These solutions are characterised by high-speed processing and low power consumption applied to AI.
A network infrastructure that is scalable, highly reliable and energy-efficient is essential not only for Beyond 5G and 6G mobile communication systems, but also for the environment and the sustainable development of the telecommunications industry.
In order to reduce overall network infrastructure cost and power consumption as well as offer the next generation of services, there is a growing need for a wireless base station virtualisation technology that offloads computationally demanding Layer 1 High-PHY processing to dedicated accelerators.
SoftBank and EdgeCortix have adapted EdgeCortix's DNA (Dynamic Neural Accelerator) IP technology to Layer 1 High-PHY processing and implemented low-density parity-check (LDPC) encoding. This is a type of wireless error correction code that has been deployed on FPGA boards. As a result, SoftBank and EdgeCortix have been able to achieve throughput of approximately 36Gbps, which is equivalent to almost 10 times the throughput compared to publicly available data of commercial IP solutions.
By expanding the application of this IP technology, SoftBank said that it will be possible to deliver more energy-efficient and low-latency wireless accelerators. In the future, SoftBank will determine the feasibility of applying this technology to the entire High-PHY process to realise a radio accelerator with higher performance.
“Improving the performance and the energy efficiency of our network infrastructure is a major challenge for the future” said Ryuji Wakikawa, Head of SoftBank’s Research Institute of Advanced Technology. “Our expectation of EdgeCortix is to be a partner who can provide both the IP and expertise that is needed to tackle these challenges simultaneously. We will continue to work with the EdgeCortix team to research the best accelerators to further improve performance and conserve power to deliver a high-quality network."
“The Open RAN acceleration opportunity with SoftBank was tailor-made for our Edge AI Platform and the Dynamic Neural Accelerator IP we’ve developed over the last three years,” added Sakyasingha Dasgupta, CEO and Founder of EdgeCortix.
“Our IP was designed to be runtime-reconfigurable, while being agnostic to AI models and the inherent complex mathematics in them. We have developed a very nimble architectural design delivered through our key assets: MERA compiler and software framework, scalable neural network engines and the SAKURA system-on-chip architecture; we are able to customise our Platform to the unique needs at the intersection of AI and the 5G space. The performance we’ve achieved on our prototype system while collaborating with SoftBank’s R&D team to date has even surprised us. In the future, we will pursue adaptation to end-to-end 5G wireless accelerator solutions that span MEC and other systems.”
