: https://softei.com/noc-ip-combines-with-ai-package-for-neural-networking/
Arteris is a fast growing silicon IP company with annual revenues the range of US$ 50 million US dollars. Arteris web site can be accessed at http://www.arteris.com.