DinoplusAI Partners with SiFive to Develop Mission-Critical AI Processor Platform for High Performance Processing with Ultra-Low Latency
Optimized for Data Centers, 5g/Edge Cloud and Autonomous Vehicles
SiFive, the leading provider of commercial RISC-V processor core IP, design platforms and silicon solutions, announced that it was selected by DinoplusAI, an innovator in artificial intelligence (AI) processors and software for high performance mission-critical applications, to develop its mission-critical AI processor platform, which is optimized for ultra-low latency applications, such as data centers, 5G/edge cloud and autonomous vehicles. The DinoplusAI processor platform combines the SiFive RISC-V E34 Management Core with a proprietary AI engine from DinoplusAI. In addition, SiFive’s Custom SoC Division will provide industry-proven robust RTL-to-physical design custom SoC implementation methodology to ensure first-time working silicon and enable high-volume production.
The DinoplusAI processor platform provides consistent ultra-low latency, mitigating the common constraints of high-performance AI processing. The DinoplusAI processor features a scalable architecture, software stack and user interfaces. It not only addresses the computing power and energy efficiency required in AI applications, but also the latency, security and reliability that designers of AI products and services rely on. The processor platform is highly consistent and predictable, and has achieved the industry’s lowest latency at low batch size, which is critical in AI applications.
“We chose to partner with SiFive because of the company’s leadership in RISC-V cores, and its proven success in ASIC design and manufacturing of fully optimized SoCs,” said Jay Hu, CEO of DinoplusAI. “This expertise, combined with our IP blocks for AI, will result in a truly unique solution that optimizes inference with a focus on performance, power efficiency, ultra-low latency, assured security and reliability.”
“The DinoplusAI processor, with superior DeepBench LSTM performance, enables more than 4,000 real-time audio streams with a computation latency of less than 1ms for Rokid’s acoustic model, an industry leading cloud speech recognition algorithm,” said Dr. Yi Rao, director of the R-lab at Rokid.
“Meeting the ultra-low latency challenges of AI applications requires a successful platform approach that comprises of a new class of silicon IP, cores and advanced physical implementation techniques,” said Naveed Sherwani, CEO of SiFive. “The DinoplusAI processor platform features powerful IP, specifically designed by DinoplusAI for next-generation AI applications. We will integrate these AI IP blocks into our 16nm FinFET design platform based on the SiFive RISC-V E34 management core to deliver a complete silicon solution.”